From 2e33637daecd10f686f10abdf9495c855a11da7e Mon Sep 17 00:00:00 2001
From: David Blanc Brioir <davidblancbrioir@gmail.com>
Date: Thu, 25 Dec 2025 12:53:14 +0100
Subject: [PATCH] Update framework configuration and clean up obsolete specs
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Configuration updates:
- Added .env.example template for environment variables
- Updated README.md with better setup instructions (.env usage)
- Enhanced .claude/settings.local.json with additional Bash permissions
- Added .claude/CLAUDE.md framework documentation

Spec cleanup:
- Removed obsolete spec files (language_selection, mistral_extensible, template, theme_customization)
- Consolidated app_spec.txt (Claude Clone example)
- Added app_spec_model.txt as reference template
- Added app_spec_library_rag_types_docs.txt
- Added coding_prompt_library.md

Framework improvements:
- Updated agent.py, autonomous_agent_demo.py, client.py with minor fixes
- Enhanced dockerize_my_project.py
- Updated prompts (initializer, initializer_bis) with better guidance
- Added docker-compose.my_project.yml example

This commit consolidates improvements made during development sessions.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
---
 .claude/CLAUDE.md                             |  234 ++++
 .claude/settings.local.json                   |   36 +-
 .env.example                                  |   11 +
 GUIDE_NEW_APP.md                              |    3 +
 README.md                                     |   42 +-
 agent.py                                      |   39 +-
 autonomous_agent_demo.py                      |   22 +-
 client.py                                     |   14 +-
 docker-compose.my_project.yml                 |   29 +
 linear_config.py                              |    8 +-
 package-lock.json                             |    6 +
 prompts.py                                    |    5 +
 prompts/app_spec.txt                          | 1171 ++++++++---------
 ...app_spec_language_selection.completion.txt |  179 ---
 prompts/app_spec_language_selection.txt       |  525 --------
 prompts/app_spec_library_rag_types_docs.txt   |  679 ++++++++++
 prompts/app_spec_mistral_extensible.txt       |  448 -------
 prompts/app_spec_model.txt                    |  681 ++++++++++
 prompts/app_spec_template.txt                 |  134 --
 prompts/app_spec_theme_customization.txt      |  403 ------
 prompts/app_spec_types_docs.backup.txt        |  679 ++++++++++
 prompts/coding_prompt_library.md              |  290 ++++
 prompts/initializer_bis_prompt.md             |    7 +
 prompts/initializer_prompt.md                 |   13 +-
 prompts/spec_embed_BAAI.txt                   |  576 ++++++++
 requirements.txt                              |    1 +
 security.py                                   |    5 +
 27 files changed, 3862 insertions(+), 2378 deletions(-)
 create mode 100644 .claude/CLAUDE.md
 create mode 100644 .env.example
 create mode 100644 docker-compose.my_project.yml
 create mode 100644 package-lock.json
 delete mode 100644 prompts/app_spec_language_selection.completion.txt
 delete mode 100644 prompts/app_spec_language_selection.txt
 create mode 100644 prompts/app_spec_library_rag_types_docs.txt
 delete mode 100644 prompts/app_spec_mistral_extensible.txt
 create mode 100644 prompts/app_spec_model.txt
 delete mode 100644 prompts/app_spec_template.txt
 delete mode 100644 prompts/app_spec_theme_customization.txt
 create mode 100644 prompts/app_spec_types_docs.backup.txt
 create mode 100644 prompts/coding_prompt_library.md
 create mode 100644 prompts/spec_embed_BAAI.txt

diff --git a/.claude/CLAUDE.md b/.claude/CLAUDE.md
new file mode 100644
index 0000000..ce017a2
--- /dev/null
+++ b/.claude/CLAUDE.md
@@ -0,0 +1,234 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Overview
+
+This is an autonomous coding agent framework that uses Claude Agent SDK with Linear integration for project management. The framework enables long-running autonomous development sessions where agents create complete applications from XML specifications.
+
+**Key Architecture**: Two-agent pattern (Initializer + Coding Agent) with Linear as the single source of truth for project state and progress tracking.
+
+## Common Commands
+
+### Running the Agent
+
+```bash
+# Fresh project initialization
+python autonomous_agent_demo.py --project-dir ./my_project
+
+# Continue existing project
+python autonomous_agent_demo.py --project-dir ./my_project
+
+# Add new features to existing project (Initializer Bis)
+python autonomous_agent_demo.py --project-dir ./my_project --new-spec app_spec_theme_customization.txt
+
+# Limit iterations for testing
+python autonomous_agent_demo.py --project-dir ./my_project --max-iterations 3
+```
+
+### Testing
+
+```bash
+# Run security hook tests
+python test_security.py
+
+# Test mypy type checking (for library projects)
+mypy path/to/module.py
+```
+
+### Environment Setup
+
+```bash
+# Generate Claude Code OAuth token
+claude setup-token
+
+# Install dependencies
+pip install -r requirements.txt
+```
+
+## High-Level Architecture
+
+### Core Agent Flow
+
+1. **First Run (Initializer Agent)**:
+   - Reads `prompts/app_spec.txt` specification
+   - Creates Linear project and ~50 issues (one per `<feature_X>` tag)
+   - Creates META issue for session tracking
+   - Initializes project structure with `init.sh`
+   - Writes `.linear_project.json` marker file
+
+2. **Subsequent Runs (Coding Agent)**:
+   - Queries Linear for highest-priority Todo issue
+   - Updates issue status to "In Progress"
+   - Implements feature using SDK tools
+   - Tests implementation (Puppeteer for web apps, pytest/mypy for libraries)
+   - Adds comment to Linear issue with implementation notes
+   - Marks issue as "Done"
+   - Updates META issue with session summary
+
+3. **Initializer Bis (Add Features)**:
+   - Triggered by `--new-spec` flag on existing projects
+   - Reads new spec file from `prompts/`
+   - Creates additional Linear issues for new features
+   - Updates existing project without re-initializing
+
+### Key Design Patterns
+
+**Session Handoff via Linear**: Agents don't use local state files for coordination. All session context, implementation notes, and progress tracking happens through Linear issues and comments. This provides:
+- Real-time visibility in Linear workspace
+- Persistent history across sessions
+- Easy debugging via issue comments
+
+**Defense-in-Depth Security** (see `security.py` and `client.py`):
+1. OS-level sandbox for bash command isolation
+2. Filesystem restrictions (operations limited to project directory)
+3. Bash command allowlist with pre-tool-use hooks
+4. Explicit MCP tool permissions
+
+**Project Type Detection** (`agent.py:is_library_project`):
+- Detects library/type-safety projects vs full-stack web apps
+- Uses different coding prompts (`coding_prompt_library.md` vs `coding_prompt.md`)
+- Keywords: "type safety", "docstrings", "mypy", "library rag"
+
+### Module Responsibilities
+
+- **`autonomous_agent_demo.py`**: Entry point, argument parsing, environment validation
+- **`agent.py`**: Core agent loop, session orchestration, project type detection
+- **`client.py`**: Claude SDK client configuration, MCP server setup (Linear + Puppeteer)
+- **`security.py`**: Bash command validation with allowlist, pre-tool-use hooks
+- **`prompts.py`**: Prompt loading utilities, spec file copying
+- **`progress.py`**: Progress tracking via `.linear_project.json` marker
+- **`linear_config.py`**: Linear API configuration constants
+
+### MCP Servers
+
+**Linear** (HTTP transport at `mcp.linear.app/mcp`):
+- Project/team management
+- Issue CRUD operations
+- Comments and status updates
+- Requires `LINEAR_API_KEY` in `.env`
+
+**Puppeteer** (stdio transport):
+- Browser automation for UI testing
+- Navigate, screenshot, click, fill, evaluate
+- Used by web app projects, not library projects
+
+## Application Specification Format
+
+Specifications use XML format in `prompts/app_spec.txt`:
+
+```xml
+<project_specification>
+  <project_name>Your App Name</project_name>
+  <overview>Detailed description...</overview>
+
+  <technology_stack>
+    <frontend>...</frontend>
+    <backend>...</backend>
+  </technology_stack>
+
+  <core_features>
+    <feature_1>
+      <title>Feature title</title>
+      <description>Detailed description</description>
+      <priority>1-4 (1=urgent, 4=low)</priority>
+      <category>frontend|backend|auth|etc</category>
+      <test_steps>
+        1. Step one
+        2. Step two
+      </test_steps>
+    </feature_1>
+    <!-- More features... -->
+  </core_features>
+</project_specification>
+```
+
+**Important**: Each `<feature_X>` tag becomes a separate Linear issue. The initializer creates exactly one issue per feature tag.
+
+## Environment Configuration
+
+All configuration via `.env` file (copy from `.env.example`):
+
+```bash
+CLAUDE_CODE_OAUTH_TOKEN='your-oauth-token'  # From: claude setup-token
+LINEAR_API_KEY='lin_api_xxxxx'              # From: linear.app/settings/api
+LINEAR_TEAM_ID='team-id'                    # Optional, agent prompts if missing
+```
+
+## Security Model
+
+### Allowed Commands (`security.py:ALLOWED_COMMANDS`)
+
+File operations: `ls`, `cat`, `head`, `tail`, `wc`, `grep`, `cp`, `mkdir`, `chmod`
+Development: `npm`, `node`, `python`, `python3`, `mypy`, `pytest`
+Version control: `git`
+Process management: `ps`, `lsof`, `sleep`, `pkill`
+Scripts: `init.sh`
+
+### Additional Validation
+
+- **`pkill`**: Only allowed for dev processes (node, npm, vite, next)
+- **`chmod`**: Only `+x` mode permitted (making scripts executable)
+- **`init.sh`**: Must be `./init.sh` or end with `/init.sh`
+
+### Adding New Commands
+
+Edit `security.py:ALLOWED_COMMANDS` and optionally add validation logic to `bash_security_hook`.
+
+## Generated Project Structure
+
+After initialization, projects contain:
+
+```
+my_project/
+├── .linear_project.json      # Linear state marker (project_id, total_issues, meta_issue_id)
+├── .claude_settings.json     # Security settings (auto-generated)
+├── app_spec.txt              # Original specification (copied from prompts/)
+├── init.sh                   # Environment setup script (executable)
+└── [generated code]          # Application files created by agent
+```
+
+## Creating New Applications
+
+1. Create `prompts/app_spec.txt` with your XML specification
+2. Use existing spec files as templates (see `prompts/app_spec.txt` for Claude Clone example)
+3. Run: `python autonomous_agent_demo.py --project-dir ./new_app`
+4. Monitor progress in Linear workspace
+
+See `GUIDE_NEW_APP.md` for detailed guide (French).
+
+## Prompt Templates
+
+Located in `prompts/`:
+
+- **`initializer_prompt.md`**: First session prompt (creates Linear project/issues)
+- **`initializer_bis_prompt.md`**: Add features prompt (extends existing project)
+- **`coding_prompt.md`**: Standard coding session (web apps with Puppeteer testing)
+- **`coding_prompt_library.md`**: Library coding session (focuses on types/docs, uses pytest/mypy)
+
+The framework automatically selects the appropriate prompt based on session type and project detection.
+
+## Important Implementation Notes
+
+### Linear Integration
+
+- All work tracked as Linear issues, not local files
+- Session handoff via Linear comments on META issue
+- Status workflow: Todo → In Progress → Done
+- Early termination: Agent stops when detecting "feature-complete" in responses
+
+### Auto-Continue Behavior
+
+Agent auto-continues with 3-second delay between sessions (`agent.py:AUTO_CONTINUE_DELAY_SECONDS`). Stops when:
+- `--max-iterations` limit reached
+- Response contains "feature-complete" or "all issues completed"
+- Fatal error occurs
+
+### Project Directory Handling
+
+Relative paths automatically placed under `generations/` directory unless absolute path provided.
+
+### Model Selection
+
+Default: `claude-opus-4-5-20251101` (Opus 4.5 for best coding performance)
+Override with: `--model claude-sonnet-4-5-20250929`
diff --git a/.claude/settings.local.json b/.claude/settings.local.json
index c85f664..047e8a8 100644
--- a/.claude/settings.local.json
+++ b/.claude/settings.local.json
@@ -3,7 +3,41 @@
     "allow": [
       "Bash(test:*)",
       "Bash(cat:*)",
-      "Bash(netstat:*)"
+      "Bash(netstat:*)",
+      "Bash(docker-compose:*)",
+      "Bash(ls:*)",
+      "Bash(rm:*)",
+      "Bash(python autonomous_agent_demo.py:*)",
+      "Bash(dir C:GitHublinear_coding_philosophia_raggenerationslibrary_rag*.py)",
+      "Bash(git add:*)",
+      "Bash(git commit -m \"$\\(cat <<''EOF''\nFix import error: rename delete_document_passages to delete_document_chunks\n\nThe function was renamed in weaviate_ingest.py but the import in __init__.py\nwas not updated, causing ImportError when using the library.\n\nChanges:\n- Updated import statement in utils/__init__.py\n- Updated __all__ export list to use correct function name\nEOF\n\\)\")",
+      "Bash(dir \"C:\\\\GitHub\\\\linear_coding_philosophia_rag\\\\generations\\\\library_rag\\\\.env\")",
+      "Bash(git commit:*)",
+      "Bash(tasklist:*)",
+      "Bash(findstr:*)",
+      "Bash(wmic process:*)",
+      "Bash(powershell -Command \"Get-Process python | Select-Object Id,Path,StartTime | Format-Table -AutoSize\")",
+      "Bash(powershell -Command \"Get-WmiObject Win32_Process -Filter \"\"name = ''python.exe''\"\" | Select-Object ProcessId, CommandLine | Format-List\")",
+      "Bash(timeout:*)",
+      "Bash(powershell -Command:*)",
+      "Bash(python:*)",
+      "Bash(dir \"C:\\\\GitHub\\\\linear_coding_library_rag\\\\generations\\\\library_rag\")",
+      "Bash(docker ps:*)",
+      "Bash(curl:*)",
+      "Bash(dir:*)",
+      "Bash(grep:*)",
+      "Bash(git push:*)",
+      "Bash(mypy:*)",
+      "WebSearch",
+      "Bash(nvidia-smi:*)",
+      "WebFetch(domain:cr.weaviate.io)",
+      "Bash(git restore:*)",
+      "Bash(git log:*)",
+      "Bash(done)",
+      "Bash(git remote set-url:*)",
+      "Bash(docker compose:*)",
+      "Bash(pytest:*)",
+      "Bash(git pull:*)"
     ]
   }
 }
diff --git a/.env.example b/.env.example
new file mode 100644
index 0000000..f9fa073
--- /dev/null
+++ b/.env.example
@@ -0,0 +1,11 @@
+# Claude Code OAuth Token
+# Run 'claude setup-token' to generate this token
+CLAUDE_CODE_OAUTH_TOKEN=your-oauth-token-here
+
+# Linear API Key
+# Get your API key from: https://linear.app/YOUR-TEAM/settings/api
+LINEAR_API_KEY=lin_api_xxxxxxxxxxxxx
+
+# Linear Team ID (optional)
+# If not set, the agent will list teams and ask you to choose
+LINEAR_TEAM_ID=
diff --git a/GUIDE_NEW_APP.md b/GUIDE_NEW_APP.md
index eda9662..db5b672 100644
--- a/GUIDE_NEW_APP.md
+++ b/GUIDE_NEW_APP.md
@@ -332,3 +332,6 @@ Pour créer une nouvelle application :
 Le framework s'occupe du reste ! 🚀
 
 
+
+
+
diff --git a/README.md b/README.md
index 4e92c64..0d3d258 100644
--- a/README.md
+++ b/README.md
@@ -26,23 +26,35 @@ pip install -r requirements.txt
 
 ### 2. Set Up Authentication
 
-You need two authentication tokens:
+Create a `.env` file in the root directory by copying the example:
 
-**Claude Code OAuth Token:**
+```bash
+cp .env.example .env
+```
+
+Then configure your credentials in the `.env` file:
+
+**1. Claude Code OAuth Token:**
 ```bash
 # Generate the token using Claude Code CLI
 claude setup-token
 
-# Set the environment variable
-export CLAUDE_CODE_OAUTH_TOKEN='your-oauth-token-here'
+# Add to .env file:
+CLAUDE_CODE_OAUTH_TOKEN='your-oauth-token-here'
 ```
 
-**Linear API Key:**
+**2. Linear API Key:**
 ```bash
 # Get your API key from: https://linear.app/YOUR-TEAM/settings/api
-export LINEAR_API_KEY='lin_api_xxxxxxxxxxxxx'
+# Add to .env file:
+LINEAR_API_KEY='lin_api_xxxxxxxxxxxxx'
+
+# Optional: Linear Team ID (if not set, agent will list teams)
+LINEAR_TEAM_ID='your-team-id'
 ```
 
+**Important:** The `.env` file is already in `.gitignore` - never commit it!
+
 ### 3. Verify Installation
 
 ```bash
@@ -142,12 +154,15 @@ Instead of local text files, agents communicate through:
 - **META Issue**: Session summaries and handoff notes
 - **Issue Status**: Todo / In Progress / Done workflow
 
-## Environment Variables
+## Configuration (.env file)
+
+All configuration is done via a `.env` file in the root directory.
 
 | Variable | Description | Required |
 |----------|-------------|----------|
 | `CLAUDE_CODE_OAUTH_TOKEN` | Claude Code OAuth token (from `claude setup-token`) | Yes |
 | `LINEAR_API_KEY` | Linear API key for MCP access | Yes |
+| `LINEAR_TEAM_ID` | Linear Team ID (if not set, agent will list teams and ask) | No |
 
 ## Command Line Options
 
@@ -268,11 +283,14 @@ Edit `security.py` to add or remove commands from `ALLOWED_COMMANDS`.
 
 ## Troubleshooting
 
-**"CLAUDE_CODE_OAUTH_TOKEN not set"**
-Run `claude setup-token` to generate a token, then export it.
+**"CLAUDE_CODE_OAUTH_TOKEN not found in .env file"**
+1. Run `claude setup-token` to generate a token
+2. Copy `.env.example` to `.env`
+3. Add your token to the `.env` file
 
-**"LINEAR_API_KEY not set"**
-Get your API key from `https://linear.app/YOUR-TEAM/settings/api`
+**"LINEAR_API_KEY not found in .env file"**
+1. Get your API key from `https://linear.app/YOUR-TEAM/settings/api`
+2. Add it to your `.env` file
 
 **"Appears to hang on first run"**
 Normal behavior. The initializer is creating a Linear project and 50 issues with detailed descriptions. Watch for `[Tool: mcp__linear__create_issue]` output.
@@ -281,7 +299,7 @@ Normal behavior. The initializer is creating a Linear project and 50 issues with
 The agent tried to run a disallowed command. Add it to `ALLOWED_COMMANDS` in `security.py` if needed.
 
 **"MCP server connection failed"**
-Verify your `LINEAR_API_KEY` is valid and has appropriate permissions. The Linear MCP server uses HTTP transport at `https://mcp.linear.app/mcp`.
+Verify your `LINEAR_API_KEY` in the `.env` file is valid and has appropriate permissions. The Linear MCP server uses HTTP transport at `https://mcp.linear.app/mcp`.
 
 ## Viewing Progress
 
diff --git a/agent.py b/agent.py
index 2e06223..8ca7b00 100644
--- a/agent.py
+++ b/agent.py
@@ -17,6 +17,7 @@ from prompts import (
     get_initializer_prompt,
     get_initializer_bis_prompt,
     get_coding_prompt,
+    get_coding_prompt_library,
     copy_spec_to_project,
     copy_new_spec_to_project,
 )
@@ -26,6 +27,34 @@ from prompts import (
 AUTO_CONTINUE_DELAY_SECONDS = 3
 
 
+def is_library_project(project_dir: Path) -> bool:
+    """
+    Detect if this is a library/type-safety project vs a full-stack web app.
+
+    Checks app_spec.txt for keywords related to type safety, documentation, or library projects.
+    """
+    app_spec_path = project_dir / "app_spec.txt"
+    if not app_spec_path.exists():
+        return False
+
+    try:
+        spec_content = app_spec_path.read_text(encoding='utf-8').lower()
+
+        # Keywords that indicate a library/type-safety project
+        library_keywords = [
+            "type safety",
+            "type annotations",
+            "docstrings",
+            "documentation enhancement",
+            "mypy",
+            "library rag",
+        ]
+
+        return any(keyword in spec_content for keyword in library_keywords)
+    except Exception:
+        return False
+
+
 async def run_agent_session(
     client: ClaudeSDKClient,
     message: str,
@@ -162,8 +191,8 @@ async def run_autonomous_agent(
         print("Fresh start - will use initializer agent")
         print()
         print("=" * 70)
-        print("  NOTE: First session takes 10-20+ minutes!")
-        print("  The agent is creating 50 Linear issues and setting up the project.")
+        print("  NOTE: First session may take several minutes!")
+        print("  The agent is creating Linear issues (one per feature in spec).")
         print("  This may appear to hang - it's working. Watch for [Tool: ...] output.")
         print("=" * 70)
         print()
@@ -213,7 +242,11 @@ async def run_autonomous_agent(
             prompt = get_initializer_bis_prompt()
             use_initializer_bis = False  # Only use initializer bis once
         else:
-            prompt = get_coding_prompt()
+            # Detect project type and use appropriate coding prompt
+            if is_library_project(project_dir):
+                prompt = get_coding_prompt_library()
+            else:
+                prompt = get_coding_prompt()
 
         # Run session with async context manager
         async with client:
diff --git a/autonomous_agent_demo.py b/autonomous_agent_demo.py
index 7d9b487..e296867 100644
--- a/autonomous_agent_demo.py
+++ b/autonomous_agent_demo.py
@@ -17,8 +17,12 @@ import asyncio
 import os
 from pathlib import Path
 
+from dotenv import load_dotenv
 from agent import run_autonomous_agent
 
+# Load environment variables from .env file
+load_dotenv()
+
 
 # Configuration
 # Using Claude Opus 4.5 as default for best coding and agentic performance
@@ -48,9 +52,10 @@ Examples:
   # Add new specifications to existing project
   python autonomous_agent_demo.py --project-dir ./claude_clone --new-spec app_spec_new1.txt
 
-Environment Variables:
+Configuration (.env file):
   CLAUDE_CODE_OAUTH_TOKEN    Claude Code OAuth token (required)
   LINEAR_API_KEY             Linear API key (required)
+  LINEAR_TEAM_ID             Linear Team ID (optional)
         """,
     )
 
@@ -91,18 +96,17 @@ def main() -> None:
 
     # Check for Claude Code OAuth token
     if not os.environ.get("CLAUDE_CODE_OAUTH_TOKEN"):
-        print("Error: CLAUDE_CODE_OAUTH_TOKEN environment variable not set")
-        print("\nRun 'claude setup-token' after installing the Claude Code CLI.")
-        print("\nThen set it:")
-        print("  export CLAUDE_CODE_OAUTH_TOKEN='your-token-here'")
+        print("Error: CLAUDE_CODE_OAUTH_TOKEN not found in .env file")
+        print("\n1. Run 'claude setup-token' after installing the Claude Code CLI")
+        print("2. Copy .env.example to .env")
+        print("3. Add your token to .env: CLAUDE_CODE_OAUTH_TOKEN='your-token-here'")
         return
 
     # Check for Linear API key
     if not os.environ.get("LINEAR_API_KEY"):
-        print("Error: LINEAR_API_KEY environment variable not set")
-        print("\nGet your API key from: https://linear.app/YOUR-TEAM/settings/api")
-        print("\nThen set it:")
-        print("  export LINEAR_API_KEY='lin_api_xxxxxxxxxxxxx'")
+        print("Error: LINEAR_API_KEY not found in .env file")
+        print("\n1. Get your API key from: https://linear.app/YOUR-TEAM/settings/api")
+        print("2. Add it to .env: LINEAR_API_KEY='lin_api_xxxxxxxxxxxxx'")
         return
 
     # Automatically place projects in generations/ directory unless already specified
diff --git a/client.py b/client.py
index 0961566..cd6867a 100644
--- a/client.py
+++ b/client.py
@@ -9,11 +9,15 @@ import json
 import os
 from pathlib import Path
 
+from dotenv import load_dotenv
 from claude_code_sdk import ClaudeCodeOptions, ClaudeSDKClient
 from claude_code_sdk.types import HookMatcher
 
 from security import bash_security_hook
 
+# Load environment variables from .env file
+load_dotenv()
+
 
 # Puppeteer MCP tools for browser automation
 PUPPETEER_TOOLS = [
@@ -85,15 +89,17 @@ def create_client(project_dir: Path, model: str) -> ClaudeSDKClient:
     api_key = os.environ.get("CLAUDE_CODE_OAUTH_TOKEN")
     if not api_key:
         raise ValueError(
-            "CLAUDE_CODE_OAUTH_TOKEN environment variable not set.\n"
-            "Run 'claude setup-token after installing the Claude Code CLI."
+            "CLAUDE_CODE_OAUTH_TOKEN not set in .env file.\n"
+            "Run 'claude setup-token' after installing the Claude Code CLI,\n"
+            "then add the token to your .env file."
         )
 
     linear_api_key = os.environ.get("LINEAR_API_KEY")
     if not linear_api_key:
         raise ValueError(
-            "LINEAR_API_KEY environment variable not set.\n"
-            "Get your API key from: https://linear.app/YOUR-TEAM/settings/api"
+            "LINEAR_API_KEY not set in .env file.\n"
+            "Get your API key from: https://linear.app/YOUR-TEAM/settings/api\n"
+            "then add it to your .env file."
         )
 
     # Create comprehensive security settings
diff --git a/docker-compose.my_project.yml b/docker-compose.my_project.yml
new file mode 100644
index 0000000..592ad00
--- /dev/null
+++ b/docker-compose.my_project.yml
@@ -0,0 +1,29 @@
+services:
+  my_project_frontend:
+    image: node:20
+    working_dir: /app
+    volumes:
+      - ./generations/my_project:/app
+      # Eviter de réutiliser les node_modules Windows dans le conteneur Linux
+      - /app/node_modules
+    command: ["sh", "-c", "npm install && npm run dev -- --host 0.0.0.0 --port 3000"]
+    ports:
+      - "4300:3000"
+    environment:
+      - NODE_ENV=development
+
+  my_project_server:
+    image: node:20
+    working_dir: /app/server
+    volumes:
+      - ./generations/my_project:/app
+      # Eviter de réutiliser les node_modules Windows dans le conteneur Linux
+      - /app/server/node_modules
+    command: ["sh", "-c", "npm install && npm start"]
+    ports:
+      - "4301:3001"
+    environment:
+      - NODE_ENV=development
+    depends_on:
+      - my_project_frontend
+
diff --git a/linear_config.py b/linear_config.py
index 434d1d2..38d4429 100644
--- a/linear_config.py
+++ b/linear_config.py
@@ -7,10 +7,14 @@ These values are used in prompts and for project state management.
 """
 
 import os
+from dotenv import load_dotenv
 
-# Environment variables (must be set before running)
+# Load environment variables from .env file
+load_dotenv()
+
+# Environment variables (loaded from .env file)
 LINEAR_API_KEY = os.environ.get("LINEAR_API_KEY")
-LINEAR_TEAM_ID  = os.environ.get("LINEAR_TEAM_ID") 
+LINEAR_TEAM_ID = os.environ.get("LINEAR_TEAM_ID") 
 
 # Default number of issues to create (can be overridden via command line)
 DEFAULT_ISSUE_COUNT = 50
diff --git a/package-lock.json b/package-lock.json
new file mode 100644
index 0000000..97cec99
--- /dev/null
+++ b/package-lock.json
@@ -0,0 +1,6 @@
+{
+  "name": "linear_coding_philosophia_rag",
+  "lockfileVersion": 3,
+  "requires": true,
+  "packages": {}
+}
diff --git a/prompts.py b/prompts.py
index f4c8a8b..cf08e91 100644
--- a/prompts.py
+++ b/prompts.py
@@ -28,6 +28,11 @@ def get_coding_prompt() -> str:
     return load_prompt("coding_prompt")
 
 
+def get_coding_prompt_library() -> str:
+    """Load the library-specific coding agent prompt (for type safety & documentation projects)."""
+    return load_prompt("coding_prompt_library")
+
+
 def copy_spec_to_project(project_dir: Path) -> None:
     """Copy the app spec file into the project directory for the agent to read."""
     spec_source = PROMPTS_DIR / "app_spec.txt"
diff --git a/prompts/app_spec.txt b/prompts/app_spec.txt
index 1e35f6d..147ea49 100644
--- a/prompts/app_spec.txt
+++ b/prompts/app_spec.txt
@@ -1,681 +1,542 @@
 <project_specification>
-  <project_name>Claude.ai Clone - AI Chat Interface</project_name>
+  <project_name>Library RAG MCP Server - PDF Ingestion & Semantic Retrieval</project_name>
 
   <overview>
-    Build a fully functional clone of claude.ai, Anthropic's conversational AI interface. The application should
-    provide a clean, modern chat interface for interacting with Claude via the API, including features like
-    conversation management, artifact rendering, project organization, multiple model selection, and advanced
-    settings. The UI should closely match claude.ai's design using Tailwind CSS with a focus on excellent
-    user experience and responsive design.
+    Serveur MCP (Model Context Protocol) exposant les capacités de Library RAG comme outils pour LLMs.
+
+    **Architecture simplifiée :**
+    - 1 outil d'ingestion : parse_pdf (configuration optimale pré-définie)
+    - 7 outils de retrieval : recherche sémantique et gestion de documents
+
+    **Configuration optimale (paramètres fixes) :**
+    - LLM : Mistral Medium (mistral-medium-latest)
+    - OCR : Mistral avec annotations (meilleure qualité TOC, 3x coût)
+    - Chunking : Sémantique intelligent (LLM-based)
+    - Ingestion : Automatique dans Weaviate
+
+    Le LLM client n'a qu'à fournir le chemin du PDF, tous les paramètres sont optimisés par défaut.
   </overview>
 
   <technology_stack>
-    <api_key>
-        You can use an API key located at /tmp/api-key for testing. You will not be allowed to read this file, but you can reference it in code.
-    </api_key>
-    <frontend>
-      <framework>React with Vite</framework>
-      <styling>Tailwind CSS (via CDN)</styling>
-      <state_management>React hooks and context</state_management>
-      <routing>React Router for navigation</routing>
-      <markdown>React Markdown for message rendering</markdown>
-      <code_highlighting>Syntax highlighting for code blocks</code_highlighting>
-      <port>Only launch on port {frontend_port}</port>
-    </frontend>
     <backend>
-      <runtime>Node.js with Express</runtime>
-      <database>SQLite with better-sqlite3</database>
-      <api_integration>Claude API for chat completions</api_integration>
-      <streaming>Server-Sent Events for streaming responses</streaming>
+      <runtime>Python 3.10+</runtime>
+      <mcp_framework>mcp Python SDK (official Anthropic implementation)</mcp_framework>
+      <vector_database>Weaviate 1.34.4 + text2vec-transformers</vector_database>
+      <ocr>Mistral OCR API with annotations</ocr>
+      <llm>Mistral API (mistral-medium-latest)</llm>
+      <type_checking>mypy strict</type_checking>
     </backend>
-    <communication>
-      <api>RESTful endpoints</api>
-      <streaming>SSE for real-time message streaming</streaming>
-      <claude_api>Integration with Claude API using Anthropic SDK</claude_api>
-    </communication>
+    <protocol>
+      <mcp_version>1.0</mcp_version>
+      <transport>stdio</transport>
+      <capabilities>tools</capabilities>
+    </protocol>
   </technology_stack>
 
-  <prerequisites>
-    <environment_setup>
-      - Repository includes .env with VITE_ANTHROPIC_API_KEY configured
-      - Frontend dependencies pre-installed via pnpm
-      - Backend code goes in /server directory
-      - Install backend dependencies as needed
-    </environment_setup>
-  </prerequisites>
-
-  <core_features>
-    <chat_interface>
-      - Clean, centered chat layout with message bubbles
-      - Streaming message responses with typing indicator
-      - Markdown rendering with proper formatting
-      - Code blocks with syntax highlighting and copy button
-      - LaTeX/math equation rendering
-      - Image upload and display in messages
-      - Multi-turn conversations with context
-      - Message editing and regeneration
-      - Stop generation button during streaming
-      - Input field with auto-resize textarea
-      - Character count and token estimation
-      - Keyboard shortcuts (Enter to send, Shift+Enter for newline)
-    </chat_interface>
-
-    <artifacts>
-      - Artifact detection and rendering in side panel
-      - Code artifact viewer with syntax highlighting
-      - HTML/SVG preview with live rendering
-      - React component preview
-      - Mermaid diagram rendering
-      - Text document artifacts
-      - Artifact editing and re-prompting
-      - Full-screen artifact view
-      - Download artifact content
-      - Artifact versioning and history
-    </artifacts>
-
-    <conversation_management>
-      - Create new conversations
-      - Conversation list in sidebar
-      - Rename conversations
-      - Delete conversations
-      - Search conversations by title/content
-      - Pin important conversations
-      - Archive conversations
-      - Conversation folders/organization
-      - Duplicate conversation
-      - Export conversation (JSON, Markdown, PDF)
-      - Conversation timestamps (created, last updated)
-      - Unread message indicators
-    </conversation_management>
-
-    <projects>
-      - Create projects to group related conversations
-      - Project knowledge base (upload documents)
-      - Project-specific custom instructions
-      - Share projects with team (mock feature)
-      - Project settings and configuration
-      - Move conversations between projects
-      - Project templates
-      - Project analytics (usage stats)
-    </projects>
-
-    <model_selection>
-      - Model selector dropdown with the following models:
-        - Claude Sonnet 4.5 (claude-sonnet-4-5-20250929) - default
-        - Claude Haiku 4.5 (claude-haiku-4-5-20251001)
-        - Claude Opus 4.1 (claude-opus-4-1-20250805)
-      - Model capabilities display
-      - Context window indicator
-      - Model-specific pricing info (display only)
-      - Switch models mid-conversation
-      - Model comparison view
-    </model_selection>
-
-    <custom_instructions>
-      - Global custom instructions
-      - Project-specific custom instructions
-      - Conversation-specific system prompts
-      - Custom instruction templates
-      - Preview how instructions affect responses
-    </custom_instructions>
-
-    <settings_preferences>
-      - Theme selection (Light, Dark, Auto)
-      - Font size adjustment
-      - Message density (compact, comfortable, spacious)
-      - Code theme selection
-      - Language preferences
-      - Accessibility options
-      - Keyboard shortcuts reference
-      - Data export options
-      - Privacy settings
-      - API key management
-    </settings_preferences>
-
-    <advanced_features>
-      - Temperature control slider
-      - Max tokens adjustment
-      - Top-p (nucleus sampling) control
-      - System prompt override
-      - Thinking/reasoning mode toggle
-      - Multi-modal input (text + images)
-      - Voice input (optional, mock UI)
-      - Response suggestions
-      - Related prompts
-      - Conversation branching
-    </advanced_features>
-
-    <collaboration>
-      - Share conversation via link (read-only)
-      - Export conversation formats
-      - Conversation templates
-      - Prompt library
-      - Share artifacts
-      - Team workspaces (mock UI)
-    </collaboration>
-
-    <search_discovery>
-      - Search across all conversations
-      - Filter by project, date, model
-      - Prompt library with categories
-      - Example conversations
-      - Quick actions menu
-      - Command palette (Cmd/Ctrl+K)
-    </search_discovery>
-
-    <usage_tracking>
-      - Token usage display per message
-      - Conversation cost estimation
-      - Daily/monthly usage dashboard
-      - Usage limits and warnings
-      - API quota tracking
-    </usage_tracking>
-
-    <onboarding>
-      - Welcome screen for new users
-      - Feature tour highlights
-      - Example prompts to get started
-      - Quick tips and best practices
-      - Keyboard shortcuts tutorial
-    </onboarding>
-
-    <accessibility>
-      - Full keyboard navigation
-      - Screen reader support
-      - ARIA labels and roles
-      - High contrast mode
-      - Focus management
-      - Reduced motion support
-    </accessibility>
-
-    <responsive_design>
-      - Mobile-first responsive layout
-      - Touch-optimized interface
-      - Collapsible sidebar on mobile
-      - Swipe gestures for navigation
-      - Adaptive artifact display
-      - Progressive Web App (PWA) support
-    </responsive_design>
-  </core_features>
-
-  <database_schema>
-    <tables>
-      <users>
-        - id, email, name, avatar_url
-        - created_at, last_login
-        - preferences (JSON: theme, font_size, etc.)
-        - custom_instructions
-      </users>
-
-      <projects>
-        - id, user_id, name, description, color
-        - custom_instructions, knowledge_base_path
-        - created_at, updated_at
-        - is_archived, is_pinned
-      </projects>
-
-      <conversations>
-        - id, user_id, project_id, title
-        - model, created_at, updated_at, last_message_at
-        - is_archived, is_pinned, is_deleted
-        - settings (JSON: temperature, max_tokens, etc.)
-        - token_count, message_count
-      </conversations>
-
-      <messages>
-        - id, conversation_id, role (user/assistant/system)
-        - content, created_at, edited_at
-        - tokens, finish_reason
-        - images (JSON array of image data)
-        - parent_message_id (for branching)
-      </messages>
-
-      <artifacts>
-        - id, message_id, conversation_id
-        - type (code/html/svg/react/mermaid/text)
-        - title, identifier, language
-        - content, version
-        - created_at, updated_at
-      </artifacts>
-
-      <shared_conversations>
-        - id, conversation_id, share_token
-        - created_at, expires_at, view_count
-        - is_public
-      </shared_conversations>
-
-      <prompt_library>
-        - id, user_id, title, description
-        - prompt_template, category, tags (JSON)
-        - is_public, usage_count
-        - created_at, updated_at
-      </prompt_library>
-
-      <conversation_folders>
-        - id, user_id, project_id, name, parent_folder_id
-        - created_at, position
-      </conversation_folders>
-
-      <conversation_folder_items>
-        - id, folder_id, conversation_id
-      </conversation_folder_items>
-
-      <usage_tracking>
-        - id, user_id, conversation_id, message_id
-        - model, input_tokens, output_tokens
-        - cost_estimate, created_at
-      </usage_tracking>
-
-      <api_keys>
-        - id, user_id, key_name, api_key_hash
-        - created_at, last_used_at
-        - is_active
-      </api_keys>
-    </tables>
-  </database_schema>
-
-  <api_endpoints_summary>
-    <authentication>
-      - POST /api/auth/login
-      - POST /api/auth/logout
-      - GET /api/auth/me
-      - PUT /api/auth/profile
-    </authentication>
-
-    <conversations>
-      - GET /api/conversations
-      - POST /api/conversations
-      - GET /api/conversations/:id
-      - PUT /api/conversations/:id
-      - DELETE /api/conversations/:id
-      - POST /api/conversations/:id/duplicate
-      - POST /api/conversations/:id/export
-      - PUT /api/conversations/:id/archive
-      - PUT /api/conversations/:id/pin
-      - POST /api/conversations/:id/branch
-    </conversations>
-
-    <messages>
-      - GET /api/conversations/:id/messages
-      - POST /api/conversations/:id/messages
-      - PUT /api/messages/:id
-      - DELETE /api/messages/:id
-      - POST /api/messages/:id/regenerate
-      - GET /api/messages/stream (SSE endpoint)
-    </messages>
-
-    <artifacts>
-      - GET /api/conversations/:id/artifacts
-      - GET /api/artifacts/:id
-      - PUT /api/artifacts/:id
-      - DELETE /api/artifacts/:id
-      - POST /api/artifacts/:id/fork
-      - GET /api/artifacts/:id/versions
-    </artifacts>
-
-    <projects>
-      - GET /api/projects
-      - POST /api/projects
-      - GET /api/projects/:id
-      - PUT /api/projects/:id
-      - DELETE /api/projects/:id
-      - POST /api/projects/:id/knowledge
-      - GET /api/projects/:id/conversations
-      - PUT /api/projects/:id/settings
-    </projects>
-
-    <sharing>
-      - POST /api/conversations/:id/share
-      - GET /api/share/:token
-      - DELETE /api/share/:token
-      - PUT /api/share/:token/settings
-    </sharing>
-
-    <prompts>
-      - GET /api/prompts/library
-      - POST /api/prompts/library
-      - GET /api/prompts/:id
-      - PUT /api/prompts/:id
-      - DELETE /api/prompts/:id
-      - GET /api/prompts/categories
-      - GET /api/prompts/examples
-    </prompts>
-
-    <search>
-      - GET /api/search/conversations?q=query
-      - GET /api/search/messages?q=query
-      - GET /api/search/artifacts?q=query
-      - GET /api/search/prompts?q=query
-    </search>
-
-    <folders>
-      - GET /api/folders
-      - POST /api/folders
-      - PUT /api/folders/:id
-      - DELETE /api/folders/:id
-      - POST /api/folders/:id/items
-      - DELETE /api/folders/:id/items/:conversationId
-    </folders>
-
-    <usage>
-      - GET /api/usage/daily
-      - GET /api/usage/monthly
-      - GET /api/usage/by-model
-      - GET /api/usage/conversations/:id
-    </usage>
-
-    <settings>
-      - GET /api/settings
-      - PUT /api/settings
-      - GET /api/settings/custom-instructions
-      - PUT /api/settings/custom-instructions
-    </settings>
-
-    <claude_api>
-      - POST /api/claude/chat (proxy to Claude API)
-      - POST /api/claude/chat/stream (streaming proxy)
-      - GET /api/claude/models
-      - POST /api/claude/images/upload
-    </claude_api>
-  </api_endpoints_summary>
-
-  <ui_layout>
-    <main_structure>
-      - Three-column layout: sidebar (conversations), main (chat), panel (artifacts)
-      - Collapsible sidebar with resize handle
-      - Responsive breakpoints: mobile (single column), tablet (two column), desktop (three column)
-      - Persistent header with project/model selector
-      - Bottom input area with send button and options
-    </main_structure>
-
-    <sidebar_left>
-      - New chat button (prominent)
-      - Project selector dropdown
-      - Search conversations input
-      - Conversations list (grouped by date: Today, Yesterday, Previous 7 days, etc.)
-      - Folder tree view (collapsible)
-      - Settings gear icon at bottom
-      - User profile at bottom
-    </sidebar_left>
-
-    <main_chat_area>
-      - Conversation title (editable inline)
-      - Model selector badge
-      - Message history (scrollable)
-      - Welcome screen for new conversations
-      - Suggested prompts (empty state)
-      - Input area with formatting toolbar
-      - Attachment button for images
-      - Send button with loading state
-      - Stop generation button
-    </main_chat_area>
-
-    <artifacts_panel>
-      - Artifact header with title and type badge
-      - Code editor or preview pane
-      - Tabs for multiple artifacts
-      - Full-screen toggle
-      - Download button
-      - Edit/Re-prompt button
-      - Version selector
-      - Close panel button
-    </artifacts_panel>
-
-    <modals_overlays>
-      - Settings modal (tabbed interface)
-      - Share conversation modal
-      - Export options modal
-      - Project settings modal
-      - Prompt library modal
-      - Command palette overlay
-      - Keyboard shortcuts reference
-    </modals_overlays>
-  </ui_layout>
-
-  <design_system>
-    <color_palette>
-      - Primary: Orange/amber accent (#CC785C claude-style)
-      - Background: White (light mode), Dark gray (#1A1A1A dark mode)
-      - Surface: Light gray (#F5F5F5 light), Darker gray (#2A2A2A dark)
-      - Text: Near black (#1A1A1A light), Off-white (#E5E5E5 dark)
-      - Borders: Light gray (#E5E5E5 light), Dark gray (#404040 dark)
-      - Code blocks: Monaco editor theme
-    </color_palette>
-
-    <typography>
-      - Sans-serif system font stack (Inter, SF Pro, Roboto, system-ui)
-      - Headings: font-semibold
-      - Body: font-normal, leading-relaxed
-      - Code: Monospace (JetBrains Mono, Consolas, Monaco)
-      - Message text: text-base (16px), comfortable line-height
-    </typography>
-
-    <components>
-      <message_bubble>
-        - User messages: Right-aligned, subtle background
-        - Assistant messages: Left-aligned, no background
-        - Markdown formatting with proper spacing
-        - Inline code with bg-gray-100 background
-        - Code blocks with syntax highlighting
-        - Copy button on code blocks
-      </message_bubble>
-
-      <buttons>
-        - Primary: Orange/amber background, white text, rounded
-        - Secondary: Border style with hover fill
-        - Icon buttons: Square with hover background
-        - Disabled state: Reduced opacity, no pointer events
-      </buttons>
-
-      <inputs>
-        - Rounded borders with focus ring
-        - Textarea auto-resize
-        - Placeholder text in gray
-        - Error states in red
-        - Character counter
-      </inputs>
-
-      <cards>
-        - Subtle border or shadow
-        - Rounded corners (8px)
-        - Padding: p-4 to p-6
-        - Hover state: slight shadow increase
-      </cards>
-    </components>
-
-    <animations>
-      - Smooth transitions (150-300ms)
-      - Fade in for new messages
-      - Slide in for sidebar
-      - Typing indicator animation
-      - Loading spinner for generation
-      - Skeleton loaders for content
-    </animations>
-  </design_system>
-
-  <key_interactions>
-    <message_flow>
-      1. User types message in input field
-      2. Optional: Attach images via button
-      3. Click send or press Enter
-      4. Message appears in chat immediately
-      5. Typing indicator shows while waiting
-      6. Response streams in word by word
-      7. Code blocks render with syntax highlighting
-      8. Artifacts detected and rendered in side panel
-      9. Message complete, enable regenerate option
-    </message_flow>
-
-    <artifact_flow>
-      1. Assistant generates artifact in response
-      2. Artifact panel slides in from right
-      3. Content renders (code with highlighting or live preview)
-      4. User can edit artifact inline
-      5. "Re-prompt" button to iterate with Claude
-      6. Download or copy artifact content
-      7. Full-screen mode for detailed work
-      8. Close panel to return to chat focus
-    </artifact_flow>
-
-    <conversation_management>
-      1. Click "New Chat" to start fresh conversation
-      2. Conversations auto-save with first message
-      3. Auto-generate title from first exchange
-      4. Click title to rename inline
-      5. Drag conversations into folders
-      6. Right-click for context menu (pin, archive, delete, export)
-      7. Search filters conversations in real-time
-      8. Click conversation to switch context
-    </conversation_management>
-  </key_interactions>
-
   <implementation_steps>
-    <step number="1">
-      <title>Setup Project Foundation and Database</title>
-      <tasks>
-        - Initialize Express server with SQLite database
-        - Set up Claude API client with streaming support
-        - Create database schema with migrations
-        - Implement authentication endpoints
-        - Set up basic CORS and middleware
-        - Create health check endpoint
-      </tasks>
-    </step>
+    <feature_1>
+      <title>MCP Server Foundation</title>
+      <description>
+        Set up the basic MCP server structure with stdio transport.
 
-    <step number="2">
-      <title>Build Core Chat Interface</title>
-      <tasks>
-        - Create main layout with sidebar and chat area
-        - Implement message display with markdown rendering
-        - Add streaming message support with SSE
-        - Build input area with auto-resize textarea
-        - Add code block syntax highlighting
-        - Implement stop generation functionality
-        - Add typing indicators and loading states
-      </tasks>
-    </step>
+        Tasks:
+        - Install mcp Python SDK: pip install mcp
+        - Create mcp_server.py with Server initialization
+        - Configure stdio transport for LLM communication
+        - Implement server lifecycle handlers (startup, shutdown)
+        - Set up basic logging with Python logging module
+        - Test server startup and graceful shutdown
+      </description>
+      <priority>1</priority>
+      <category>infrastructure</category>
+      <test_steps>
+        1. Run: python mcp_server.py
+        2. Verify server starts without errors
+        3. Check that stdio transport is initialized
+        4. Send SIGTERM and verify graceful shutdown
+        5. Check logs are created and formatted correctly
+      </test_steps>
+    </feature_1>
 
-    <step number="3">
-      <title>Conversation Management</title>
-      <tasks>
-        - Create conversation list in sidebar
-        - Implement new conversation creation
-        - Add conversation switching
-        - Build conversation rename functionality
-        - Implement delete with confirmation
-        - Add conversation search
-        - Create conversation grouping by date
-      </tasks>
-    </step>
+    <feature_2>
+      <title>Configuration Management</title>
+      <description>
+        Implement configuration loading and validation.
 
-    <step number="4">
-      <title>Artifacts System</title>
-      <tasks>
-        - Build artifact detection from Claude responses
-        - Create artifact rendering panel
-        - Implement code artifact viewer
-        - Add HTML/SVG live preview
-        - Build artifact editing interface
-        - Add artifact versioning
-        - Implement full-screen artifact view
-      </tasks>
-    </step>
+        Tasks:
+        - Verify mcp_config.py exists (already created)
+        - Test MCPConfig.from_env() loads .env correctly
+        - Validate MISTRAL_API_KEY is required
+        - Test default values for optional settings
+        - Create .env.example file with all variables
+        - Document all environment variables
+      </description>
+      <priority>1</priority>
+      <category>infrastructure</category>
+      <test_steps>
+        1. Create .env file with MISTRAL_API_KEY
+        2. Run: python -c "from mcp_config import MCPConfig; MCPConfig.from_env()"
+        3. Verify config loads successfully
+        4. Remove MISTRAL_API_KEY from .env
+        5. Verify ValueError is raised
+        6. Test all default values are applied
+      </test_steps>
+    </feature_2>
 
-    <step number="5">
-      <title>Projects and Organization</title>
-      <tasks>
-        - Create projects CRUD endpoints
-        - Build project selector UI
-        - Implement project-specific custom instructions
-        - Add folder system for conversations
-        - Create drag-and-drop organization
-        - Build project settings panel
-      </tasks>
-    </step>
+    <feature_3>
+      <title>Pydantic Schemas for All Tools</title>
+      <description>
+        Define all tool input/output schemas using Pydantic models.
 
-    <step number="6">
-      <title>Advanced Features</title>
-      <tasks>
-        - Add model selection dropdown
-        - Implement temperature and parameter controls
-        - Build image upload functionality
-        - Create message editing and regeneration
-        - Add conversation branching
-        - Implement export functionality
-      </tasks>
-    </step>
+        Tasks:
+        - Create mcp_tools/ directory
+        - Create mcp_tools/__init__.py
+        - Create mcp_tools/schemas.py
+        - Define ParsePdfInput model with validation
+        - Define ParsePdfOutput model
+        - Define SearchChunksInput/Output models
+        - Define schemas for all 7 retrieval tools
+        - Add Field() validation (min/max, regex, enum)
+        - Add docstrings for JSON schema generation
+        - Verify mypy --strict passes
+      </description>
+      <priority>1</priority>
+      <category>infrastructure</category>
+      <test_steps>
+        1. Run: mypy mcp_tools/schemas.py --strict
+        2. Verify no type errors
+        3. Test ParsePdfInput validation with invalid inputs
+        4. Test ParsePdfInput validation with valid inputs
+        5. Generate JSON schema from Pydantic models
+        6. Verify all fields have descriptions
+      </test_steps>
+    </feature_3>
 
-    <step number="7">
-      <title>Settings and Customization</title>
-      <tasks>
-        - Build settings modal with tabs
-        - Implement theme switching (light/dark)
-        - Add custom instructions management
-        - Create keyboard shortcuts
-        - Build prompt library
-        - Add usage tracking dashboard
-      </tasks>
-    </step>
+    <feature_4>
+      <title>Parsing Tool - parse_pdf Implementation</title>
+      <description>
+        Implement the parse_pdf tool with optimal parameters pre-configured.
 
-    <step number="8">
-      <title>Sharing and Collaboration</title>
-      <tasks>
-        - Implement conversation sharing with tokens
-        - Create public share view
-        - Add export to multiple formats
-        - Build prompt templates
-        - Create example conversations
-      </tasks>
-    </step>
+        Tasks:
+        - Create mcp_tools/parsing_tools.py
+        - Implement parse_pdf tool handler
+        - Fixed parameters:
+          - llm_provider="mistral"
+          - llm_model="mistral-medium-latest"
+          - use_semantic_chunking=True
+          - use_ocr_annotations=True
+          - ingest_to_weaviate=True
+        - Wrapper around pdf_pipeline.process_pdf_bytes()
+        - Handle file downloads for URL inputs
+        - Return comprehensive results (metadata, costs, file paths)
+        - Add error handling and logging
+        - Register tool with MCP server
+      </description>
+      <priority>1</priority>
+      <category>functional</category>
+      <test_steps>
+        1. Mock pdf_pipeline.process_pdf_bytes()
+        2. Call parse_pdf with local PDF path
+        3. Verify fixed parameters are used
+        4. Call parse_pdf with URL
+        5. Verify file download works
+        6. Check output contains all required fields
+        7. Verify costs are tracked and returned
+      </test_steps>
+    </feature_4>
 
-    <step number="9">
-      <title>Polish and Optimization</title>
-      <tasks>
-        - Optimize for mobile responsiveness
-        - Add command palette (Cmd+K)
-        - Implement comprehensive keyboard navigation
-        - Add onboarding flow
-        - Create accessibility improvements
-        - Performance optimization and caching
-      </tasks>
-    </step>
+    <feature_5>
+      <title>Retrieval Tool - search_chunks</title>
+      <description>
+        Implement semantic search on text chunks.
+
+        Tasks:
+        - Create mcp_tools/retrieval_tools.py
+        - Implement search_chunks tool handler
+        - Weaviate near_text query on Chunk collection
+        - Support filters: author, work, language, min_similarity
+        - Handle nested object properties (work.author, work.title)
+        - Return results with similarity scores
+        - Add error handling for Weaviate connection
+        - Register tool with MCP server
+      </description>
+      <priority>2</priority>
+      <category>functional</category>
+      <test_steps>
+        1. Mock Weaviate client
+        2. Call search_chunks with query="justice"
+        3. Verify near_text query is called
+        4. Test author_filter applies correct nested filter
+        5. Test work_filter applies correct nested filter
+        6. Test min_similarity threshold works
+        7. Verify results include all metadata fields
+      </test_steps>
+    </feature_5>
+
+    <feature_6>
+      <title>Retrieval Tool - search_summaries</title>
+      <description>
+        Implement search in chapter/section summaries.
+
+        Tasks:
+        - Add search_summaries handler to retrieval_tools.py
+        - Query Summary collection with near_text
+        - Support level filters (min_level, max_level)
+        - Return summaries with hierarchical metadata
+        - Add error handling
+        - Register tool with MCP server
+      </description>
+      <priority>2</priority>
+      <category>functional</category>
+      <test_steps>
+        1. Mock Weaviate Summary collection
+        2. Call search_summaries with query
+        3. Verify near_text query on Summary
+        4. Test min_level filter
+        5. Test max_level filter
+        6. Verify results include sectionPath and concepts
+      </test_steps>
+    </feature_6>
+
+    <feature_7>
+      <title>Retrieval Tool - get_document</title>
+      <description>
+        Retrieve complete document metadata and chunks by sourceId.
+
+        Tasks:
+        - Add get_document handler to retrieval_tools.py
+        - Query Document collection by sourceId
+        - Optionally fetch related chunks
+        - Support chunk_limit parameter
+        - Return complete document data with TOC and hierarchy
+        - Add error handling for missing documents
+        - Register tool with MCP server
+      </description>
+      <priority>2</priority>
+      <category>functional</category>
+      <test_steps>
+        1. Mock Weaviate Document collection
+        2. Call get_document with valid sourceId
+        3. Verify Document query by sourceId
+        4. Test include_chunks=True fetches chunks
+        5. Test include_chunks=False skips chunks
+        6. Test chunk_limit parameter
+        7. Verify error on missing document
+      </test_steps>
+    </feature_7>
+
+    <feature_8>
+      <title>Retrieval Tool - list_documents</title>
+      <description>
+        List all documents with filtering and pagination.
+
+        Tasks:
+        - Add list_documents handler to retrieval_tools.py
+        - Query Document collection with filters
+        - Support author, work, language filters
+        - Implement pagination (limit, offset)
+        - Return document summaries with counts
+        - Add error handling
+        - Register tool with MCP server
+      </description>
+      <priority>2</priority>
+      <category>functional</category>
+      <test_steps>
+        1. Mock Weaviate Document collection
+        2. Call list_documents without filters
+        3. Verify all documents returned
+        4. Test author_filter
+        5. Test work_filter
+        6. Test pagination (limit=10, offset=5)
+        7. Verify total count is accurate
+      </test_steps>
+    </feature_8>
+
+    <feature_9>
+      <title>Retrieval Tool - get_chunks_by_document</title>
+      <description>
+        Retrieve all chunks for a document in sequential order.
+
+        Tasks:
+        - Add get_chunks_by_document handler to retrieval_tools.py
+        - Query Chunk collection filtered by document.sourceId
+        - Order results by orderIndex
+        - Support pagination (limit, offset)
+        - Support section_filter for specific sections
+        - Return ordered chunks with document metadata
+        - Add error handling
+        - Register tool with MCP server
+      </description>
+      <priority>2</priority>
+      <category>functional</category>
+      <test_steps>
+        1. Mock Weaviate Chunk collection
+        2. Call get_chunks_by_document with document_id
+        3. Verify filter by document.sourceId
+        4. Verify ordering by orderIndex
+        5. Test pagination
+        6. Test section_filter parameter
+        7. Verify document metadata is included
+      </test_steps>
+    </feature_9>
+
+    <feature_10>
+      <title>Retrieval Tool - filter_by_author</title>
+      <description>
+        Get all works and documents by a specific author.
+
+        Tasks:
+        - Add filter_by_author handler to retrieval_tools.py
+        - Query Work collection by author
+        - Fetch related Documents for each work
+        - Optionally aggregate chunk counts
+        - Return hierarchical structure (works → documents)
+        - Add error handling
+        - Register tool with MCP server
+      </description>
+      <priority>2</priority>
+      <category>functional</category>
+      <test_steps>
+        1. Mock Weaviate Work and Document collections
+        2. Call filter_by_author with author="Platon"
+        3. Verify Work query by author
+        4. Verify Documents are fetched for each work
+        5. Test include_chunks=True aggregates counts
+        6. Test include_chunks=False skips counts
+        7. Verify hierarchical structure in output
+      </test_steps>
+    </feature_10>
+
+    <feature_11>
+      <title>Retrieval Tool - delete_document</title>
+      <description>
+        Delete a document and all its chunks from Weaviate.
+
+        Tasks:
+        - Add delete_document handler to retrieval_tools.py
+        - Use weaviate_ingest.delete_document_chunks()
+        - Require explicit confirmation flag
+        - Return deletion statistics (chunks, summaries, document)
+        - Add safety checks
+        - Add error handling
+        - Register tool with MCP server
+      </description>
+      <priority>3</priority>
+      <category>functional</category>
+      <test_steps>
+        1. Mock weaviate_ingest.delete_document_chunks()
+        2. Call delete_document without confirm=True
+        3. Verify error is raised
+        4. Call delete_document with confirm=True
+        5. Verify delete_document_chunks is called
+        6. Verify deletion statistics are returned
+        7. Test error handling for missing document
+      </test_steps>
+    </feature_11>
+
+    <feature_12>
+      <title>Error Handling & Structured Logging</title>
+      <description>
+        Implement comprehensive error handling and logging across all tools.
+
+        Tasks:
+        - Define custom exception classes:
+          - WeaviateConnectionError
+          - PDFProcessingError
+          - ValidationError
+          - MCPToolError
+        - Add try-except in all tool handlers
+        - Convert Python exceptions to MCP error responses
+        - Implement structured JSON logging
+        - Log all tool invocations (name, inputs, duration, costs)
+        - Log Weaviate queries and results
+        - Configure log level from environment
+        - Never expose sensitive data in logs
+      </description>
+      <priority>2</priority>
+      <category>infrastructure</category>
+      <test_steps>
+        1. Test WeaviateConnectionError is raised on connection failure
+        2. Verify error is converted to MCP error format
+        3. Test PDFProcessingError on invalid PDF
+        4. Verify all errors are logged with context
+        5. Check logs contain tool name, inputs, duration
+        6. Verify sensitive data (API keys) not in logs
+        7. Test LOG_LEVEL environment variable works
+      </test_steps>
+    </feature_12>
+
+    <feature_13>
+      <title>MCP Server Documentation</title>
+      <description>
+        Create comprehensive documentation for MCP server usage.
+
+        Tasks:
+        - Create MCP_README.md with:
+          - Overview of capabilities
+          - Installation instructions
+          - Environment variable configuration
+          - Tool descriptions with examples
+          - Claude Desktop integration guide
+          - Troubleshooting section
+        - Add docstrings to all tool handlers
+        - Create .env.example file
+        - Document error codes and messages
+        - Add usage examples for each tool
+      </description>
+      <priority>3</priority>
+      <category>documentation</category>
+      <test_steps>
+        1. Review MCP_README.md for completeness
+        2. Verify installation instructions work
+        3. Check all tools documented with examples
+        4. Test Claude Desktop integration steps
+        5. Verify .env.example has all variables
+        6. Check troubleshooting covers common issues
+      </test_steps>
+    </feature_13>
+
+    <feature_14>
+      <title>Unit Tests - Parsing Tool</title>
+      <description>
+        Implement unit tests for parse_pdf tool.
+
+        Tasks:
+        - Create tests/mcp/ directory
+        - Create tests/mcp/conftest.py with fixtures
+        - Create tests/mcp/test_parsing_tools.py
+        - Test parse_pdf with valid PDF path
+        - Test parse_pdf with URL (mock download)
+        - Test parse_pdf error handling
+        - Mock pdf_pipeline.process_pdf_bytes()
+        - Verify fixed parameters are used
+        - Test cost tracking in output
+        - Use pytest-mock for mocking
+        - Target >80% coverage
+      </description>
+      <priority>3</priority>
+      <category>testing</category>
+      <test_steps>
+        1. Run: pytest tests/mcp/test_parsing_tools.py -v
+        2. Verify all tests pass
+        3. Run: pytest tests/mcp/test_parsing_tools.py --cov
+        4. Verify coverage >80%
+        5. Check mocks are used (no real API calls)
+        6. Verify error cases are tested
+      </test_steps>
+    </feature_14>
+
+    <feature_15>
+      <title>Unit Tests - Retrieval Tools</title>
+      <description>
+        Implement unit tests for all 7 retrieval tools.
+
+        Tasks:
+        - Create tests/mcp/test_retrieval_tools.py
+        - Test search_chunks with filters
+        - Test search_summaries with level filters
+        - Test get_document with/without chunks
+        - Test list_documents pagination
+        - Test get_chunks_by_document ordering
+        - Test filter_by_author aggregation
+        - Test delete_document confirmation
+        - Mock all Weaviate queries
+        - Test error handling for all tools
+        - Target >80% coverage
+      </description>
+      <priority>3</priority>
+      <category>testing</category>
+      <test_steps>
+        1. Run: pytest tests/mcp/test_retrieval_tools.py -v
+        2. Verify all tests pass
+        3. Run: pytest tests/mcp/test_retrieval_tools.py --cov
+        4. Verify coverage >80%
+        5. Check all 7 tools have tests
+        6. Verify error cases are tested
+        7. Check no real Weaviate connections
+      </test_steps>
+    </feature_15>
   </implementation_steps>
 
+  <testing_strategy>
+    <unit_tests>
+      <structure>
+        tests/mcp/
+        ├── conftest.py (fixtures pour mocks)
+        ├── test_parsing_tools.py
+        ├── test_retrieval_tools.py
+        ├── test_schemas.py
+        └── test_config.py
+      </structure>
+
+      <mocking>
+        - pytest-mock pour mocking
+        - Mock Mistral API responses
+        - Mock Weaviate client
+        - Fixtures pour test data
+        - Pas d'appels API réels en unit tests
+      </mocking>
+    </unit_tests>
+
+    <no_ui_tests>
+      <critical>
+        **PAS DE TESTS PUPPETEER**
+
+        Serveur MCP sans interface web.
+        Focus sur :
+        - Fonctionnalité des outils (unit tests)
+        - Validation schemas (Pydantic)
+        - Error handling
+      </critical>
+    </no_ui_tests>
+  </testing_strategy>
+
+  <deployment>
+    <claude_desktop_integration>
+      Configuration dans ~/.config/claude/claude_desktop_config.json :
+
+      {
+        "mcpServers": {
+          "library-rag": {
+            "command": "python",
+            "args": ["C:/GitHub/linear_coding_library_rag/generations/library_rag/mcp_server.py"],
+            "env": {
+              "MISTRAL_API_KEY": "your-key-here"
+            }
+          }
+        }
+      }
+    </claude_desktop_integration>
+  </deployment>
+
   <success_criteria>
-    <functionality>
-      - Streaming chat responses work smoothly
-      - Artifact detection and rendering accurate
-      - Conversation management intuitive and reliable
-      - Project organization clear and useful
-      - Image upload and display working
-      - All CRUD operations functional
-    </functionality>
+    <functional>
+      - 8 outils fonctionnels (1 parsing + 7 retrieval)
+      - parse_pdf avec paramètres optimaux fixes
+      - Recherche sémantique Weaviate fonctionnelle
+      - Error handling complet
+    </functional>
 
-    <user_experience>
-      - Interface matches claude.ai design language
-      - Responsive on all device sizes
-      - Smooth animations and transitions
-      - Fast response times and minimal lag
-      - Intuitive navigation and workflows
-      - Clear feedback for all actions
-    </user_experience>
+    <quality>
+      - mypy --strict passes
+      - Coverage >80%
+      - Docstrings complets
+    </quality>
 
-    <technical_quality>
-      - Clean, maintainable code structure
-      - Proper error handling throughout
-      - Secure API key management
-      - Optimized database queries
-      - Efficient streaming implementation
-      - Comprehensive testing coverage
-    </technical_quality>
-
-    <design_polish>
-      - Consistent with claude.ai visual design
-      - Beautiful typography and spacing
-      - Smooth animations and micro-interactions
-      - Excellent contrast and accessibility
-      - Professional, polished appearance
-      - Dark mode fully implemented
-    </design_polish>
+    <performance>
+      - parse_pdf : <2min par document
+      - search_chunks : <1s
+    </performance>
   </success_criteria>
+
+  <cost_tracking>
+    <total_estimate>
+      Document 300 pages avec configuration optimale :
+      - OCR : ~2.70€
+      - LLM : ~0.15€
+      - Total : ~2.85€
+
+      Retourné dans parse_pdf output pour tracking.
+    </total_estimate>
+  </cost_tracking>
 </project_specification>
diff --git a/prompts/app_spec_language_selection.completion.txt b/prompts/app_spec_language_selection.completion.txt
deleted file mode 100644
index 656ccb9..0000000
--- a/prompts/app_spec_language_selection.completion.txt
+++ /dev/null
@@ -1,179 +0,0 @@
-<project_specification>
-  <project_name>Language selection & i18n completion (FR default, EN/FR only)</project_name>
-
-  <overview>
-    This specification complements the existing "app_spec_language_selection.txt" file.
-    It does NOT replace the original spec. Instead, it adds additional requirements and
-    corrective steps to fully complete the language selection and i18n implementation.
-
-    Main goals:
-    - Support exactly two UI languages: English ("en") and French ("fr").
-    - Make French ("fr") the default language when no preference exists.
-    - Ensure that all user-facing text is translated via the i18n system (no hardcoded strings).
-    - Align the language selector UI with the actual supported languages.
-  </overview>
-
-  <relationship_to_original_spec>
-    - The original file "app_spec_language_selection.txt" defines the initial language selection
-      feature and i18n architecture (context, translation files, etc.).
-    - This completion spec:
-      * keeps that architecture,
-      * tightens some requirements (FR as default),
-      * and adds missing work items (removal of hardcoded English strings, cleanup of extra languages).
-    - The original spec remains valid; this completion spec should be applied on top of it.
-  </relationship_to_original_spec>
-
-  <constraints>
-    - Officially supported UI languages:
-      * English ("en")
-      * French ("fr")
-    - Default language:
-      * French ("fr") MUST be the default language when there is no stored preference.
-    - No other languages (es, de, ja, etc.) are considered part of this completion scope.
-      They may be added later in a separate spec with full translation coverage.
-    - The existing i18n architecture (LanguageContext, useLanguage hook, en.json, fr.json)
-      must be reused, not replaced.
-  </constraints>
-
-  <current_state_summary>
-    - LanguageContext and useLanguage already exist and manage language + translations.
-    - en.json and fr.json exist with a significant subset of strings translated.
-    - Some components already call t('...') correctly (e.g. welcome screen, many settings labels).
-    - However:
-      * Many UI strings are still hardcoded in English in "src/App.jsx".
-      * The language selector UI mentions more languages than are actually implemented.
-      * The default language behavior is not explicitly enforced as French.
-  </current_state_summary>
-
-  <target_state>
-    - French is used as the default language for new/anonymous users.
-    - Only English and French appear in the language selector.
-    - All user-facing UI strings in "src/App.jsx" and its inline components use t('key').
-    - Every key used by the UI is defined in both en.json and fr.json.
-    - No leftover English UI text appears when French is selected.
-  </target_state>
-
-  <implementation_details>
-    <default_language>
-      - In the language context code:
-        * Ensure there is a constant DEFAULT_LANGUAGE set to "fr".
-          Example:
-            const DEFAULT_LANGUAGE = 'fr';
-      - Initial language resolution MUST follow this order:
-        1. If a valid language ("en" or "fr") is found in localStorage, use it.
-        2. Otherwise, fall back to DEFAULT_LANGUAGE = "fr".
-      - This guarantees that first-time users and users without a stored preference see the UI in French.
-    </default_language>
-
-    <supported_languages>
-      - SUPPORTED_LANGUAGES must contain exactly:
-        * { code: 'en', name: 'English', nativeName: 'English' }
-        * { code: 'fr', name: 'French', nativeName: 'Français' }
-      - The Settings > Language dropdown must iterate only over SUPPORTED_LANGUAGES.
-      - Any explicit references to "es", "de", "ja" as selectable languages must be removed
-        or commented out as "future languages" (but not shown to users).
-    </supported_languages>
-
-    <hardcoded_strings_audit>
-      - Perform a systematic audit of "src/App.jsx" to identify every user-visible English string
-        that is still hardcoded. Typical areas include:
-        * ThemePreview sample messages (e.g. “Hello! Can you help me with something?”).
-        * About section in Settings > General (product name, description, “Built with …” text).
-        * Default model description and option labels.
-        * Project modals: “Cancel”, “Save Changes”, etc.
-        * Any toasts, confirmation messages, help texts, or labels still in English.
-      - For each identified string:
-        * Define a stable translation key (e.g. "themePreview.sampleUser1",
-          "settings.defaultModelDescription", "projectModal.cancel", "projectModal.saveChanges").
-        * Add this key to both en.json and fr.json.
-    </hardcoded_strings_audit>
-
-    <refactor_to_use_t>
-      - Replace each hardcoded string with a call to the translation function, for example:
-        BEFORE:
-          <p>Hello! Can you help me with something?</p>
-        AFTER:
-          <p>{t('themePreview.sampleUser1')}</p>
-      - Ensure that:
-        * The component (or function) imports useLanguage.
-        * const { t } = useLanguage() is declared in the correct scope.
-      - Apply this systematically across:
-        * Settings / General and Appearance sections.
-        * Theme preview component.
-        * Project-related modals.
-        * Any remaining banners, tooltips, or messages defined inside App.jsx.
-    </refactor_to_use_t>
-
-    <translation_files_update>
-      - Update translations/en.json:
-        * Add all new keys with natural English text.
-      - Update translations/fr.json:
-        * Add the same keys with accurate French translations.
-      - Goal:
-        * For every key used in code, both en.json and fr.json must contain a value.
-    </translation_files_update>
-
-    <fallback_behavior>
-      - Keep existing fallback behavior in LanguageContext:
-        * If a key is missing in the current language, fall back to English.
-        * If the key is also missing in English, return the key and log a warning.
-      - However, after this completion spec is implemented:
-        * No fallback warnings should appear in normal operation, because all keys are defined.
-    </fallback_behavior>
-
-    <settings_language_section>
-      - In the Settings > General tab:
-        * The language section heading must be translated via t('settings.language').
-        * Any helper text/description for the language selector must also use t('...').
-        * The select's value is bound to the language from useLanguage.
-        * The onChange handler calls setLanguage(newLanguageCode).
-      - Expected behavior:
-        * Switching to French instantly updates the UI and saves "fr" in localStorage.
-        * Switching to English instantly updates the UI and saves "en" in localStorage.
-    </settings_language_section>
-  </implementation_details>
-
-  <testing_plan>
-    <manual_tests>
-      1. Clear the language preference from localStorage.
-      2. Load the application:
-         - Confirm that the UI is initially in French (FR as default).
-      3. Open the Settings modal and navigate to the General tab.
-         - Verify that the language selector shows only "Français" and "English".
-      4. Switch to English:
-         - Verify that Sidebar, Settings, Welcome screen, Chat area, and modals are all in English.
-      5. Refresh the page:
-         - Confirm that the UI stays in English (preference persisted).
-      6. Switch back to French and repeat quick checks to confirm all UI text is in French.
-    </manual_tests>
-
-    <coverage_checks>
-      - Check in both languages:
-        * Main/empty state (welcome screen).
-        * Chat area (input placeholder, send/stop/regenerate buttons).
-        * Sidebar (navigation sections, search placeholder, pinned/archived labels).
-        * Settings (all tabs).
-        * Project creation and edit modals.
-        * Delete/confirmation dialogs and any share/export flows.
-      - Confirm:
-        * In French, there is no remaining English UI text.
-        * In English, there is no accidental French UI text.
-    </coverage_checks>
-
-    <regression>
-      - Verify:
-        * Chat behavior is unchanged except for translated labels/text.
-        * Project operations (create/update/delete) still work.
-        * No new console errors appear when switching languages or reloading.
-    </regression>
-  </testing_plan>
-
-  <success_criteria>
-    - "app_spec_language_selection.txt" remains the original base spec.
-    - This completion spec ("app_spec_language_selection.completion.txt") is fully implemented.
-    - French is used as default language when no preference exists.
-    - Only English and French are presented in the language selector.
-    - All user-facing strings in App.jsx go through t('key') and exist in both en.json and fr.json.
-    - No stray English text is visible when the French language is selected.
-  </success_criteria>
-</project_specification>
\ No newline at end of file
diff --git a/prompts/app_spec_language_selection.txt b/prompts/app_spec_language_selection.txt
deleted file mode 100644
index 7c5f689..0000000
--- a/prompts/app_spec_language_selection.txt
+++ /dev/null
@@ -1,525 +0,0 @@
-<project_specification>
-  <project_name>Claude.ai Clone - Language Selection Bug Fix</project_name>
-
-  <overview>
-    This specification fixes a bug in the language selection functionality. The feature was
-    originally planned in the initial app_spec.txt (line 127: "Language preferences") and a UI
-    component already exists in the settings panel (App.jsx lines 1412-1419), but the functionality
-    is incomplete and non-functional.
-    
-    Currently, there is a language selector dropdown in the settings with options for English,
-    Español, Français, Deutsch, and 日本語, but it lacks:
-    - State management for the selected language
-    - Event handlers (onChange) to handle language changes
-    - A translation system (i18n)
-    - Translation files (en.json, fr.json, etc.)
-    - Language context/provider
-    - Persistence of language preference
-    
-    This bug fix will complete the implementation by adding the missing functionality so that when
-    a language is selected, the entire interface updates immediately to display all text in the
-    chosen language. The language preference should persist across sessions.
-    
-    Focus will be on English (default) and French as the primary languages, with the existing
-    UI supporting additional languages for future expansion.
-  </overview>
-
-  <current_state>
-    <existing_ui>
-      Location: src/App.jsx, lines 1412-1419
-      Component: Language selector dropdown in settings panel (General/Preferences section)
-      Current options: English (en), Español (es), Français (fr), Deutsch (de), 日本語 (ja)
-      Status: UI exists but is non-functional (no onChange handler, no state, no translations)
-    </existing_ui>
-    
-    <specification_reference>
-      Original spec: prompts/app_spec.txt, line 127
-      Mentioned as: "Language preferences" in settings_preferences section
-      Status: Feature was planned but not fully implemented
-    </specification_reference>
-  </current_state>
-
-  <safety_requirements>
-    <critical>
-      - DO NOT remove or modify the existing language selector UI (lines 1412-1419 in App.jsx)
-      - DO NOT break existing functionality when language is changed
-      - English must remain the default language
-      - Language changes should apply immediately without page refresh
-      - All translations must be complete (no missing translations)
-      - Maintain backward compatibility with existing code
-      - Language preference should be stored and persist across sessions
-      - Keep the existing dropdown structure and styling
-      - Connect the existing select element to the new translation system
-    </critical>
-  </safety_requirements>
-
-  <bug_fixes>
-    <fix_language_selection>
-      <title>Fix Language Selection Functionality</title>
-      <description>
-        Complete the implementation of the existing language selector in the settings menu.
-        The UI already exists (App.jsx lines 1412-1419) but needs to be made functional.
-        
-        The fix should:
-        - Connect the existing select element to state management
-        - Add onChange handler to the existing select element
-        - Display current selected language (load from localStorage on mount)
-        - Apply language changes immediately to the entire interface
-        - Save language preference to localStorage
-        - Persist language choice across sessions
-        
-        The existing selector is already in the correct location (General/Preferences section
-        of settings panel) and has the correct styling, so only the functionality needs to be added.
-      </description>
-      <priority>1</priority>
-      <category>bug_fix</category>
-      <type>completion_of_existing_feature</type>
-      <implementation_approach>
-        - Keep the existing select element in App.jsx (lines 1412-1419)
-        - Add useState hook to manage selected language state
-        - Add value prop to select element (bound to state)
-        - Add onChange handler to select element
-        - Load language preference from localStorage on component mount
-        - Save language preference to localStorage on change
-        - Create translation files/dictionaries for each language
-        - Implement language context/provider to manage current language
-        - Create translation utility function to retrieve translated strings
-        - Update all hardcoded text to use translation function
-        - Apply language changes reactively throughout the application
-      </implementation_approach>
-      <test_steps>
-        1. Open settings menu
-        2. Navigate to "General" or "Preferences" section
-        3. Locate the existing "Language" selector (should already be visible)
-        4. Verify the select element now has a value bound to state (not empty)
-        5. Verify default language is "English" (en) on first load
-        6. Select "Français" (fr) from the existing language dropdown
-        7. Verify onChange handler fires and updates state
-        8. Verify entire interface updates immediately to French
-        9. Check that all UI elements are translated (buttons, labels, menus)
-        10. Navigate to different pages and verify translations persist
-        11. Refresh the page and verify language preference is maintained (loaded from localStorage)
-        12. Switch back to "English" and verify interface returns to English
-        13. Test with new conversations and verify messages/UI are in selected language
-        14. Verify the existing select element styling and structure remain unchanged
-      </test_steps>
-    </fix_language_selection>
-
-    <fix_translation_system>
-      <title>Translation System Infrastructure</title>
-      <description>
-        Implement a translation system that:
-        - Stores translations for English and French
-        - Provides a translation function/utility to retrieve translated strings
-        - Supports dynamic language switching
-        - Handles missing translations gracefully (fallback to English)
-        - Organizes translations by feature/component
-        
-        Translation keys should be organized logically:
-        - Common UI elements (buttons, labels, placeholders)
-        - Settings panel
-        - Chat interface
-        - Navigation menus
-        - Error messages
-        - Success messages
-        - Tooltips and help text
-      </description>
-      <priority>1</priority>
-      <category>infrastructure</category>
-      <type>new_implementation</type>
-      <implementation_approach>
-        - Create translation files (JSON or JS objects):
-          * translations/en.json (English)
-          * translations/fr.json (French)
-        - Create translation context/provider (React Context)
-        - Create useTranslation hook for components
-        - Create translation utility function (t() or translate())
-        - Organize translations by namespace/feature
-        - Implement fallback mechanism for missing translations
-        - Ensure type safety for translation keys (TypeScript if applicable)
-      </implementation_approach>
-      <test_steps>
-        1. Verify translation files exist for both languages
-        2. Test translation function with valid keys
-        3. Test translation function with invalid keys (should fallback)
-        4. Verify all translation keys have values in both languages
-        5. Test language switching updates all components
-        6. Verify no console errors when switching languages
-      </test_steps>
-    </fix_translation_system>
-
-    <fix_ui_translations>
-      <title>Complete UI Translation Coverage</title>
-      <description>
-        Translate all user-facing text in the application to support both English and French:
-        
-        Navigation & Menus:
-        - Sidebar navigation items
-        - Menu labels
-        - Breadcrumbs
-        
-        Chat Interface:
-        - Input placeholder text
-        - Send button
-        - Message status indicators
-        - Empty state messages
-        - Loading states
-        
-        Settings:
-        - All setting section titles
-        - Setting option labels
-        - Setting descriptions
-        - Save/Cancel buttons
-        
-        Buttons & Actions:
-        - Primary action buttons
-        - Secondary buttons
-        - Delete/Remove actions
-        - Edit actions
-        - Save actions
-        
-        Messages & Notifications:
-        - Success messages
-        - Error messages
-        - Warning messages
-        - Info messages
-        
-        Forms:
-        - Form labels
-        - Input placeholders
-        - Validation messages
-        - Help text
-        
-        Modals & Dialogs:
-        - Modal titles
-        - Modal content
-        - Confirmation dialogs
-        - Cancel/Confirm buttons
-      </description>
-      <priority>1</priority>
-      <category>ui</category>
-      <type>translation_implementation</type>
-      <implementation_approach>
-        - Audit all hardcoded text in the application
-        - Replace all hardcoded strings with translation function calls
-        - Create translation keys for each text element
-        - Add French translations for all keys
-        - Test each screen/page to ensure complete translation coverage
-        - Verify no English text remains when French is selected
-      </implementation_approach>
-      <test_steps>
-        1. Set language to French
-        2. Navigate through all pages/screens
-        3. Verify every text element is translated
-        4. Check all buttons, labels, placeholders
-        5. Test all modals and dialogs
-        6. Verify form validation messages
-        7. Check error and success notifications
-        8. Verify no English text appears when French is selected
-        9. Repeat test with English to ensure nothing broke
-      </test_steps>
-    </fix_ui_translations>
-
-    <fix_language_persistence>
-      <title>Language Preference Persistence</title>
-      <description>
-        Ensure that the selected language preference is saved and persists across:
-        - Page refreshes
-        - Browser sessions
-        - Tab closures
-        - Application restarts
-        
-        The language preference should be:
-        - Stored in localStorage (client-side) or backend user preferences
-        - Loaded on application startup
-        - Applied immediately when the app loads
-        - Synchronized if user is logged in (multi-device support)
-      </description>
-      <priority>1</priority>
-      <category>persistence</category>
-      <type>bug_fix</type>
-      <implementation_approach>
-        - Save language selection to localStorage on change
-        - Load language preference on app initialization
-        - Apply saved language before rendering UI
-        - Optionally sync with backend user preferences if available
-        - Handle case where no preference is saved (default to English)
-      </implementation_approach>
-      <test_steps>
-        1. Select French language
-        2. Refresh the page
-        3. Verify interface is still in French
-        4. Close browser tab and reopen
-        5. Verify language preference persists
-        6. Clear localStorage and verify defaults to English
-        7. Select language again and verify it saves
-      </test_steps>
-    </fix_language_persistence>
-  </bug_fixes>
-
-  <implementation_notes>
-    <existing_code>
-      Location: src/App.jsx, lines 1412-1419
-      Current code:
-      ```jsx
-      <div>
-        <h3 className="text-sm font-medium text-gray-900 dark:text-gray-100 mb-3">Language</h3>
-        <select className="w-full px-3 py-2 bg-white dark:bg-gray-700 border border-gray-300 dark:border-gray-600 rounded-lg text-sm text-gray-900 dark:text-gray-100 focus:ring-2 focus:ring-claude-orange focus:border-transparent">
-          <option value="en">English</option>
-          <option value="es">Español</option>
-          <option value="fr">Français</option>
-          <option value="de">Deutsch</option>
-          <option value="ja">日本語</option>
-        </select>
-      </div>
-      ```
-      
-      Required changes:
-      - Add value={language} to select element
-      - Add onChange={(e) => setLanguage(e.target.value)} to select element
-      - Add useState for language state
-      - Load from localStorage on mount
-      - Save to localStorage on change
-    </existing_code>
-    
-    <code_structure>
-      frontend/
-        src/
-          App.jsx                      # UPDATE: Add language state and connect existing select
-        components/
-          LanguageSelector.jsx        # Optional: Extract to component if needed (NEW)
-        contexts/
-          LanguageContext.jsx         # Language context provider (NEW)
-        hooks/
-          useLanguage.js              # Hook to access language and translations (NEW)
-        utils/
-          translations.js             # Translation utility functions (NEW)
-        translations/
-          en.json                     # English translations (NEW)
-          fr.json                     # French translations (NEW)
-    </code_structure>
-    
-    <translation_structure>
-      Translation files should be organized by feature/namespace:
-      {
-        "common": {
-          "save": "Save",
-          "cancel": "Cancel",
-          "delete": "Delete",
-          "edit": "Edit",
-          ...
-        },
-        "settings": {
-          "title": "Settings",
-          "language": "Language",
-          "theme": "Theme",
-          ...
-        },
-        "chat": {
-          "placeholder": "Message Claude...",
-          "send": "Send",
-          ...
-        },
-        ...
-      }
-    </translation_structure>
-    
-    <storage_approach>
-      Store language preference in:
-      - localStorage key: "app_language" (value: "en" or "fr")
-      - Or backend user preferences if available:
-        {
-          language: "en" | "fr"
-        }
-      
-      Default value: "en" (English)
-    </storage_approach>
-    
-    <translation_function>
-      Example implementation:
-      - useTranslation() hook returns { t, language, setLanguage }
-      - t(key) function retrieves translation for current language
-      - t("common.save") returns "Save" (en) or "Enregistrer" (fr)
-      - Supports nested keys: t("settings.general.title")
-      - Falls back to English if translation missing
-    </translation_function>
-    
-    <safety_guidelines>
-      - Keep all existing functionality intact
-      - Default to English if no language preference set
-      - Gracefully handle missing translations (fallback to English)
-      - Ensure language changes don't cause re-renders that break functionality
-      - Test thoroughly to ensure no English text remains when French is selected
-      - Maintain code readability with clear translation key naming
-    </safety_guidelines>
-  </implementation_notes>
-
-  <ui_components>
-    <language_selector>
-      <description>Language selector component in settings (ALREADY EXISTS - needs functionality)</description>
-      <location>Settings > General/Preferences section (App.jsx lines 1412-1419)</location>
-      <current_state>
-        - UI exists with dropdown/select element
-        - Has 5 language options: English (en), Español (es), Français (fr), Deutsch (de), 日本語 (ja)
-        - Styling is already correct
-        - Missing: value binding, onChange handler, state management
-      </current_state>
-      <required_changes>
-        - Add value prop bound to language state
-        - Add onChange handler to update language state
-        - Connect to translation system
-        - Add persistence (localStorage)
-      </required_changes>
-      <display>
-        - Keep existing dropdown/select element (no UI changes needed)
-        - Shows current selection (via value prop)
-        - Updates interface immediately on change (via onChange)
-      </display>
-    </language_selector>
-  </ui_components>
-
-  <translation_coverage>
-    <required_translations>
-      All text visible to users must be translated:
-      - Navigation menu items
-      - Page titles and headers
-      - Button labels
-      - Form labels and placeholders
-      - Input field labels
-      - Error messages
-      - Success messages
-      - Tooltips
-      - Help text
-      - Modal titles and content
-      - Dialog confirmations
-      - Empty states
-      - Loading states
-      - Settings labels and descriptions
-      - Chat interface elements
-    </required_translations>
-    
-    <translation_examples>
-      English -> French:
-      - "Settings" -> "Paramètres"
-      - "Save" -> "Enregistrer"
-      - "Cancel" -> "Annuler"
-      - "Delete" -> "Supprimer"
-      - "Language" -> "Langue"
-      - "Theme" -> "Thème"
-      - "Send" -> "Envoyer"
-      - "New Conversation" -> "Nouvelle conversation"
-      - "Message Claude..." -> "Message à Claude..."
-    </translation_examples>
-  </translation_coverage>
-
-  <api_endpoints>
-    <if_backend_storage>
-      If storing language preference in backend:
-      - GET /api/user/preferences - Get user preferences (includes language)
-      - PUT /api/user/preferences - Update user preferences (includes language)
-      - GET /api/user/preferences/language - Get language preference only
-      - PUT /api/user/preferences/language - Update language preference only
-    </if_backend_storage>
-    
-    <note>
-      If using localStorage only, no API endpoints needed.
-      Backend storage is optional but recommended for multi-device sync.
-    </note>
-  </api_endpoints>
-
-  <accessibility_requirements>
-    - Language selector must be keyboard navigable
-    - Language changes must be announced to screen readers
-    - Translation quality must be accurate (no machine translation errors)
-    - Text direction should be handled correctly (LTR for both languages)
-    - Font rendering should support both languages properly
-  </accessibility_requirements>
-
-  <testing_requirements>
-    <regression_tests>
-      - Verify existing functionality works in both languages
-      - Verify language change doesn't break any features
-      - Test that default language (English) still works as before
-      - Verify all existing features are accessible in both languages
-    </regression_tests>
-    
-    <feature_tests>
-      - Test language selector in settings
-      - Test immediate language change on selection
-      - Test language persistence across page refresh
-      - Test language persistence across browser sessions
-      - Test all UI elements are translated
-      - Test translation fallback for missing keys
-      - Test switching between languages multiple times
-      - Verify no English text appears when French is selected
-      - Verify all pages/screens are translated
-    </feature_tests>
-    
-    <translation_tests>
-      - Verify all translation keys have values in both languages
-      - Test translation accuracy (no machine translation errors)
-      - Verify consistent terminology across the application
-      - Test special characters and accents in French
-      - Verify text doesn't overflow UI elements in French (may be longer)
-    </translation_tests>
-    
-    <compatibility_tests>
-      - Test with different browsers (Chrome, Firefox, Safari, Edge)
-      - Test with different screen sizes (responsive design)
-      - Test language switching during active conversations
-      - Test language switching with modals open
-      - Verify language preference syncs across tabs (if applicable)
-    </compatibility_tests>
-  </testing_requirements>
-
-  <summary>
-    <bug_description>
-      The language selection feature was planned in the original specification (app_spec.txt line 127)
-      and a UI component was created (App.jsx lines 1412-1419), but the implementation is incomplete.
-      The select dropdown exists but has no functionality - it lacks state management, event handlers,
-      and a translation system.
-    </bug_description>
-    
-    <fix_scope>
-      This is a bug fix that completes the existing feature by:
-      1. Connecting the existing UI to state management
-      2. Adding the missing translation system
-      3. Implementing language persistence
-      4. Translating all UI text to support English and French
-    </fix_scope>
-    
-    <key_principle>
-      DO NOT remove or significantly modify the existing language selector UI. Only add the
-      missing functionality to make it work.
-    </key_principle>
-  </summary>
-
-  <success_criteria>
-    <functionality>
-      - Users can select language from the existing settings dropdown (English or French)
-      - Language changes apply immediately to entire interface
-      - Language preference persists across sessions
-      - All UI elements are translated when language is changed
-      - English remains the default language
-      - No functionality is broken by language changes
-      - The existing select element in App.jsx (lines 1412-1419) is now functional
-    </functionality>
-    
-    <user_experience>
-      - Language selector is easy to find in settings
-      - Language change is instant and smooth
-      - All text is properly translated (no English text in French mode)
-      - Translations are accurate and natural
-      - Interface layout works well with both languages
-    </user_experience>
-    
-    <technical>
-      - Translation system is well-organized and maintainable
-      - Translation keys are logically structured
-      - Language preference is stored reliably
-      - No performance degradation with language switching
-      - Code is clean and follows existing patterns
-      - Easy to add more languages in the future
-    </technical>
-  </success_criteria>
-</project_specification>
\ No newline at end of file
diff --git a/prompts/app_spec_library_rag_types_docs.txt b/prompts/app_spec_library_rag_types_docs.txt
new file mode 100644
index 0000000..0fe4fa6
--- /dev/null
+++ b/prompts/app_spec_library_rag_types_docs.txt
@@ -0,0 +1,679 @@
+<project_specification>
+  <project_name>Library RAG - Type Safety & Documentation Enhancement</project_name>
+
+  <overview>
+    Enhance the Library RAG application (philosophical texts indexing and semantic search) by adding
+    strict type annotations and comprehensive Google-style docstrings to all Python modules. This will
+    improve code maintainability, enable static type checking with mypy, and provide clear documentation
+    for all functions, classes, and modules.
+
+    The application is a RAG pipeline that processes PDF documents through OCR, LLM-based extraction,
+    semantic chunking, and ingestion into Weaviate vector database. It includes a Flask web interface
+    for document upload, processing, and semantic search.
+  </overview>
+
+  <technology_stack>
+    <backend>
+      <runtime>Python 3.10+</runtime>
+      <web_framework>Flask 3.0</web_framework>
+      <vector_database>Weaviate 1.34.4 with text2vec-transformers</vector_database>
+      <ocr>Mistral OCR API</ocr>
+      <llm>Ollama (local) or Mistral API</llm>
+      <type_checking>mypy with strict configuration</type_checking>
+    </backend>
+    <infrastructure>
+      <containerization>Docker Compose (Weaviate + transformers)</containerization>
+      <dependencies>weaviate-client, flask, mistralai, python-dotenv</dependencies>
+    </infrastructure>
+  </technology_stack>
+
+  <current_state>
+    <project_structure>
+      - flask_app.py: Main Flask application (640 lines)
+      - schema.py: Weaviate schema definition (383 lines)
+      - utils/: 16+ modules for PDF processing pipeline
+        - pdf_pipeline.py: Main orchestration (879 lines)
+        - mistral_client.py: OCR API client
+        - ocr_processor.py: OCR processing
+        - markdown_builder.py: Markdown generation
+        - llm_metadata.py: Metadata extraction via LLM
+        - llm_toc.py: Table of contents extraction
+        - llm_classifier.py: Section classification
+        - llm_chunker.py: Semantic chunking
+        - llm_cleaner.py: Chunk cleaning
+        - llm_validator.py: Document validation
+        - weaviate_ingest.py: Database ingestion
+        - hierarchy_parser.py: Document hierarchy parsing
+        - image_extractor.py: Image extraction from PDFs
+        - toc_extractor*.py: Various TOC extraction methods
+      - templates/: Jinja2 templates for Flask UI
+      - tests/utils2/: Minimal test coverage (3 test files)
+    </project_structure>
+
+    <issues>
+      - Inconsistent type annotations across modules (some have partial types, many have none)
+      - Missing or incomplete docstrings (no Google-style format)
+      - No mypy configuration for strict type checking
+      - Type hints missing on function parameters and return values
+      - Dict[str, Any] used extensively without proper typing
+      - No type stubs for complex nested structures
+    </issues>
+  </current_state>
+
+  <core_features>
+    <type_annotations>
+      <strict_typing>
+        - Add complete type annotations to ALL functions and methods
+        - Use proper generic types (List, Dict, Optional, Union) from typing module
+        - Add TypedDict for complex dictionary structures
+        - Add Protocol types for duck-typed interfaces
+        - Use Literal types for string constants
+        - Add ParamSpec and TypeVar where appropriate
+        - Type all class attributes and instance variables
+        - Add type annotations to lambda functions where possible
+      </strict_typing>
+
+      <mypy_configuration>
+        - Create mypy.ini with strict configuration
+        - Enable: check_untyped_defs, disallow_untyped_defs, disallow_incomplete_defs
+        - Enable: disallow_untyped_calls, disallow_untyped_decorators
+        - Enable: warn_return_any, warn_redundant_casts
+        - Enable: strict_equality, strict_optional
+        - Set python_version to 3.10
+        - Configure per-module overrides if needed for gradual migration
+      </mypy_configuration>
+
+      <type_stubs>
+        - Create TypedDict definitions for common data structures:
+          - OCR response structures
+          - Metadata dictionaries
+          - TOC entries
+          - Chunk objects
+          - Weaviate objects
+          - Pipeline results
+        - Add NewType for semantic type safety (DocumentName, ChunkId, etc.)
+        - Create Protocol types for callback functions
+      </type_stubs>
+
+      <specific_improvements>
+        - pdf_pipeline.py: Type all 10 pipeline steps, callbacks, result dictionaries
+        - flask_app.py: Type all route handlers, request/response types
+        - schema.py: Type Weaviate configuration objects
+        - llm_*.py: Type LLM request/response structures
+        - mistral_client.py: Type API client methods and responses
+        - weaviate_ingest.py: Type ingestion functions and batch operations
+      </specific_improvements>
+    </type_annotations>
+
+    <documentation>
+      <google_style_docstrings>
+        - Add comprehensive Google-style docstrings to ALL:
+          - Module-level docstrings explaining purpose and usage
+          - Class docstrings with Attributes section
+          - Function/method docstrings with Args, Returns, Raises sections
+          - Complex algorithm explanations with Examples section
+        - Include code examples for public APIs
+        - Document all exceptions that can be raised
+        - Add Notes section for important implementation details
+        - Add See Also section for related functions
+      </google_style_docstrings>
+
+      <module_documentation>
+        <utils_modules>
+          - pdf_pipeline.py: Document the 10-step pipeline, each step's purpose
+          - mistral_client.py: Document OCR API usage, cost calculation
+          - llm_metadata.py: Document metadata extraction logic
+          - llm_toc.py: Document TOC extraction strategies
+          - llm_classifier.py: Document section classification types
+          - llm_chunker.py: Document semantic vs basic chunking
+          - llm_cleaner.py: Document cleaning rules and validation
+          - llm_validator.py: Document validation criteria
+          - weaviate_ingest.py: Document ingestion process, nested objects
+          - hierarchy_parser.py: Document hierarchy building algorithm
+        </utils_modules>
+
+        <flask_app>
+          - Document all routes with request/response examples
+          - Document SSE (Server-Sent Events) implementation
+          - Document Weaviate query patterns
+          - Document upload processing workflow
+          - Document background job management
+        </flask_app>
+
+        <schema>
+          - Document Weaviate schema design decisions
+          - Document each collection's purpose and relationships
+          - Document nested object structure
+          - Document vectorization strategy
+        </schema>
+      </module_documentation>
+
+      <inline_comments>
+        - Add inline comments for complex logic only (don't over-comment)
+        - Explain WHY not WHAT (code should be self-documenting)
+        - Document performance considerations
+        - Document cost implications (OCR, LLM API calls)
+        - Document error handling strategies
+      </inline_comments>
+    </documentation>
+
+    <validation>
+      <type_checking>
+        - All modules must pass mypy --strict
+        - No # type: ignore comments without justification
+        - CI/CD should run mypy checks
+        - Type coverage should be 100%
+      </type_checking>
+
+      <documentation_quality>
+        - All public functions must have docstrings
+        - All docstrings must follow Google style
+        - Examples should be executable and tested
+        - Documentation should be clear and concise
+      </documentation_quality>
+    </validation>
+  </core_features>
+
+  <implementation_priority>
+    <critical_modules>
+      Priority 1 (Most used, most complex):
+      1. utils/pdf_pipeline.py - Main orchestration
+      2. flask_app.py - Web application entry point
+      3. utils/weaviate_ingest.py - Database operations
+      4. schema.py - Schema definition
+
+      Priority 2 (Core LLM modules):
+      5. utils/llm_metadata.py
+      6. utils/llm_toc.py
+      7. utils/llm_classifier.py
+      8. utils/llm_chunker.py
+      9. utils/llm_cleaner.py
+      10. utils/llm_validator.py
+
+      Priority 3 (OCR and parsing):
+      11. utils/mistral_client.py
+      12. utils/ocr_processor.py
+      13. utils/markdown_builder.py
+      14. utils/hierarchy_parser.py
+      15. utils/image_extractor.py
+
+      Priority 4 (Supporting modules):
+      16. utils/toc_extractor.py
+      17. utils/toc_extractor_markdown.py
+      18. utils/toc_extractor_visual.py
+      19. utils/llm_structurer.py (legacy)
+    </critical_modules>
+  </implementation_priority>
+
+  <implementation_steps>
+    <feature_1>
+      <title>Setup Type Checking Infrastructure</title>
+      <description>
+        Configure mypy with strict settings and create foundational type definitions
+      </description>
+      <tasks>
+        - Create mypy.ini configuration file with strict settings
+        - Add mypy to requirements.txt or dev dependencies
+        - Create utils/types.py module for common TypedDict definitions
+        - Define core types: OCRResponse, Metadata, TOCEntry, ChunkData, PipelineResult
+        - Add NewType definitions for semantic types: DocumentName, ChunkId, SectionPath
+        - Create Protocol types for callbacks (ProgressCallback, etc.)
+        - Document type definitions in utils/types.py module docstring
+        - Test mypy configuration on a single module to verify settings
+      </tasks>
+      <acceptance_criteria>
+        - mypy.ini exists with strict configuration
+        - utils/types.py contains all foundational types with docstrings
+        - mypy runs without errors on utils/types.py
+        - Type definitions are comprehensive and reusable
+      </acceptance_criteria>
+    </feature_1>
+
+    <feature_2>
+      <title>Add Types to PDF Pipeline Orchestration</title>
+      <description>
+        Add complete type annotations to pdf_pipeline.py (879 lines, most complex module)
+      </description>
+      <tasks>
+        - Add type annotations to all function signatures in pdf_pipeline.py
+        - Type the 10-step pipeline: OCR, Markdown, Metadata, TOC, Classify, Chunk, Clean, Validate, Weaviate
+        - Type progress_callback parameter with Protocol or Callable
+        - Add TypedDict for pipeline options dictionary
+        - Add TypedDict for pipeline result dictionary structure
+        - Type all helper functions (extract_document_metadata_legacy, etc.)
+        - Add proper return types for process_pdf_v2, process_pdf, process_pdf_bytes
+        - Fix any mypy errors that arise
+        - Verify mypy --strict passes on pdf_pipeline.py
+      </tasks>
+      <acceptance_criteria>
+        - All functions in pdf_pipeline.py have complete type annotations
+        - progress_callback is properly typed with Protocol
+        - All Dict[str, Any] replaced with TypedDict where appropriate
+        - mypy --strict pdf_pipeline.py passes with zero errors
+        - No # type: ignore comments (or justified if absolutely necessary)
+      </acceptance_criteria>
+    </feature_2>
+
+    <feature_3>
+      <title>Add Types to Flask Application</title>
+      <description>
+        Add complete type annotations to flask_app.py and type all routes
+      </description>
+      <tasks>
+        - Add type annotations to all Flask route handlers
+        - Type request.args, request.form, request.files usage
+        - Type jsonify() return values
+        - Type get_weaviate_client context manager
+        - Type get_collection_stats, get_all_chunks, search_chunks functions
+        - Add TypedDict for Weaviate query results
+        - Type background job processing functions (run_processing_job)
+        - Type SSE generator function (upload_progress)
+        - Add type hints for template rendering
+        - Verify mypy --strict passes on flask_app.py
+      </tasks>
+      <acceptance_criteria>
+        - All Flask routes have complete type annotations
+        - Request/response types are clear and documented
+        - Weaviate query functions are properly typed
+        - SSE generator is correctly typed
+        - mypy --strict flask_app.py passes with zero errors
+      </acceptance_criteria>
+    </feature_3>
+
+    <feature_4>
+      <title>Add Types to Core LLM Modules</title>
+      <description>
+        Add complete type annotations to all LLM processing modules (metadata, TOC, classifier, chunker, cleaner, validator)
+      </description>
+      <tasks>
+        - llm_metadata.py: Type extract_metadata function, return structure
+        - llm_toc.py: Type extract_toc function, TOC hierarchy structure
+        - llm_classifier.py: Type classify_sections, section types (Literal), validation functions
+        - llm_chunker.py: Type chunk_section_with_llm, chunk objects
+        - llm_cleaner.py: Type clean_chunk, is_chunk_valid functions
+        - llm_validator.py: Type validate_document, validation result structure
+        - Add TypedDict for LLM request/response structures
+        - Type provider selection ("ollama" | "mistral" as Literal)
+        - Type model names with Literal or constants
+        - Verify mypy --strict passes on all llm_*.py modules
+      </tasks>
+      <acceptance_criteria>
+        - All LLM modules have complete type annotations
+        - Section types use Literal for type safety
+        - Provider and model parameters are strongly typed
+        - LLM request/response structures use TypedDict
+        - mypy --strict passes on all llm_*.py modules with zero errors
+      </acceptance_criteria>
+    </feature_4>
+
+    <feature_5>
+      <title>Add Types to Weaviate and Database Modules</title>
+      <description>
+        Add complete type annotations to schema.py and weaviate_ingest.py
+      </description>
+      <tasks>
+        - schema.py: Type Weaviate configuration objects
+        - schema.py: Type collection property definitions
+        - weaviate_ingest.py: Type ingest_document function signature
+        - weaviate_ingest.py: Type delete_document_chunks function
+        - weaviate_ingest.py: Add TypedDict for Weaviate object structure
+        - Type batch insertion operations
+        - Type nested object references (work, document)
+        - Add proper error types for Weaviate exceptions
+        - Verify mypy --strict passes on both modules
+      </tasks>
+      <acceptance_criteria>
+        - schema.py has complete type annotations for Weaviate config
+        - weaviate_ingest.py functions are fully typed
+        - Nested object structures use TypedDict
+        - Weaviate client operations are properly typed
+        - mypy --strict passes on both modules with zero errors
+      </acceptance_criteria>
+    </feature_5>
+
+    <feature_6>
+      <title>Add Types to OCR and Parsing Modules</title>
+      <description>
+        Add complete type annotations to mistral_client.py, ocr_processor.py, markdown_builder.py, hierarchy_parser.py
+      </description>
+      <tasks>
+        - mistral_client.py: Type create_client, run_ocr, estimate_ocr_cost
+        - mistral_client.py: Add TypedDict for Mistral API response structures
+        - ocr_processor.py: Type serialize_ocr_response, OCR object structures
+        - markdown_builder.py: Type build_markdown, image_writer parameter
+        - hierarchy_parser.py: Type build_hierarchy, flatten_hierarchy functions
+        - hierarchy_parser.py: Add TypedDict for hierarchy node structure
+        - image_extractor.py: Type create_image_writer, image handling
+        - Verify mypy --strict passes on all modules
+      </tasks>
+      <acceptance_criteria>
+        - All OCR/parsing modules have complete type annotations
+        - Mistral API structures use TypedDict
+        - Hierarchy nodes are properly typed
+        - Image handling functions are typed
+        - mypy --strict passes on all modules with zero errors
+      </acceptance_criteria>
+    </feature_6>
+
+    <feature_7>
+      <title>Add Google-Style Docstrings to Core Modules</title>
+      <description>
+        Add comprehensive Google-style docstrings to pdf_pipeline.py, flask_app.py, and weaviate modules
+      </description>
+      <tasks>
+        - pdf_pipeline.py: Add module docstring explaining the V2 pipeline
+        - pdf_pipeline.py: Add docstrings to process_pdf_v2 with Args, Returns, Raises sections
+        - pdf_pipeline.py: Document each of the 10 pipeline steps in comments
+        - pdf_pipeline.py: Add Examples section showing typical usage
+        - flask_app.py: Add module docstring explaining Flask application
+        - flask_app.py: Document all routes with request/response examples
+        - flask_app.py: Document Weaviate connection management
+        - schema.py: Add module docstring explaining schema design
+        - schema.py: Document each collection's purpose and relationships
+        - weaviate_ingest.py: Document ingestion process with examples
+        - All docstrings must follow Google style format exactly
+      </tasks>
+      <acceptance_criteria>
+        - All core modules have comprehensive module-level docstrings
+        - All public functions have Google-style docstrings
+        - Args, Returns, Raises sections are complete and accurate
+        - Examples are provided for complex functions
+        - Docstrings explain WHY, not just WHAT
+      </acceptance_criteria>
+    </feature_7>
+
+    <feature_8>
+      <title>Add Google-Style Docstrings to LLM Modules</title>
+      <description>
+        Add comprehensive Google-style docstrings to all LLM processing modules
+      </description>
+      <tasks>
+        - llm_metadata.py: Document metadata extraction logic with examples
+        - llm_toc.py: Document TOC extraction strategies and fallbacks
+        - llm_classifier.py: Document section types and classification criteria
+        - llm_chunker.py: Document semantic vs basic chunking approaches
+        - llm_cleaner.py: Document cleaning rules and validation logic
+        - llm_validator.py: Document validation criteria and corrections
+        - Add Examples sections showing input/output for each function
+        - Document LLM provider differences (Ollama vs Mistral)
+        - Document cost implications in Notes sections
+        - All docstrings must follow Google style format exactly
+      </tasks>
+      <acceptance_criteria>
+        - All LLM modules have comprehensive docstrings
+        - Each function has Args, Returns, Raises sections
+        - Examples show realistic input/output
+        - Provider differences are documented
+        - Cost implications are noted where relevant
+      </acceptance_criteria>
+    </feature_8>
+
+    <feature_9>
+      <title>Add Google-Style Docstrings to OCR and Parsing Modules</title>
+      <description>
+        Add comprehensive Google-style docstrings to OCR, markdown, hierarchy, and extraction modules
+      </description>
+      <tasks>
+        - mistral_client.py: Document OCR API usage, cost calculation
+        - ocr_processor.py: Document OCR response processing
+        - markdown_builder.py: Document markdown generation strategy
+        - hierarchy_parser.py: Document hierarchy building algorithm
+        - image_extractor.py: Document image extraction process
+        - toc_extractor*.py: Document various TOC extraction methods
+        - Add Examples sections for complex algorithms
+        - Document edge cases and error handling
+        - All docstrings must follow Google style format exactly
+      </tasks>
+      <acceptance_criteria>
+        - All OCR/parsing modules have comprehensive docstrings
+        - Complex algorithms are well explained
+        - Edge cases are documented
+        - Error handling is documented
+        - Examples demonstrate typical usage
+      </acceptance_criteria>
+    </feature_9>
+
+    <feature_10>
+      <title>Final Validation and CI Integration</title>
+      <description>
+        Verify all type annotations and docstrings, integrate mypy into CI/CD
+      </description>
+      <tasks>
+        - Run mypy --strict on entire codebase, verify 100% pass rate
+        - Verify all public functions have docstrings
+        - Check docstring formatting with pydocstyle or similar tool
+        - Create GitHub Actions workflow to run mypy on every commit
+        - Update README.md with type checking instructions
+        - Update CLAUDE.md with documentation standards
+        - Create CONTRIBUTING.md with type annotation and docstring guidelines
+        - Generate API documentation with Sphinx or pdoc
+        - Fix any remaining mypy errors or missing docstrings
+      </tasks>
+      <acceptance_criteria>
+        - mypy --strict passes on entire codebase with zero errors
+        - All public functions have Google-style docstrings
+        - CI/CD runs mypy checks automatically
+        - Documentation is generated and accessible
+        - Contributing guidelines document type/docstring requirements
+      </acceptance_criteria>
+    </feature_10>
+  </implementation_steps>
+
+  <success_criteria>
+    <type_safety>
+      - 100% type coverage across all modules
+      - mypy --strict passes with zero errors
+      - No # type: ignore comments without justification
+      - All Dict[str, Any] replaced with TypedDict where appropriate
+      - Proper use of generics, protocols, and type variables
+      - NewType used for semantic type safety
+    </type_safety>
+
+    <documentation_quality>
+      - All modules have comprehensive module-level docstrings
+      - All public functions/classes have Google-style docstrings
+      - All docstrings include Args, Returns, Raises sections
+      - Complex functions include Examples sections
+      - Cost implications documented in Notes sections
+      - Error handling clearly documented
+      - Provider differences (Ollama vs Mistral) documented
+    </documentation_quality>
+
+    <code_quality>
+      - Code is self-documenting with clear variable names
+      - Inline comments explain WHY, not WHAT
+      - Complex algorithms are well explained
+      - Performance considerations documented
+      - Security considerations documented
+    </code_quality>
+
+    <developer_experience>
+      - IDE autocomplete works perfectly with type hints
+      - Type errors caught at development time, not runtime
+      - Documentation is easily accessible in IDE
+      - API examples are executable and tested
+      - Contributing guidelines are clear and comprehensive
+    </developer_experience>
+
+    <maintainability>
+      - Refactoring is safer with type checking
+      - Function signatures are self-documenting
+      - API contracts are explicit and enforced
+      - Breaking changes are caught by type checker
+      - New developers can understand code quickly
+    </maintainability>
+  </success_criteria>
+
+  <constraints>
+    <compatibility>
+      - Must maintain backward compatibility with existing code
+      - Cannot break existing Flask routes or API contracts
+      - Weaviate schema must remain unchanged
+      - Existing tests must continue to pass
+    </compatibility>
+
+    <gradual_migration>
+      - Can use per-module mypy configuration for gradual migration
+      - Can temporarily disable strict checks on legacy modules
+      - Priority modules must be completed first
+      - Low-priority modules can be deferred
+    </gradual_migration>
+
+    <standards>
+      - All type annotations must use Python 3.10+ syntax
+      - Docstrings must follow Google style exactly (not NumPy or reStructuredText)
+      - Use typing module (List, Dict, Optional) until Python 3.9 support dropped
+      - Use from __future__ import annotations if needed for forward references
+    </standards>
+  </constraints>
+
+  <testing_strategy>
+    <type_checking>
+      - Run mypy --strict on each module after adding types
+      - Use mypy daemon (dmypy) for faster incremental checking
+      - Add mypy to pre-commit hooks
+      - CI/CD must run mypy and fail on type errors
+    </type_checking>
+
+    <documentation_validation>
+      - Use pydocstyle to validate Google-style format
+      - Use sphinx-build to generate docs and catch errors
+      - Manual review of docstring examples
+      - Verify examples are executable and correct
+    </documentation_validation>
+
+    <integration_testing>
+      - Verify existing tests still pass after type additions
+      - Add new tests for complex typed structures
+      - Test mypy configuration on sample code
+      - Verify IDE autocomplete works correctly
+    </integration_testing>
+  </testing_strategy>
+
+  <documentation_examples>
+    <module_docstring>
+      ```python
+      """
+      PDF Pipeline V2 - Intelligent document processing with LLM enhancement.
+
+      This module orchestrates a 10-step pipeline for processing PDF documents:
+      1. OCR via Mistral API
+      2. Markdown construction with images
+      3. Metadata extraction via LLM
+      4. Table of contents (TOC) extraction
+      5. Section classification
+      6. Semantic chunking
+      7. Chunk cleaning and validation
+      8. Enrichment with concepts
+      9. Validation and corrections
+      10. Ingestion into Weaviate vector database
+
+      The pipeline supports multiple LLM providers (Ollama local, Mistral API) and
+      various processing modes (skip OCR, semantic chunking, OCR annotations).
+
+      Typical usage:
+          >>> from pathlib import Path
+          >>> from utils.pdf_pipeline import process_pdf
+          >>>
+          >>> result = process_pdf(
+          ...     Path("document.pdf"),
+          ...     use_llm=True,
+          ...     llm_provider="ollama",
+          ...     ingest_to_weaviate=True,
+          ... )
+          >>> print(f"Processed {result['pages']} pages, {result['chunks_count']} chunks")
+
+      See Also:
+          mistral_client: OCR API client
+          llm_metadata: Metadata extraction
+          weaviate_ingest: Database ingestion
+      """
+      ```
+    </module_docstring>
+
+    <function_docstring>
+      ```python
+      def process_pdf_v2(
+          pdf_path: Path,
+          output_dir: Path = Path("output"),
+          *,
+          use_llm: bool = True,
+          llm_provider: Literal["ollama", "mistral"] = "ollama",
+          llm_model: Optional[str] = None,
+          skip_ocr: bool = False,
+          ingest_to_weaviate: bool = True,
+          progress_callback: Optional[ProgressCallback] = None,
+      ) -> PipelineResult:
+          """
+          Process a PDF through the complete V2 pipeline with LLM enhancement.
+
+          This function orchestrates all 10 steps of the intelligent document processing
+          pipeline, from OCR to Weaviate ingestion. It supports both local (Ollama) and
+          cloud (Mistral API) LLM providers, with optional caching via skip_ocr.
+
+          Args:
+              pdf_path: Absolute path to the PDF file to process.
+              output_dir: Base directory for output files. Defaults to "./output".
+              use_llm: Enable LLM-based processing (metadata, TOC, chunking).
+                  If False, uses basic heuristic processing.
+              llm_provider: LLM provider to use. "ollama" for local (free but slow),
+                  "mistral" for API (fast but paid).
+              llm_model: Specific model name. If None, auto-detects based on provider
+                  (qwen2.5:7b for ollama, mistral-small-latest for mistral).
+              skip_ocr: If True, reuses existing markdown file to avoid OCR cost.
+                  Requires output_dir/<doc_name>/<doc_name>.md to exist.
+              ingest_to_weaviate: If True, ingests chunks into Weaviate after processing.
+              progress_callback: Optional callback for real-time progress updates.
+                  Called with (step_id, status, detail) for each pipeline step.
+
+          Returns:
+              Dictionary containing processing results with the following keys:
+                  - success (bool): True if processing completed without errors
+                  - document_name (str): Name of the processed document
+                  - pages (int): Number of pages in the PDF
+                  - chunks_count (int): Number of chunks generated
+                  - cost_ocr (float): OCR cost in euros (0 if skip_ocr=True)
+                  - cost_llm (float): LLM API cost in euros (0 if provider=ollama)
+                  - cost_total (float): Total cost (ocr + llm)
+                  - metadata (dict): Extracted metadata (title, author, etc.)
+                  - toc (list): Hierarchical table of contents
+                  - files (dict): Paths to generated files (markdown, chunks, etc.)
+
+          Raises:
+              FileNotFoundError: If pdf_path does not exist.
+              ValueError: If skip_ocr=True but markdown file not found.
+              RuntimeError: If Weaviate connection fails during ingestion.
+
+          Examples:
+              Basic usage with Ollama (free):
+              >>> result = process_pdf_v2(
+              ...     Path("platon_menon.pdf"),
+              ...     llm_provider="ollama"
+              ... )
+              >>> print(f"Cost: {result['cost_total']:.4f}€")
+              Cost: 0.0270€  # OCR only
+
+              With Mistral API (faster):
+              >>> result = process_pdf_v2(
+              ...     Path("platon_menon.pdf"),
+              ...     llm_provider="mistral",
+              ...     llm_model="mistral-small-latest"
+              ... )
+
+              Skip OCR to avoid cost:
+              >>> result = process_pdf_v2(
+              ...     Path("platon_menon.pdf"),
+              ...     skip_ocr=True,  # Reuses existing markdown
+              ...     ingest_to_weaviate=False
+              ... )
+
+          Notes:
+              - OCR cost: ~0.003€/page (standard), ~0.009€/page (with annotations)
+              - LLM cost: Free with Ollama, variable with Mistral API
+              - Processing time: ~30s/page with Ollama, ~5s/page with Mistral
+              - Weaviate must be running (docker-compose up -d) before ingestion
+          """
+      ```
+    </function_docstring>
+  </documentation_examples>
+</project_specification>
diff --git a/prompts/app_spec_mistral_extensible.txt b/prompts/app_spec_mistral_extensible.txt
deleted file mode 100644
index 0abcc86..0000000
--- a/prompts/app_spec_mistral_extensible.txt
+++ /dev/null
@@ -1,448 +0,0 @@
-<project_specification>
-  <project_name>Claude.ai Clone - Multi-Provider Support (Mistral + Extensible)</project_name>
-
-  <overview>
-    This specification adds Mistral AI model support AND creates an extensible provider architecture
-    that makes it easy to add additional AI providers (OpenAI, Gemini, etc.) in the future.
-    This uses the "Open/Closed Principle" - open for extension, closed for modification.
-    
-    All changes are additive and backward-compatible. Existing Claude functionality remains unchanged.
-  </overview>
-
-  <safety_requirements>
-    <critical>
-      - DO NOT modify existing Claude API integration code directly
-      - DO NOT change existing model selection logic for Claude models
-      - DO NOT modify existing database schema without safe migrations
-      - DO NOT break existing conversations or messages
-      - All new code must be in separate files/modules when possible
-      - Test thoroughly before marking issues as complete
-      - Maintain backward compatibility at all times
-      - Refactor Claude code to use BaseProvider WITHOUT changing functionality
-    </critical>
-  </safety_requirements>
-
-  <architecture_design>
-    <provider_pattern>
-      Create an abstract provider interface that all AI providers implement:
-      - BaseProvider (abstract class/interface) - defines common interface
-      - ClaudeProvider (existing code refactored to extend BaseProvider)
-      - MistralProvider (new, extends BaseProvider)
-      - OpenAIProvider (future, extends BaseProvider - easy to add)
-      - GeminiProvider (future, extends BaseProvider - easy to add)
-    </provider_pattern>
-    
-    <benefits>
-      - Easy to add new providers without modifying existing code
-      - Consistent interface across all providers
-      - Isolated error handling per provider
-      - Unified model selection UI
-      - Shared functionality (streaming, error handling, logging)
-      - Future-proof architecture
-    </benefits>
-  </architecture_design>
-
-  <new_features>
-    <feature_provider_architecture>
-      <title>Extensible Provider Architecture (Foundation)</title>
-      <description>
-        Create a provider abstraction layer that allows easy addition of multiple AI providers.
-        This is the foundation that makes adding OpenAI, Gemini, etc. trivial in the future.
-        
-        BaseProvider abstract class should define:
-        - sendMessage(messages, options) -&gt; Promise&lt;response&gt;
-        - streamMessage(messages, options) -&gt; AsyncGenerator&lt;chunk&gt;
-        - getModels() -&gt; Promise&lt;array&gt; of available models
-        - validateApiKey(key) -&gt; Promise&lt;boolean&gt;
-        - getCapabilities() -&gt; object with provider capabilities
-        - getName() -&gt; string (provider name: 'claude', 'mistral', 'openai', etc.)
-        - getDefaultModel() -&gt; string (default model ID for this provider)
-        
-        ProviderRegistry should:
-        - Register all available providers
-        - Provide list of all providers
-        - Check which providers are configured (have API keys)
-        - Enable/disable providers
-        
-        ProviderFactory should:
-        - Create provider instances based on model ID or provider name
-        - Handle provider selection logic
-        - Route requests to correct provider
-      </description>
-      <priority>1</priority>
-      <category>functional</category>
-      <implementation_approach>
-        - Create server/providers/BaseProvider.js (abstract base class)
-        - Refactor existing Claude code to server/providers/ClaudeProvider.js (extends BaseProvider)
-        - Create server/providers/ProviderRegistry.js (manages all providers)
-        - Create server/providers/ProviderFactory.js (creates provider instances)
-        - Update existing routes to use ProviderFactory instead of direct Claude calls
-        - Keep all provider code in server/providers/ directory
-      </implementation_approach>
-      <test_steps>
-        1. Verify Claude still works after refactoring to use BaseProvider
-        2. Test that ProviderFactory creates ClaudeProvider correctly
-        3. Test that ProviderRegistry lists Claude provider
-        4. Verify error handling works correctly
-        5. Test that adding a mock provider is straightforward
-        6. Verify no regression in existing Claude functionality
-      </test_steps>
-    </feature_provider_architecture>
-
-    <feature_mistral_provider>
-      <title>Mistral Provider Implementation</title>
-      <description>
-        Implement MistralProvider extending BaseProvider. This should:
-        - Implement all BaseProvider abstract methods
-        - Handle Mistral-specific API calls (https://api.mistral.ai/v1/chat/completions)
-        - Support Mistral streaming (Server-Sent Events)
-        - Handle Mistral-specific error codes and messages
-        - Provide Mistral model list:
-          * mistral-large-latest (default)
-          * mistral-medium-latest
-          * mistral-small-latest
-          * mistral-7b-instruct
-        - Manage Mistral API authentication
-        - Return responses in unified format (same as Claude)
-      </description>
-      <priority>2</priority>
-      <category>functional</category>
-      <implementation_approach>
-        - Create server/providers/MistralProvider.js
-        - Extend BaseProvider class
-        - Implement Mistral API integration using fetch or axios
-        - Register in ProviderRegistry
-        - Use same response format as ClaudeProvider for consistency
-      </implementation_approach>
-      <test_steps>
-        1. Test MistralProvider.sendMessage() works with valid API key
-        2. Test MistralProvider.streamMessage() works
-        3. Test MistralProvider.getModels() returns correct models
-        4. Test error handling for invalid API key
-        5. Test error handling for API rate limits
-        6. Verify it integrates with ProviderFactory
-        7. Verify responses match expected format
-      </test_steps>
-    </feature_mistral_provider>
-
-    <feature_unified_model_selector>
-      <title>Unified Model Selector (All Providers)</title>
-      <description>
-        Update model selector to dynamically load models from all registered providers.
-        The selector should:
-        - Query all providers for available models via GET /api/models
-        - Group models by provider (Claude, Mistral, etc.)
-        - Display provider badges/icons next to model names
-        - Show which provider each model belongs to
-        - Filter models by provider (optional toggle)
-        - Show provider-specific capabilities (streaming, images, etc.)
-        - Only show models from providers with configured API keys
-        - Handle providers gracefully (show "Configure API key" if not set)
-      </description>
-      <priority>2</priority>
-      <category>functional</category>
-      <implementation_approach>
-        - Create API endpoint: GET /api/models (returns all models from all providers)
-        - Update frontend ModelSelector component to handle multiple providers
-        - Add provider grouping/filtering in UI
-        - Show provider badges/icons next to model names
-        - Group models by provider with collapsible sections
-        - Show provider status (configured/not configured)
-      </implementation_approach>
-      <test_steps>
-        1. Verify model selector shows Claude models (existing functionality)
-        2. Verify model selector shows Mistral models (if key configured)
-        3. Test grouping by provider works
-        4. Test filtering by provider works
-        5. Verify provider badges display correctly
-        6. Test that providers without API keys show "Configure" message
-        7. Verify selecting a model works for both providers
-      </test_steps>
-    </feature_unified_model_selector>
-
-    <feature_provider_settings>
-      <title>Multi-Provider API Key Management</title>
-      <description>
-        Create unified API key management that supports multiple providers. Users should be able to:
-        - Manage API keys for each provider separately (Claude, Mistral, OpenAI, etc.)
-        - See which providers are available
-        - See which providers are configured (have API keys)
-        - Test each provider's API key independently
-        - Enable/disable providers (hide models if key not configured)
-        - See provider status indicators (configured/not configured/error)
-        - Update or remove API keys for any provider
-        - See usage statistics per provider
-      </description>
-      <priority>2</priority>
-      <category>functional</category>
-      <implementation_approach>
-        - Create server/routes/providers.js with unified provider management
-        - Update settings UI to show provider cards (one per provider)
-        - Each provider card has:
-          * Provider name and logo/icon
-          * API key input field (masked)
-          * "Test Connection" button
-          * Status indicator (green/yellow/red)
-          * Enable/disable toggle
-        - Store keys in api_keys table with key_name = 'claude_api_key', 'mistral_api_key', etc.
-        - Use same encryption method for all providers
-      </implementation_approach>
-      <test_steps>
-        1. Configure Claude API key (verify existing functionality still works)
-        2. Configure Mistral API key
-        3. Verify both keys are stored separately
-        4. Test each provider's "Test Connection" button
-        5. Remove one key and verify only that provider's models are hidden
-        6. Verify provider status indicators update correctly
-        7. Test that disabling a provider hides its models
-      </test_steps>
-    </feature_provider_settings>
-
-    <feature_database_provider_support>
-      <title>Database Support for Multiple Providers (Future-Proof)</title>
-      <description>
-        Update database schema to support multiple providers in a future-proof way.
-        This should:
-        - Add provider field to conversations table (TEXT, default: 'claude')
-        - Add provider field to messages/usage_tracking (TEXT, default: 'claude')
-        - Use TEXT field (not ENUM) to allow easy addition of new providers without schema changes
-        - Migration should be safe, idempotent, and backward compatible
-        - All existing records default to 'claude' provider
-        - Add indexes for performance on provider queries
-      </description>
-      <priority>1</priority>
-      <category>functional</category>
-      <implementation_approach>
-        - Create migration: server/migrations/add_provider_support.sql
-        - Use TEXT field (not ENUM) for provider name (allows 'claude', 'mistral', 'openai', etc.)
-        - Default all existing records to 'claude'
-        - Add indexes on provider columns for performance
-        - Make migration idempotent (can run multiple times safely)
-        - Create rollback script if needed
-      </implementation_approach>
-      <test_steps>
-        1. Backup existing database
-        2. Run migration script
-        3. Verify all existing conversations have provider='claude'
-        4. Verify all existing messages have provider='claude' (via usage_tracking)
-        5. Create new conversation with Mistral provider
-        6. Verify provider='mistral' is saved correctly
-        7. Query conversations by provider (test index performance)
-        8. Verify existing Claude conversations still work
-        9. Test rollback script if needed
-      </test_steps>
-    </feature_database_provider_support>
-
-    <feature_unified_chat_endpoint>
-      <title>Unified Chat Endpoint (Works with Any Provider)</title>
-      <description>
-        Update chat endpoints to use ProviderFactory, making them work with any provider.
-        The endpoint should:
-        - Accept provider or model ID in request
-        - Use ProviderFactory to get correct provider
-        - Route request to appropriate provider
-        - Return unified response format
-        - Handle provider-specific errors gracefully
-        - Support streaming for all providers that support it
-      </description>
-      <priority>1</priority>
-      <category>functional</category>
-      <implementation_approach>
-        - Update POST /api/chat to use ProviderFactory
-        - Update POST /api/chat/stream to use ProviderFactory
-        - Extract provider from model ID or accept provider parameter
-        - Route to correct provider instance
-        - Return unified response format
-      </implementation_approach>
-      <test_steps>
-        1. Test POST /api/chat with Claude model (verify no regression)
-        2. Test POST /api/chat with Mistral model
-        3. Test POST /api/chat/stream with Claude (verify streaming still works)
-        4. Test POST /api/chat/stream with Mistral
-        5. Test error handling for invalid provider
-        6. Test error handling for missing API key
-      </test_steps>
-    </feature_unified_chat_endpoint>
-  </new_features>
-
-  <future_extensibility>
-    <openai_provider_example>
-      <title>How to Add OpenAI in the Future</title>
-      <description>
-        To add OpenAI support later, simply follow these steps (NO changes to existing code needed):
-        
-        1. Create server/providers/OpenAIProvider.js extending BaseProvider
-        2. Implement OpenAI API calls (https://api.openai.com/v1/chat/completions)
-        3. Register in ProviderRegistry: ProviderRegistry.register('openai', OpenAIProvider)
-        4. That's it! OpenAI models will automatically appear in model selector.
-        
-        Example OpenAIProvider structure:
-        - Extends BaseProvider
-        - Implements sendMessage() using OpenAI API
-        - Implements streamMessage() for streaming support
-        - Returns models: gpt-4, gpt-3.5-turbo, etc.
-        - Handles OpenAI-specific authentication and errors
-      </description>
-    </openai_provider_example>
-    
-    <other_providers>
-      <note>
-        Same pattern works for any AI provider:
-        - Google Gemini (GeminiProvider)
-        - Cohere (CohereProvider)
-        - Any other AI API that follows similar patterns
-        Just create a new Provider class extending BaseProvider and register it.
-      </note>
-    </other_providers>
-  </future_extensibility>
-
-  <implementation_notes>
-    <code_structure>
-      server/
-        providers/
-          BaseProvider.js          # Abstract base class (NEW)
-          ClaudeProvider.js        # Refactored Claude (extends BaseProvider)
-          MistralProvider.js       # New Mistral (extends BaseProvider)
-          ProviderRegistry.js      # Manages all providers (NEW)
-          ProviderFactory.js       # Creates provider instances (NEW)
-        routes/
-          providers.js            # Unified provider management (NEW)
-          chat.js                  # Updated to use ProviderFactory
-        migrations/
-          add_provider_support.sql # Database migration (NEW)
-    </code_structure>
-    
-    <safety_guidelines>
-      - Refactor Claude code to use BaseProvider WITHOUT changing functionality
-      - All providers are isolated - errors in one don't affect others
-      - Database changes are backward compatible (TEXT field, not ENUM)
-      - Existing conversations default to 'claude' provider
-      - Test Claude thoroughly after refactoring
-      - Use feature flags if needed to enable/disable providers
-      - Log all provider operations separately for debugging
-    </safety_guidelines>
-    
-    <error_handling>
-      - Each provider handles its own errors
-      - Provider errors should NOT affect other providers
-      - Show user-friendly error messages
-      - Log errors with provider context
-      - Don't throw unhandled exceptions
-    </error_handling>
-  </implementation_notes>
-
-  <database_changes>
-    <safe_migrations>
-      <migration_1>
-        <description>Add provider support (TEXT field for extensibility)</description>
-        <sql>
-          -- Add provider column to conversations (TEXT allows any provider name)
-          -- Default to 'claude' for backward compatibility
-          ALTER TABLE conversations 
-          ADD COLUMN provider TEXT DEFAULT 'claude';
-          
-          -- Add provider column to usage_tracking
-          ALTER TABLE usage_tracking
-          ADD COLUMN provider TEXT DEFAULT 'claude';
-          
-          -- Add indexes for performance
-          CREATE INDEX IF NOT EXISTS idx_conversations_provider 
-          ON conversations(provider);
-          
-          CREATE INDEX IF NOT EXISTS idx_usage_tracking_provider 
-          ON usage_tracking(provider);
-        </sql>
-        <rollback>
-          -- Rollback script (use with caution - may cause data issues)
-          DROP INDEX IF EXISTS idx_conversations_provider;
-          DROP INDEX IF EXISTS idx_usage_tracking_provider;
-          -- Note: SQLite doesn't support DROP COLUMN easily
-          -- Would need to recreate table without provider column
-        </rollback>
-        <note>
-          Using TEXT instead of ENUM allows adding new providers (OpenAI, Gemini, etc.)
-          without database schema changes in the future. This is future-proof.
-        </note>
-      </migration_1>
-    </safe_migrations>
-    
-    <data_integrity>
-      - All existing conversations default to provider='claude'
-      - All existing messages default to provider='claude'
-      - Migration is idempotent (can run multiple times safely)
-      - No data loss during migration
-      - Existing queries continue to work
-    </data_integrity>
-  </database_changes>
-
-  <api_endpoints>
-    <new_endpoints>
-      - GET /api/models - Get all models from all configured providers
-      - GET /api/providers - Get list of available providers and their status
-      - POST /api/providers/:provider/key - Set API key for specific provider
-      - POST /api/providers/:provider/test - Test provider API key
-      - GET /api/providers/:provider/status - Get provider configuration status
-      - DELETE /api/providers/:provider/key - Remove provider API key
-    </new_endpoints>
-    
-    <updated_endpoints>
-      - POST /api/chat - Updated to use ProviderFactory (works with any provider)
-        * Accepts: { model: 'model-id', messages: [...], ... }
-        * Provider is determined from model ID or can be specified
-      - POST /api/chat/stream - Updated to use ProviderFactory (streaming for any provider)
-        * Same interface, works with any provider that supports streaming
-    </updated_endpoints>
-  </api_endpoints>
-
-  <dependencies>
-    <backend>
-      - No new dependencies required (use native fetch for Mistral API)
-      - Optional: @mistralai/mistralai (only if provides significant value)
-      - Keep dependencies minimal to avoid conflicts
-    </backend>
-  </dependencies>
-
-  <testing_requirements>
-    <regression_tests>
-      - Verify all existing Claude functionality still works
-      - Test that existing conversations load correctly
-      - Verify Claude model selection still works
-      - Test Claude API endpoints are unaffected
-      - Verify database queries for Claude still work
-      - Test Claude streaming still works
-    </regression_tests>
-    
-    <integration_tests>
-      - Test switching between Claude and Mistral models
-      - Test conversations with different providers
-      - Test error handling doesn't affect other providers
-      - Test migration doesn't break existing data
-      - Test ProviderFactory routes correctly
-      - Test unified model selector with multiple providers
-    </integration_tests>
-    
-    <extensibility_tests>
-      - Verify adding a mock provider is straightforward
-      - Test that ProviderFactory correctly routes to providers
-      - Verify provider isolation (errors don't propagate)
-      - Test that new providers automatically appear in UI
-    </extensibility_tests>
-  </testing_requirements>
-
-  <success_criteria>
-    <functionality>
-      - Claude functionality works exactly as before (no regression)
-      - Mistral models appear in selector and work correctly
-      - Users can switch between Claude and Mistral seamlessly
-      - API key management works for both providers
-      - Database migration is safe and backward compatible
-    </functionality>
-    
-    <extensibility>
-      - Adding a new provider (like OpenAI) requires only creating one new file
-      - No changes needed to existing code when adding providers
-      - Provider architecture is documented and easy to follow
-      - Code is organized and maintainable
-    </extensibility>
-  </success_criteria>
-</project_specification>
diff --git a/prompts/app_spec_model.txt b/prompts/app_spec_model.txt
new file mode 100644
index 0000000..1e35f6d
--- /dev/null
+++ b/prompts/app_spec_model.txt
@@ -0,0 +1,681 @@
+<project_specification>
+  <project_name>Claude.ai Clone - AI Chat Interface</project_name>
+
+  <overview>
+    Build a fully functional clone of claude.ai, Anthropic's conversational AI interface. The application should
+    provide a clean, modern chat interface for interacting with Claude via the API, including features like
+    conversation management, artifact rendering, project organization, multiple model selection, and advanced
+    settings. The UI should closely match claude.ai's design using Tailwind CSS with a focus on excellent
+    user experience and responsive design.
+  </overview>
+
+  <technology_stack>
+    <api_key>
+        You can use an API key located at /tmp/api-key for testing. You will not be allowed to read this file, but you can reference it in code.
+    </api_key>
+    <frontend>
+      <framework>React with Vite</framework>
+      <styling>Tailwind CSS (via CDN)</styling>
+      <state_management>React hooks and context</state_management>
+      <routing>React Router for navigation</routing>
+      <markdown>React Markdown for message rendering</markdown>
+      <code_highlighting>Syntax highlighting for code blocks</code_highlighting>
+      <port>Only launch on port {frontend_port}</port>
+    </frontend>
+    <backend>
+      <runtime>Node.js with Express</runtime>
+      <database>SQLite with better-sqlite3</database>
+      <api_integration>Claude API for chat completions</api_integration>
+      <streaming>Server-Sent Events for streaming responses</streaming>
+    </backend>
+    <communication>
+      <api>RESTful endpoints</api>
+      <streaming>SSE for real-time message streaming</streaming>
+      <claude_api>Integration with Claude API using Anthropic SDK</claude_api>
+    </communication>
+  </technology_stack>
+
+  <prerequisites>
+    <environment_setup>
+      - Repository includes .env with VITE_ANTHROPIC_API_KEY configured
+      - Frontend dependencies pre-installed via pnpm
+      - Backend code goes in /server directory
+      - Install backend dependencies as needed
+    </environment_setup>
+  </prerequisites>
+
+  <core_features>
+    <chat_interface>
+      - Clean, centered chat layout with message bubbles
+      - Streaming message responses with typing indicator
+      - Markdown rendering with proper formatting
+      - Code blocks with syntax highlighting and copy button
+      - LaTeX/math equation rendering
+      - Image upload and display in messages
+      - Multi-turn conversations with context
+      - Message editing and regeneration
+      - Stop generation button during streaming
+      - Input field with auto-resize textarea
+      - Character count and token estimation
+      - Keyboard shortcuts (Enter to send, Shift+Enter for newline)
+    </chat_interface>
+
+    <artifacts>
+      - Artifact detection and rendering in side panel
+      - Code artifact viewer with syntax highlighting
+      - HTML/SVG preview with live rendering
+      - React component preview
+      - Mermaid diagram rendering
+      - Text document artifacts
+      - Artifact editing and re-prompting
+      - Full-screen artifact view
+      - Download artifact content
+      - Artifact versioning and history
+    </artifacts>
+
+    <conversation_management>
+      - Create new conversations
+      - Conversation list in sidebar
+      - Rename conversations
+      - Delete conversations
+      - Search conversations by title/content
+      - Pin important conversations
+      - Archive conversations
+      - Conversation folders/organization
+      - Duplicate conversation
+      - Export conversation (JSON, Markdown, PDF)
+      - Conversation timestamps (created, last updated)
+      - Unread message indicators
+    </conversation_management>
+
+    <projects>
+      - Create projects to group related conversations
+      - Project knowledge base (upload documents)
+      - Project-specific custom instructions
+      - Share projects with team (mock feature)
+      - Project settings and configuration
+      - Move conversations between projects
+      - Project templates
+      - Project analytics (usage stats)
+    </projects>
+
+    <model_selection>
+      - Model selector dropdown with the following models:
+        - Claude Sonnet 4.5 (claude-sonnet-4-5-20250929) - default
+        - Claude Haiku 4.5 (claude-haiku-4-5-20251001)
+        - Claude Opus 4.1 (claude-opus-4-1-20250805)
+      - Model capabilities display
+      - Context window indicator
+      - Model-specific pricing info (display only)
+      - Switch models mid-conversation
+      - Model comparison view
+    </model_selection>
+
+    <custom_instructions>
+      - Global custom instructions
+      - Project-specific custom instructions
+      - Conversation-specific system prompts
+      - Custom instruction templates
+      - Preview how instructions affect responses
+    </custom_instructions>
+
+    <settings_preferences>
+      - Theme selection (Light, Dark, Auto)
+      - Font size adjustment
+      - Message density (compact, comfortable, spacious)
+      - Code theme selection
+      - Language preferences
+      - Accessibility options
+      - Keyboard shortcuts reference
+      - Data export options
+      - Privacy settings
+      - API key management
+    </settings_preferences>
+
+    <advanced_features>
+      - Temperature control slider
+      - Max tokens adjustment
+      - Top-p (nucleus sampling) control
+      - System prompt override
+      - Thinking/reasoning mode toggle
+      - Multi-modal input (text + images)
+      - Voice input (optional, mock UI)
+      - Response suggestions
+      - Related prompts
+      - Conversation branching
+    </advanced_features>
+
+    <collaboration>
+      - Share conversation via link (read-only)
+      - Export conversation formats
+      - Conversation templates
+      - Prompt library
+      - Share artifacts
+      - Team workspaces (mock UI)
+    </collaboration>
+
+    <search_discovery>
+      - Search across all conversations
+      - Filter by project, date, model
+      - Prompt library with categories
+      - Example conversations
+      - Quick actions menu
+      - Command palette (Cmd/Ctrl+K)
+    </search_discovery>
+
+    <usage_tracking>
+      - Token usage display per message
+      - Conversation cost estimation
+      - Daily/monthly usage dashboard
+      - Usage limits and warnings
+      - API quota tracking
+    </usage_tracking>
+
+    <onboarding>
+      - Welcome screen for new users
+      - Feature tour highlights
+      - Example prompts to get started
+      - Quick tips and best practices
+      - Keyboard shortcuts tutorial
+    </onboarding>
+
+    <accessibility>
+      - Full keyboard navigation
+      - Screen reader support
+      - ARIA labels and roles
+      - High contrast mode
+      - Focus management
+      - Reduced motion support
+    </accessibility>
+
+    <responsive_design>
+      - Mobile-first responsive layout
+      - Touch-optimized interface
+      - Collapsible sidebar on mobile
+      - Swipe gestures for navigation
+      - Adaptive artifact display
+      - Progressive Web App (PWA) support
+    </responsive_design>
+  </core_features>
+
+  <database_schema>
+    <tables>
+      <users>
+        - id, email, name, avatar_url
+        - created_at, last_login
+        - preferences (JSON: theme, font_size, etc.)
+        - custom_instructions
+      </users>
+
+      <projects>
+        - id, user_id, name, description, color
+        - custom_instructions, knowledge_base_path
+        - created_at, updated_at
+        - is_archived, is_pinned
+      </projects>
+
+      <conversations>
+        - id, user_id, project_id, title
+        - model, created_at, updated_at, last_message_at
+        - is_archived, is_pinned, is_deleted
+        - settings (JSON: temperature, max_tokens, etc.)
+        - token_count, message_count
+      </conversations>
+
+      <messages>
+        - id, conversation_id, role (user/assistant/system)
+        - content, created_at, edited_at
+        - tokens, finish_reason
+        - images (JSON array of image data)
+        - parent_message_id (for branching)
+      </messages>
+
+      <artifacts>
+        - id, message_id, conversation_id
+        - type (code/html/svg/react/mermaid/text)
+        - title, identifier, language
+        - content, version
+        - created_at, updated_at
+      </artifacts>
+
+      <shared_conversations>
+        - id, conversation_id, share_token
+        - created_at, expires_at, view_count
+        - is_public
+      </shared_conversations>
+
+      <prompt_library>
+        - id, user_id, title, description
+        - prompt_template, category, tags (JSON)
+        - is_public, usage_count
+        - created_at, updated_at
+      </prompt_library>
+
+      <conversation_folders>
+        - id, user_id, project_id, name, parent_folder_id
+        - created_at, position
+      </conversation_folders>
+
+      <conversation_folder_items>
+        - id, folder_id, conversation_id
+      </conversation_folder_items>
+
+      <usage_tracking>
+        - id, user_id, conversation_id, message_id
+        - model, input_tokens, output_tokens
+        - cost_estimate, created_at
+      </usage_tracking>
+
+      <api_keys>
+        - id, user_id, key_name, api_key_hash
+        - created_at, last_used_at
+        - is_active
+      </api_keys>
+    </tables>
+  </database_schema>
+
+  <api_endpoints_summary>
+    <authentication>
+      - POST /api/auth/login
+      - POST /api/auth/logout
+      - GET /api/auth/me
+      - PUT /api/auth/profile
+    </authentication>
+
+    <conversations>
+      - GET /api/conversations
+      - POST /api/conversations
+      - GET /api/conversations/:id
+      - PUT /api/conversations/:id
+      - DELETE /api/conversations/:id
+      - POST /api/conversations/:id/duplicate
+      - POST /api/conversations/:id/export
+      - PUT /api/conversations/:id/archive
+      - PUT /api/conversations/:id/pin
+      - POST /api/conversations/:id/branch
+    </conversations>
+
+    <messages>
+      - GET /api/conversations/:id/messages
+      - POST /api/conversations/:id/messages
+      - PUT /api/messages/:id
+      - DELETE /api/messages/:id
+      - POST /api/messages/:id/regenerate
+      - GET /api/messages/stream (SSE endpoint)
+    </messages>
+
+    <artifacts>
+      - GET /api/conversations/:id/artifacts
+      - GET /api/artifacts/:id
+      - PUT /api/artifacts/:id
+      - DELETE /api/artifacts/:id
+      - POST /api/artifacts/:id/fork
+      - GET /api/artifacts/:id/versions
+    </artifacts>
+
+    <projects>
+      - GET /api/projects
+      - POST /api/projects
+      - GET /api/projects/:id
+      - PUT /api/projects/:id
+      - DELETE /api/projects/:id
+      - POST /api/projects/:id/knowledge
+      - GET /api/projects/:id/conversations
+      - PUT /api/projects/:id/settings
+    </projects>
+
+    <sharing>
+      - POST /api/conversations/:id/share
+      - GET /api/share/:token
+      - DELETE /api/share/:token
+      - PUT /api/share/:token/settings
+    </sharing>
+
+    <prompts>
+      - GET /api/prompts/library
+      - POST /api/prompts/library
+      - GET /api/prompts/:id
+      - PUT /api/prompts/:id
+      - DELETE /api/prompts/:id
+      - GET /api/prompts/categories
+      - GET /api/prompts/examples
+    </prompts>
+
+    <search>
+      - GET /api/search/conversations?q=query
+      - GET /api/search/messages?q=query
+      - GET /api/search/artifacts?q=query
+      - GET /api/search/prompts?q=query
+    </search>
+
+    <folders>
+      - GET /api/folders
+      - POST /api/folders
+      - PUT /api/folders/:id
+      - DELETE /api/folders/:id
+      - POST /api/folders/:id/items
+      - DELETE /api/folders/:id/items/:conversationId
+    </folders>
+
+    <usage>
+      - GET /api/usage/daily
+      - GET /api/usage/monthly
+      - GET /api/usage/by-model
+      - GET /api/usage/conversations/:id
+    </usage>
+
+    <settings>
+      - GET /api/settings
+      - PUT /api/settings
+      - GET /api/settings/custom-instructions
+      - PUT /api/settings/custom-instructions
+    </settings>
+
+    <claude_api>
+      - POST /api/claude/chat (proxy to Claude API)
+      - POST /api/claude/chat/stream (streaming proxy)
+      - GET /api/claude/models
+      - POST /api/claude/images/upload
+    </claude_api>
+  </api_endpoints_summary>
+
+  <ui_layout>
+    <main_structure>
+      - Three-column layout: sidebar (conversations), main (chat), panel (artifacts)
+      - Collapsible sidebar with resize handle
+      - Responsive breakpoints: mobile (single column), tablet (two column), desktop (three column)
+      - Persistent header with project/model selector
+      - Bottom input area with send button and options
+    </main_structure>
+
+    <sidebar_left>
+      - New chat button (prominent)
+      - Project selector dropdown
+      - Search conversations input
+      - Conversations list (grouped by date: Today, Yesterday, Previous 7 days, etc.)
+      - Folder tree view (collapsible)
+      - Settings gear icon at bottom
+      - User profile at bottom
+    </sidebar_left>
+
+    <main_chat_area>
+      - Conversation title (editable inline)
+      - Model selector badge
+      - Message history (scrollable)
+      - Welcome screen for new conversations
+      - Suggested prompts (empty state)
+      - Input area with formatting toolbar
+      - Attachment button for images
+      - Send button with loading state
+      - Stop generation button
+    </main_chat_area>
+
+    <artifacts_panel>
+      - Artifact header with title and type badge
+      - Code editor or preview pane
+      - Tabs for multiple artifacts
+      - Full-screen toggle
+      - Download button
+      - Edit/Re-prompt button
+      - Version selector
+      - Close panel button
+    </artifacts_panel>
+
+    <modals_overlays>
+      - Settings modal (tabbed interface)
+      - Share conversation modal
+      - Export options modal
+      - Project settings modal
+      - Prompt library modal
+      - Command palette overlay
+      - Keyboard shortcuts reference
+    </modals_overlays>
+  </ui_layout>
+
+  <design_system>
+    <color_palette>
+      - Primary: Orange/amber accent (#CC785C claude-style)
+      - Background: White (light mode), Dark gray (#1A1A1A dark mode)
+      - Surface: Light gray (#F5F5F5 light), Darker gray (#2A2A2A dark)
+      - Text: Near black (#1A1A1A light), Off-white (#E5E5E5 dark)
+      - Borders: Light gray (#E5E5E5 light), Dark gray (#404040 dark)
+      - Code blocks: Monaco editor theme
+    </color_palette>
+
+    <typography>
+      - Sans-serif system font stack (Inter, SF Pro, Roboto, system-ui)
+      - Headings: font-semibold
+      - Body: font-normal, leading-relaxed
+      - Code: Monospace (JetBrains Mono, Consolas, Monaco)
+      - Message text: text-base (16px), comfortable line-height
+    </typography>
+
+    <components>
+      <message_bubble>
+        - User messages: Right-aligned, subtle background
+        - Assistant messages: Left-aligned, no background
+        - Markdown formatting with proper spacing
+        - Inline code with bg-gray-100 background
+        - Code blocks with syntax highlighting
+        - Copy button on code blocks
+      </message_bubble>
+
+      <buttons>
+        - Primary: Orange/amber background, white text, rounded
+        - Secondary: Border style with hover fill
+        - Icon buttons: Square with hover background
+        - Disabled state: Reduced opacity, no pointer events
+      </buttons>
+
+      <inputs>
+        - Rounded borders with focus ring
+        - Textarea auto-resize
+        - Placeholder text in gray
+        - Error states in red
+        - Character counter
+      </inputs>
+
+      <cards>
+        - Subtle border or shadow
+        - Rounded corners (8px)
+        - Padding: p-4 to p-6
+        - Hover state: slight shadow increase
+      </cards>
+    </components>
+
+    <animations>
+      - Smooth transitions (150-300ms)
+      - Fade in for new messages
+      - Slide in for sidebar
+      - Typing indicator animation
+      - Loading spinner for generation
+      - Skeleton loaders for content
+    </animations>
+  </design_system>
+
+  <key_interactions>
+    <message_flow>
+      1. User types message in input field
+      2. Optional: Attach images via button
+      3. Click send or press Enter
+      4. Message appears in chat immediately
+      5. Typing indicator shows while waiting
+      6. Response streams in word by word
+      7. Code blocks render with syntax highlighting
+      8. Artifacts detected and rendered in side panel
+      9. Message complete, enable regenerate option
+    </message_flow>
+
+    <artifact_flow>
+      1. Assistant generates artifact in response
+      2. Artifact panel slides in from right
+      3. Content renders (code with highlighting or live preview)
+      4. User can edit artifact inline
+      5. "Re-prompt" button to iterate with Claude
+      6. Download or copy artifact content
+      7. Full-screen mode for detailed work
+      8. Close panel to return to chat focus
+    </artifact_flow>
+
+    <conversation_management>
+      1. Click "New Chat" to start fresh conversation
+      2. Conversations auto-save with first message
+      3. Auto-generate title from first exchange
+      4. Click title to rename inline
+      5. Drag conversations into folders
+      6. Right-click for context menu (pin, archive, delete, export)
+      7. Search filters conversations in real-time
+      8. Click conversation to switch context
+    </conversation_management>
+  </key_interactions>
+
+  <implementation_steps>
+    <step number="1">
+      <title>Setup Project Foundation and Database</title>
+      <tasks>
+        - Initialize Express server with SQLite database
+        - Set up Claude API client with streaming support
+        - Create database schema with migrations
+        - Implement authentication endpoints
+        - Set up basic CORS and middleware
+        - Create health check endpoint
+      </tasks>
+    </step>
+
+    <step number="2">
+      <title>Build Core Chat Interface</title>
+      <tasks>
+        - Create main layout with sidebar and chat area
+        - Implement message display with markdown rendering
+        - Add streaming message support with SSE
+        - Build input area with auto-resize textarea
+        - Add code block syntax highlighting
+        - Implement stop generation functionality
+        - Add typing indicators and loading states
+      </tasks>
+    </step>
+
+    <step number="3">
+      <title>Conversation Management</title>
+      <tasks>
+        - Create conversation list in sidebar
+        - Implement new conversation creation
+        - Add conversation switching
+        - Build conversation rename functionality
+        - Implement delete with confirmation
+        - Add conversation search
+        - Create conversation grouping by date
+      </tasks>
+    </step>
+
+    <step number="4">
+      <title>Artifacts System</title>
+      <tasks>
+        - Build artifact detection from Claude responses
+        - Create artifact rendering panel
+        - Implement code artifact viewer
+        - Add HTML/SVG live preview
+        - Build artifact editing interface
+        - Add artifact versioning
+        - Implement full-screen artifact view
+      </tasks>
+    </step>
+
+    <step number="5">
+      <title>Projects and Organization</title>
+      <tasks>
+        - Create projects CRUD endpoints
+        - Build project selector UI
+        - Implement project-specific custom instructions
+        - Add folder system for conversations
+        - Create drag-and-drop organization
+        - Build project settings panel
+      </tasks>
+    </step>
+
+    <step number="6">
+      <title>Advanced Features</title>
+      <tasks>
+        - Add model selection dropdown
+        - Implement temperature and parameter controls
+        - Build image upload functionality
+        - Create message editing and regeneration
+        - Add conversation branching
+        - Implement export functionality
+      </tasks>
+    </step>
+
+    <step number="7">
+      <title>Settings and Customization</title>
+      <tasks>
+        - Build settings modal with tabs
+        - Implement theme switching (light/dark)
+        - Add custom instructions management
+        - Create keyboard shortcuts
+        - Build prompt library
+        - Add usage tracking dashboard
+      </tasks>
+    </step>
+
+    <step number="8">
+      <title>Sharing and Collaboration</title>
+      <tasks>
+        - Implement conversation sharing with tokens
+        - Create public share view
+        - Add export to multiple formats
+        - Build prompt templates
+        - Create example conversations
+      </tasks>
+    </step>
+
+    <step number="9">
+      <title>Polish and Optimization</title>
+      <tasks>
+        - Optimize for mobile responsiveness
+        - Add command palette (Cmd+K)
+        - Implement comprehensive keyboard navigation
+        - Add onboarding flow
+        - Create accessibility improvements
+        - Performance optimization and caching
+      </tasks>
+    </step>
+  </implementation_steps>
+
+  <success_criteria>
+    <functionality>
+      - Streaming chat responses work smoothly
+      - Artifact detection and rendering accurate
+      - Conversation management intuitive and reliable
+      - Project organization clear and useful
+      - Image upload and display working
+      - All CRUD operations functional
+    </functionality>
+
+    <user_experience>
+      - Interface matches claude.ai design language
+      - Responsive on all device sizes
+      - Smooth animations and transitions
+      - Fast response times and minimal lag
+      - Intuitive navigation and workflows
+      - Clear feedback for all actions
+    </user_experience>
+
+    <technical_quality>
+      - Clean, maintainable code structure
+      - Proper error handling throughout
+      - Secure API key management
+      - Optimized database queries
+      - Efficient streaming implementation
+      - Comprehensive testing coverage
+    </technical_quality>
+
+    <design_polish>
+      - Consistent with claude.ai visual design
+      - Beautiful typography and spacing
+      - Smooth animations and micro-interactions
+      - Excellent contrast and accessibility
+      - Professional, polished appearance
+      - Dark mode fully implemented
+    </design_polish>
+  </success_criteria>
+</project_specification>
diff --git a/prompts/app_spec_template.txt b/prompts/app_spec_template.txt
deleted file mode 100644
index 677d2db..0000000
--- a/prompts/app_spec_template.txt
+++ /dev/null
@@ -1,134 +0,0 @@
-<project_specification>
-  <project_name>Votre Nom d'Application</project_name>
-
-  <overview>
-    Description complète de votre application. Expliquez :
-    - Ce que fait l'application
-    - Qui sont les utilisateurs cibles
-    - Les objectifs principaux
-    - Les fonctionnalités clés en quelques phrases
-  </overview>
-
-  <technology_stack>
-    <api_key>
-      Note: Vous pouvez utiliser une clé API située à /tmp/api-key pour les tests.
-      Vous ne pourrez pas lire ce fichier, mais vous pouvez le référencer dans le code.
-    </api_key>
-    <frontend>
-      <framework>React avec Vite</framework>
-      <styling>Tailwind CSS (via CDN)</styling>
-      <state_management>React hooks et context</state_management>
-      <routing>React Router pour la navigation</routing>
-      <port>Lancer uniquement sur le port {frontend_port}</port>
-    </frontend>
-    <backend>
-      <runtime>Node.js avec Express</runtime>
-      <database>SQLite avec better-sqlite3</database>
-      <api_integration>Intégration avec les APIs nécessaires</api_integration>
-      <streaming>Server-Sent Events pour les réponses en temps réel (si nécessaire)</streaming>
-    </backend>
-    <communication>
-      <api>Endpoints RESTful</api>
-      <streaming>SSE pour le streaming en temps réel (si nécessaire)</streaming>
-    </communication>
-  </technology_stack>
-
-  <prerequisites>
-    <environment_setup>
-      - Repository inclut .env avec les clés API configurées
-      - Dépendances frontend pré-installées via npm/pnpm
-      - Code backend dans le répertoire /server
-      - Installer les dépendances backend au besoin
-    </environment_setup>
-  </prerequisites>
-
-  <core_features>
-    <!-- 
-      IMPORTANT: Créez une balise <feature_X> pour chaque fonctionnalité.
-      L'agent initializer créera une issue Linear pour chaque feature.
-      Utilisez des numéros séquentiels: feature_1, feature_2, feature_3, etc.
-    -->
-
-    <feature_1>
-      <title>Nom de la fonctionnalité 1</title>
-      <description>
-        Description détaillée de ce que fait cette fonctionnalité.
-        Incluez les détails techniques importants, les cas d'usage, et les
-        interactions avec d'autres parties de l'application.
-      </description>
-      <priority>1</priority>
-      <category>frontend</category>
-      <test_steps>
-        1. Étape de test 1 - décrire l'action
-        2. Étape de test 2 - décrire l'action
-        3. Étape de test 3 - vérifier le résultat attendu
-        4. Étape de test 4 - tester les cas d'erreur
-      </test_steps>
-    </feature_1>
-
-    <feature_2>
-      <title>Nom de la fonctionnalité 2</title>
-      <description>
-        Description de la fonctionnalité 2...
-      </description>
-      <priority>2</priority>
-      <category>backend</category>
-      <test_steps>
-        1. Étape de test 1
-        2. Étape de test 2
-      </test_steps>
-    </feature_2>
-
-    <!-- 
-      Continuez à ajouter des features...
-      L'agent créera environ 50 issues au total, donc détaillez bien vos fonctionnalités
-      en les divisant en sous-fonctionnalités si nécessaire.
-    -->
-  </core_features>
-
-  <ui_design>
-    <!-- 
-      Optionnel: Décrivez le design UI si nécessaire
-      - Layout général
-      - Couleurs et thème
-      - Composants réutilisables
-      - Responsive design
-    -->
-  </ui_design>
-
-  <api_endpoints>
-    <!-- 
-      Optionnel: Liste des endpoints API si nécessaire
-      <endpoint>
-        <method>POST</method>
-        <path>/api/users</path>
-        <description>Créer un nouvel utilisateur</description>
-        <request_body>JSON avec email, password</request_body>
-        <response>JSON avec user_id, email</response>
-      </endpoint>
-    -->
-  </api_endpoints>
-
-  <database_schema>
-    <!-- 
-      Optionnel: Schéma de base de données si nécessaire
-      <table>
-        <name>users</name>
-        <columns>
-          <column>id INTEGER PRIMARY KEY</column>
-          <column>email TEXT UNIQUE</column>
-          <column>password_hash TEXT</column>
-          <column>created_at DATETIME</column>
-        </columns>
-      </table>
-    -->
-  </database_schema>
-
-  <deployment>
-    <!-- 
-      Optionnel: Instructions de déploiement si nécessaire
-    -->
-  </deployment>
-</project_specification>
-
-
diff --git a/prompts/app_spec_theme_customization.txt b/prompts/app_spec_theme_customization.txt
deleted file mode 100644
index f95e5d3..0000000
--- a/prompts/app_spec_theme_customization.txt
+++ /dev/null
@@ -1,403 +0,0 @@
-<project_specification>
-  <project_name>Claude.ai Clone - Advanced Theme Customization</project_name>
-
-  <overview>
-    This specification adds advanced theme customization features to the Claude.ai clone application.
-    Users will be able to customize accent colors, font sizes, message spacing, and choose from
-    preset color themes. All changes are additive and backward-compatible with existing theme functionality.
-    
-    The existing light/dark mode toggle remains unchanged and functional.
-  </overview>
-
-  <safety_requirements>
-    <critical>
-      - DO NOT modify existing light/dark mode functionality
-      - DO NOT break existing theme persistence
-      - DO NOT change existing CSS classes without ensuring backward compatibility
-      - All new theme options must be optional (defaults should match current behavior)
-      - Test thoroughly to ensure existing themes still work
-      - Maintain backward compatibility at all times
-      - New theme preferences should be stored separately from existing theme settings
-    </critical>
-  </safety_requirements>
-
-  <new_features>
-    <feature_6_theme_customization>
-      <title>Advanced Theme Customization</title>
-      <description>
-        Add advanced theme customization options. Users should be able to:
-        - Customize accent colors (beyond just light/dark mode)
-        - Choose from preset color themes (blue, green, purple, orange)
-        - Adjust font size globally (small, medium, large)
-        - Adjust message spacing (compact, comfortable, spacious)
-        - Preview theme changes before applying
-        - Save custom theme preferences
-        
-        The customization interface should be intuitive and provide real-time preview
-        of changes before they are applied. All preferences should persist across sessions.
-      </description>
-      <priority>3</priority>
-      <category>style</category>
-      <implementation_approach>
-        - Create a new "Appearance" or "Theme" section in settings
-        - Add accent color picker with preset options (blue, green, purple, orange)
-        - Add font size slider/selector (small, medium, large)
-        - Add message spacing selector (compact, comfortable, spacious)
-        - Implement preview functionality that shows changes in real-time
-        - Store theme preferences in localStorage or backend (user preferences)
-        - Apply theme using CSS custom properties (CSS variables)
-        - Ensure theme works with both light and dark modes
-      </implementation_approach>
-      <test_steps>
-        1. Open settings menu
-        2. Navigate to "Appearance" or "Theme" section
-        3. Select a different accent color (e.g., green)
-        4. Verify accent color changes are visible in preview
-        5. Adjust font size slider to "large"
-        6. Verify font size changes in preview
-        7. Adjust message spacing option to "spacious"
-        8. Verify spacing changes in preview
-        9. Click "Preview" to see changes applied temporarily
-        10. Click "Apply" to save changes permanently
-        11. Verify changes persist after page refresh
-        12. Test with both light and dark mode
-        13. Test reset to default theme
-        14. Verify existing conversations display correctly with new theme
-      </test_steps>
-    </feature_6_theme_customization>
-
-    <feature_accent_colors>
-      <title>Accent Color Customization</title>
-      <description>
-        Allow users to customize the accent color used throughout the application.
-        This includes:
-        - Primary button colors
-        - Link colors
-        - Focus states
-        - Active states
-        - Selection highlights
-        - Progress indicators
-        
-        Preset options:
-        - Blue (default, matches Claude.ai)
-        - Green
-        - Purple
-        - Orange
-        
-        Users should be able to see a preview of each color before applying.
-      </description>
-      <priority>3</priority>
-      <category>style</category>
-      <implementation_approach>
-        - Define accent colors as CSS custom properties
-        - Create color palette for each preset (light and dark variants)
-        - Add color picker UI component in settings
-        - Update all accent color usages to use CSS variables
-        - Ensure colors have proper contrast ratios for accessibility
-        - Store selected accent color in user preferences
-      </implementation_approach>
-      <test_steps>
-        1. Open theme settings
-        2. Select "Green" accent color
-        3. Verify buttons, links, and highlights use green
-        4. Switch to dark mode and verify green accent still works
-        5. Test all preset colors (blue, green, purple, orange)
-        6. Verify color persists after refresh
-        7. Test accessibility (contrast ratios)
-      </test_steps>
-    </feature_accent_colors>
-
-    <feature_font_size>
-      <title>Global Font Size Adjustment</title>
-      <description>
-        Allow users to adjust the global font size for better readability.
-        Options:
-        - Small (12px base)
-        - Medium (14px base, default)
-        - Large (16px base)
-        
-        Font size should scale proportionally across all text elements:
-        - Message text
-        - UI labels
-        - Input fields
-        - Buttons
-        - Sidebar text
-      </description>
-      <priority>3</priority>
-      <category>style</category>
-      <implementation_approach>
-        - Use CSS rem units for all font sizes
-        - Set base font size on root element
-        - Create font size presets (small, medium, large)
-        - Add font size selector in settings
-        - Store preference in user settings
-        - Ensure responsive design still works with different font sizes
-      </implementation_approach>
-      <test_steps>
-        1. Open theme settings
-        2. Select "Small" font size
-        3. Verify all text is smaller throughout the app
-        4. Select "Large" font size
-        5. Verify all text is larger throughout the app
-        6. Verify layout doesn't break with different font sizes
-        7. Test with long messages to ensure wrapping works
-        8. Verify preference persists after refresh
-      </test_steps>
-    </feature_font_size>
-
-    <feature_message_spacing>
-      <title>Message Spacing Customization</title>
-      <description>
-        Allow users to adjust the spacing between messages and within message bubbles.
-        Options:
-        - Compact: Minimal spacing (for users who prefer dense layouts)
-        - Comfortable: Default spacing (current behavior)
-        - Spacious: Increased spacing (for better readability)
-        
-        This affects:
-        - Vertical spacing between messages
-        - Padding within message bubbles
-        - Spacing between message elements (avatar, text, timestamp)
-      </description>
-      <priority>3</priority>
-      <category>style</category>
-      <implementation_approach>
-        - Define spacing scale using CSS custom properties
-        - Create spacing presets (compact, comfortable, spacious)
-        - Apply spacing to message containers and bubbles
-        - Add spacing selector in settings
-        - Store preference in user settings
-        - Ensure spacing works well with different font sizes
-      </implementation_approach>
-      <test_steps>
-        1. Open theme settings
-        2. Select "Compact" spacing
-        3. Verify messages are closer together
-        4. Select "Spacious" spacing
-        5. Verify messages have more space between them
-        6. Test with long conversations to ensure scrolling works
-        7. Verify spacing preference persists after refresh
-        8. Test with different font sizes to ensure compatibility
-      </test_steps>
-    </feature_message_spacing>
-
-    <feature_theme_preview>
-      <title>Theme Preview Functionality</title>
-      <description>
-        Allow users to preview theme changes before applying them permanently.
-        The preview should:
-        - Show a sample conversation with the new theme applied
-        - Update in real-time as settings are changed
-        - Allow users to cancel and revert to previous theme
-        - Show both light and dark mode previews if applicable
-        
-        Users should be able to:
-        - See preview immediately when changing settings
-        - Click "Apply" to save changes
-        - Click "Cancel" to discard changes
-        - Click "Reset" to return to default theme
-      </description>
-      <priority>3</priority>
-      <category>functional</category>
-      <implementation_approach>
-        - Create preview component showing sample conversation
-        - Apply theme changes temporarily to preview
-        - Store original theme state for cancel functionality
-        - Update preview in real-time as settings change
-        - Only persist changes when "Apply" is clicked
-        - Show clear visual feedback for preview vs. applied state
-      </implementation_approach>
-      <test_steps>
-        1. Open theme settings
-        2. Change accent color to green
-        3. Verify preview updates immediately
-        4. Change font size to large
-        5. Verify preview updates with new font size
-        6. Click "Cancel" and verify changes are reverted
-        7. Make changes again and click "Apply"
-        8. Verify changes are saved and applied to actual interface
-        9. Test preview with both light and dark mode
-      </test_steps>
-    </feature_theme_preview>
-  </new_features>
-
-  <implementation_notes>
-    <code_structure>
-      frontend/
-        components/
-          ThemeSettings.jsx        # New theme customization UI (NEW)
-          ThemePreview.jsx         # Preview component (NEW)
-        styles/
-          theme-variables.css      # CSS custom properties for themes (NEW)
-          accent-colors.css        # Accent color definitions (NEW)
-        hooks/
-          useTheme.js              # Updated to handle new theme options
-        utils/
-          themeStorage.js          # Theme preference persistence (NEW)
-    </code_structure>
-    
-    <css_architecture>
-      Use CSS custom properties (CSS variables) for all theme values:
-      - --accent-color-primary
-      - --accent-color-hover
-      - --font-size-base
-      - --message-spacing-vertical
-      - --message-padding
-      
-      This allows easy theme switching without JavaScript manipulation.
-    </css_architecture>
-    
-    <storage_approach>
-      Store theme preferences in:
-      - localStorage for client-side persistence
-      - Or backend user preferences table if available
-      
-      Structure:
-      {
-        accentColor: 'blue' | 'green' | 'purple' | 'orange',
-        fontSize: 'small' | 'medium' | 'large',
-        messageSpacing: 'compact' | 'comfortable' | 'spacious',
-        theme: 'light' | 'dark' (existing)
-      }
-    </storage_approach>
-    
-    <safety_guidelines>
-      - Keep existing theme functionality intact
-      - Default values should match current behavior
-      - Use feature detection for new theme features
-      - Gracefully degrade if CSS custom properties not supported
-      - Test with existing conversations and UI elements
-      - Ensure accessibility standards are maintained
-    </safety_guidelines>
-  </implementation_notes>
-
-  <ui_components>
-    <theme_settings_panel>
-      <description>Settings panel for theme customization</description>
-      <sections>
-        - Accent Color: Radio buttons or color swatches for preset colors
-        - Font Size: Slider or dropdown (small, medium, large)
-        - Message Spacing: Radio buttons (compact, comfortable, spacious)
-        - Preview: Live preview of theme changes
-        - Actions: Apply, Cancel, Reset buttons
-      </sections>
-    </theme_settings_panel>
-    
-    <theme_preview>
-      <description>Preview component showing sample conversation</description>
-      <elements>
-        - Sample user message
-        - Sample AI response
-        - Shows current accent color
-        - Shows current font size
-        - Shows current spacing
-        - Updates in real-time
-      </elements>
-    </theme_preview>
-  </ui_components>
-
-  <css_custom_properties>
-    <accent_colors>
-      Define CSS variables for each accent color preset:
-      --accent-blue: #2563eb;
-      --accent-green: #10b981;
-      --accent-purple: #8b5cf6;
-      --accent-orange: #f59e0b;
-      
-      Each should have hover, active, and focus variants.
-    </accent_colors>
-    
-    <font_sizes>
-      Define base font sizes:
-      --font-size-small: 0.75rem;   (12px)
-      --font-size-medium: 0.875rem; (14px, default)
-      --font-size-large: 1rem;      (16px)
-    </font_sizes>
-    
-    <spacing>
-      Define spacing scales:
-      --spacing-compact: 0.5rem;
-      --spacing-comfortable: 1rem; (default)
-      --spacing-spacious: 1.5rem;
-    </spacing>
-  </css_custom_properties>
-
-  <api_endpoints>
-    <if_backend_storage>
-      If storing preferences in backend:
-      - GET /api/user/preferences - Get user theme preferences
-      - PUT /api/user/preferences - Update user theme preferences
-      - GET /api/user/preferences/theme - Get theme preferences only
-    </if_backend_storage>
-    
-    <note>
-      If using localStorage only, no API endpoints needed.
-      Backend storage is optional but recommended for multi-device sync.
-    </note>
-  </api_endpoints>
-
-  <accessibility_requirements>
-    - All accent colors must meet WCAG AA contrast ratios (4.5:1 for text)
-    - Font size changes must not break screen reader compatibility
-    - Theme settings must be keyboard navigable
-    - Color choices should not be the only way to convey information
-    - Provide high contrast mode option if possible
-  </accessibility_requirements>
-
-  <testing_requirements>
-    <regression_tests>
-      - Verify existing light/dark mode toggle still works
-      - Verify existing theme persistence still works
-      - Test that default theme matches current behavior
-      - Verify existing conversations display correctly
-      - Test that all UI elements are styled correctly
-    </regression_tests>
-    
-    <feature_tests>
-      - Test each accent color preset
-      - Test each font size option
-      - Test each spacing option
-      - Test theme preview functionality
-      - Test theme persistence (localStorage/backend)
-      - Test theme reset to defaults
-      - Test theme with both light and dark modes
-      - Test theme changes in real-time
-    </feature_tests>
-    
-    <compatibility_tests>
-      - Test with different browsers (Chrome, Firefox, Safari, Edge)
-      - Test with different screen sizes (responsive design)
-      - Test with long conversations
-      - Test with different message types (text, code, artifacts)
-      - Test accessibility with screen readers
-    </compatibility_tests>
-  </testing_requirements>
-
-  <success_criteria>
-    <functionality>
-      - Users can customize accent colors from preset options
-      - Users can adjust global font size (small, medium, large)
-      - Users can adjust message spacing (compact, comfortable, spacious)
-      - Theme preview shows changes in real-time
-      - Theme preferences persist across sessions
-      - Existing light/dark mode functionality works unchanged
-      - All theme options work together harmoniously
-    </functionality>
-    
-    <user_experience>
-      - Theme customization is intuitive and easy to use
-      - Preview provides clear feedback before applying changes
-      - Changes apply smoothly without flickering
-      - Settings are easy to find and access
-      - Reset to defaults is easily accessible
-    </user_experience>
-    
-    <technical>
-      - Code is well-organized and maintainable
-      - CSS custom properties are used consistently
-      - Theme preferences are stored reliably
-      - No performance degradation with theme changes
-      - Backward compatibility is maintained
-    </technical>
-  </success_criteria>
-</project_specification>
diff --git a/prompts/app_spec_types_docs.backup.txt b/prompts/app_spec_types_docs.backup.txt
new file mode 100644
index 0000000..0fe4fa6
--- /dev/null
+++ b/prompts/app_spec_types_docs.backup.txt
@@ -0,0 +1,679 @@
+<project_specification>
+  <project_name>Library RAG - Type Safety & Documentation Enhancement</project_name>
+
+  <overview>
+    Enhance the Library RAG application (philosophical texts indexing and semantic search) by adding
+    strict type annotations and comprehensive Google-style docstrings to all Python modules. This will
+    improve code maintainability, enable static type checking with mypy, and provide clear documentation
+    for all functions, classes, and modules.
+
+    The application is a RAG pipeline that processes PDF documents through OCR, LLM-based extraction,
+    semantic chunking, and ingestion into Weaviate vector database. It includes a Flask web interface
+    for document upload, processing, and semantic search.
+  </overview>
+
+  <technology_stack>
+    <backend>
+      <runtime>Python 3.10+</runtime>
+      <web_framework>Flask 3.0</web_framework>
+      <vector_database>Weaviate 1.34.4 with text2vec-transformers</vector_database>
+      <ocr>Mistral OCR API</ocr>
+      <llm>Ollama (local) or Mistral API</llm>
+      <type_checking>mypy with strict configuration</type_checking>
+    </backend>
+    <infrastructure>
+      <containerization>Docker Compose (Weaviate + transformers)</containerization>
+      <dependencies>weaviate-client, flask, mistralai, python-dotenv</dependencies>
+    </infrastructure>
+  </technology_stack>
+
+  <current_state>
+    <project_structure>
+      - flask_app.py: Main Flask application (640 lines)
+      - schema.py: Weaviate schema definition (383 lines)
+      - utils/: 16+ modules for PDF processing pipeline
+        - pdf_pipeline.py: Main orchestration (879 lines)
+        - mistral_client.py: OCR API client
+        - ocr_processor.py: OCR processing
+        - markdown_builder.py: Markdown generation
+        - llm_metadata.py: Metadata extraction via LLM
+        - llm_toc.py: Table of contents extraction
+        - llm_classifier.py: Section classification
+        - llm_chunker.py: Semantic chunking
+        - llm_cleaner.py: Chunk cleaning
+        - llm_validator.py: Document validation
+        - weaviate_ingest.py: Database ingestion
+        - hierarchy_parser.py: Document hierarchy parsing
+        - image_extractor.py: Image extraction from PDFs
+        - toc_extractor*.py: Various TOC extraction methods
+      - templates/: Jinja2 templates for Flask UI
+      - tests/utils2/: Minimal test coverage (3 test files)
+    </project_structure>
+
+    <issues>
+      - Inconsistent type annotations across modules (some have partial types, many have none)
+      - Missing or incomplete docstrings (no Google-style format)
+      - No mypy configuration for strict type checking
+      - Type hints missing on function parameters and return values
+      - Dict[str, Any] used extensively without proper typing
+      - No type stubs for complex nested structures
+    </issues>
+  </current_state>
+
+  <core_features>
+    <type_annotations>
+      <strict_typing>
+        - Add complete type annotations to ALL functions and methods
+        - Use proper generic types (List, Dict, Optional, Union) from typing module
+        - Add TypedDict for complex dictionary structures
+        - Add Protocol types for duck-typed interfaces
+        - Use Literal types for string constants
+        - Add ParamSpec and TypeVar where appropriate
+        - Type all class attributes and instance variables
+        - Add type annotations to lambda functions where possible
+      </strict_typing>
+
+      <mypy_configuration>
+        - Create mypy.ini with strict configuration
+        - Enable: check_untyped_defs, disallow_untyped_defs, disallow_incomplete_defs
+        - Enable: disallow_untyped_calls, disallow_untyped_decorators
+        - Enable: warn_return_any, warn_redundant_casts
+        - Enable: strict_equality, strict_optional
+        - Set python_version to 3.10
+        - Configure per-module overrides if needed for gradual migration
+      </mypy_configuration>
+
+      <type_stubs>
+        - Create TypedDict definitions for common data structures:
+          - OCR response structures
+          - Metadata dictionaries
+          - TOC entries
+          - Chunk objects
+          - Weaviate objects
+          - Pipeline results
+        - Add NewType for semantic type safety (DocumentName, ChunkId, etc.)
+        - Create Protocol types for callback functions
+      </type_stubs>
+
+      <specific_improvements>
+        - pdf_pipeline.py: Type all 10 pipeline steps, callbacks, result dictionaries
+        - flask_app.py: Type all route handlers, request/response types
+        - schema.py: Type Weaviate configuration objects
+        - llm_*.py: Type LLM request/response structures
+        - mistral_client.py: Type API client methods and responses
+        - weaviate_ingest.py: Type ingestion functions and batch operations
+      </specific_improvements>
+    </type_annotations>
+
+    <documentation>
+      <google_style_docstrings>
+        - Add comprehensive Google-style docstrings to ALL:
+          - Module-level docstrings explaining purpose and usage
+          - Class docstrings with Attributes section
+          - Function/method docstrings with Args, Returns, Raises sections
+          - Complex algorithm explanations with Examples section
+        - Include code examples for public APIs
+        - Document all exceptions that can be raised
+        - Add Notes section for important implementation details
+        - Add See Also section for related functions
+      </google_style_docstrings>
+
+      <module_documentation>
+        <utils_modules>
+          - pdf_pipeline.py: Document the 10-step pipeline, each step's purpose
+          - mistral_client.py: Document OCR API usage, cost calculation
+          - llm_metadata.py: Document metadata extraction logic
+          - llm_toc.py: Document TOC extraction strategies
+          - llm_classifier.py: Document section classification types
+          - llm_chunker.py: Document semantic vs basic chunking
+          - llm_cleaner.py: Document cleaning rules and validation
+          - llm_validator.py: Document validation criteria
+          - weaviate_ingest.py: Document ingestion process, nested objects
+          - hierarchy_parser.py: Document hierarchy building algorithm
+        </utils_modules>
+
+        <flask_app>
+          - Document all routes with request/response examples
+          - Document SSE (Server-Sent Events) implementation
+          - Document Weaviate query patterns
+          - Document upload processing workflow
+          - Document background job management
+        </flask_app>
+
+        <schema>
+          - Document Weaviate schema design decisions
+          - Document each collection's purpose and relationships
+          - Document nested object structure
+          - Document vectorization strategy
+        </schema>
+      </module_documentation>
+
+      <inline_comments>
+        - Add inline comments for complex logic only (don't over-comment)
+        - Explain WHY not WHAT (code should be self-documenting)
+        - Document performance considerations
+        - Document cost implications (OCR, LLM API calls)
+        - Document error handling strategies
+      </inline_comments>
+    </documentation>
+
+    <validation>
+      <type_checking>
+        - All modules must pass mypy --strict
+        - No # type: ignore comments without justification
+        - CI/CD should run mypy checks
+        - Type coverage should be 100%
+      </type_checking>
+
+      <documentation_quality>
+        - All public functions must have docstrings
+        - All docstrings must follow Google style
+        - Examples should be executable and tested
+        - Documentation should be clear and concise
+      </documentation_quality>
+    </validation>
+  </core_features>
+
+  <implementation_priority>
+    <critical_modules>
+      Priority 1 (Most used, most complex):
+      1. utils/pdf_pipeline.py - Main orchestration
+      2. flask_app.py - Web application entry point
+      3. utils/weaviate_ingest.py - Database operations
+      4. schema.py - Schema definition
+
+      Priority 2 (Core LLM modules):
+      5. utils/llm_metadata.py
+      6. utils/llm_toc.py
+      7. utils/llm_classifier.py
+      8. utils/llm_chunker.py
+      9. utils/llm_cleaner.py
+      10. utils/llm_validator.py
+
+      Priority 3 (OCR and parsing):
+      11. utils/mistral_client.py
+      12. utils/ocr_processor.py
+      13. utils/markdown_builder.py
+      14. utils/hierarchy_parser.py
+      15. utils/image_extractor.py
+
+      Priority 4 (Supporting modules):
+      16. utils/toc_extractor.py
+      17. utils/toc_extractor_markdown.py
+      18. utils/toc_extractor_visual.py
+      19. utils/llm_structurer.py (legacy)
+    </critical_modules>
+  </implementation_priority>
+
+  <implementation_steps>
+    <feature_1>
+      <title>Setup Type Checking Infrastructure</title>
+      <description>
+        Configure mypy with strict settings and create foundational type definitions
+      </description>
+      <tasks>
+        - Create mypy.ini configuration file with strict settings
+        - Add mypy to requirements.txt or dev dependencies
+        - Create utils/types.py module for common TypedDict definitions
+        - Define core types: OCRResponse, Metadata, TOCEntry, ChunkData, PipelineResult
+        - Add NewType definitions for semantic types: DocumentName, ChunkId, SectionPath
+        - Create Protocol types for callbacks (ProgressCallback, etc.)
+        - Document type definitions in utils/types.py module docstring
+        - Test mypy configuration on a single module to verify settings
+      </tasks>
+      <acceptance_criteria>
+        - mypy.ini exists with strict configuration
+        - utils/types.py contains all foundational types with docstrings
+        - mypy runs without errors on utils/types.py
+        - Type definitions are comprehensive and reusable
+      </acceptance_criteria>
+    </feature_1>
+
+    <feature_2>
+      <title>Add Types to PDF Pipeline Orchestration</title>
+      <description>
+        Add complete type annotations to pdf_pipeline.py (879 lines, most complex module)
+      </description>
+      <tasks>
+        - Add type annotations to all function signatures in pdf_pipeline.py
+        - Type the 10-step pipeline: OCR, Markdown, Metadata, TOC, Classify, Chunk, Clean, Validate, Weaviate
+        - Type progress_callback parameter with Protocol or Callable
+        - Add TypedDict for pipeline options dictionary
+        - Add TypedDict for pipeline result dictionary structure
+        - Type all helper functions (extract_document_metadata_legacy, etc.)
+        - Add proper return types for process_pdf_v2, process_pdf, process_pdf_bytes
+        - Fix any mypy errors that arise
+        - Verify mypy --strict passes on pdf_pipeline.py
+      </tasks>
+      <acceptance_criteria>
+        - All functions in pdf_pipeline.py have complete type annotations
+        - progress_callback is properly typed with Protocol
+        - All Dict[str, Any] replaced with TypedDict where appropriate
+        - mypy --strict pdf_pipeline.py passes with zero errors
+        - No # type: ignore comments (or justified if absolutely necessary)
+      </acceptance_criteria>
+    </feature_2>
+
+    <feature_3>
+      <title>Add Types to Flask Application</title>
+      <description>
+        Add complete type annotations to flask_app.py and type all routes
+      </description>
+      <tasks>
+        - Add type annotations to all Flask route handlers
+        - Type request.args, request.form, request.files usage
+        - Type jsonify() return values
+        - Type get_weaviate_client context manager
+        - Type get_collection_stats, get_all_chunks, search_chunks functions
+        - Add TypedDict for Weaviate query results
+        - Type background job processing functions (run_processing_job)
+        - Type SSE generator function (upload_progress)
+        - Add type hints for template rendering
+        - Verify mypy --strict passes on flask_app.py
+      </tasks>
+      <acceptance_criteria>
+        - All Flask routes have complete type annotations
+        - Request/response types are clear and documented
+        - Weaviate query functions are properly typed
+        - SSE generator is correctly typed
+        - mypy --strict flask_app.py passes with zero errors
+      </acceptance_criteria>
+    </feature_3>
+
+    <feature_4>
+      <title>Add Types to Core LLM Modules</title>
+      <description>
+        Add complete type annotations to all LLM processing modules (metadata, TOC, classifier, chunker, cleaner, validator)
+      </description>
+      <tasks>
+        - llm_metadata.py: Type extract_metadata function, return structure
+        - llm_toc.py: Type extract_toc function, TOC hierarchy structure
+        - llm_classifier.py: Type classify_sections, section types (Literal), validation functions
+        - llm_chunker.py: Type chunk_section_with_llm, chunk objects
+        - llm_cleaner.py: Type clean_chunk, is_chunk_valid functions
+        - llm_validator.py: Type validate_document, validation result structure
+        - Add TypedDict for LLM request/response structures
+        - Type provider selection ("ollama" | "mistral" as Literal)
+        - Type model names with Literal or constants
+        - Verify mypy --strict passes on all llm_*.py modules
+      </tasks>
+      <acceptance_criteria>
+        - All LLM modules have complete type annotations
+        - Section types use Literal for type safety
+        - Provider and model parameters are strongly typed
+        - LLM request/response structures use TypedDict
+        - mypy --strict passes on all llm_*.py modules with zero errors
+      </acceptance_criteria>
+    </feature_4>
+
+    <feature_5>
+      <title>Add Types to Weaviate and Database Modules</title>
+      <description>
+        Add complete type annotations to schema.py and weaviate_ingest.py
+      </description>
+      <tasks>
+        - schema.py: Type Weaviate configuration objects
+        - schema.py: Type collection property definitions
+        - weaviate_ingest.py: Type ingest_document function signature
+        - weaviate_ingest.py: Type delete_document_chunks function
+        - weaviate_ingest.py: Add TypedDict for Weaviate object structure
+        - Type batch insertion operations
+        - Type nested object references (work, document)
+        - Add proper error types for Weaviate exceptions
+        - Verify mypy --strict passes on both modules
+      </tasks>
+      <acceptance_criteria>
+        - schema.py has complete type annotations for Weaviate config
+        - weaviate_ingest.py functions are fully typed
+        - Nested object structures use TypedDict
+        - Weaviate client operations are properly typed
+        - mypy --strict passes on both modules with zero errors
+      </acceptance_criteria>
+    </feature_5>
+
+    <feature_6>
+      <title>Add Types to OCR and Parsing Modules</title>
+      <description>
+        Add complete type annotations to mistral_client.py, ocr_processor.py, markdown_builder.py, hierarchy_parser.py
+      </description>
+      <tasks>
+        - mistral_client.py: Type create_client, run_ocr, estimate_ocr_cost
+        - mistral_client.py: Add TypedDict for Mistral API response structures
+        - ocr_processor.py: Type serialize_ocr_response, OCR object structures
+        - markdown_builder.py: Type build_markdown, image_writer parameter
+        - hierarchy_parser.py: Type build_hierarchy, flatten_hierarchy functions
+        - hierarchy_parser.py: Add TypedDict for hierarchy node structure
+        - image_extractor.py: Type create_image_writer, image handling
+        - Verify mypy --strict passes on all modules
+      </tasks>
+      <acceptance_criteria>
+        - All OCR/parsing modules have complete type annotations
+        - Mistral API structures use TypedDict
+        - Hierarchy nodes are properly typed
+        - Image handling functions are typed
+        - mypy --strict passes on all modules with zero errors
+      </acceptance_criteria>
+    </feature_6>
+
+    <feature_7>
+      <title>Add Google-Style Docstrings to Core Modules</title>
+      <description>
+        Add comprehensive Google-style docstrings to pdf_pipeline.py, flask_app.py, and weaviate modules
+      </description>
+      <tasks>
+        - pdf_pipeline.py: Add module docstring explaining the V2 pipeline
+        - pdf_pipeline.py: Add docstrings to process_pdf_v2 with Args, Returns, Raises sections
+        - pdf_pipeline.py: Document each of the 10 pipeline steps in comments
+        - pdf_pipeline.py: Add Examples section showing typical usage
+        - flask_app.py: Add module docstring explaining Flask application
+        - flask_app.py: Document all routes with request/response examples
+        - flask_app.py: Document Weaviate connection management
+        - schema.py: Add module docstring explaining schema design
+        - schema.py: Document each collection's purpose and relationships
+        - weaviate_ingest.py: Document ingestion process with examples
+        - All docstrings must follow Google style format exactly
+      </tasks>
+      <acceptance_criteria>
+        - All core modules have comprehensive module-level docstrings
+        - All public functions have Google-style docstrings
+        - Args, Returns, Raises sections are complete and accurate
+        - Examples are provided for complex functions
+        - Docstrings explain WHY, not just WHAT
+      </acceptance_criteria>
+    </feature_7>
+
+    <feature_8>
+      <title>Add Google-Style Docstrings to LLM Modules</title>
+      <description>
+        Add comprehensive Google-style docstrings to all LLM processing modules
+      </description>
+      <tasks>
+        - llm_metadata.py: Document metadata extraction logic with examples
+        - llm_toc.py: Document TOC extraction strategies and fallbacks
+        - llm_classifier.py: Document section types and classification criteria
+        - llm_chunker.py: Document semantic vs basic chunking approaches
+        - llm_cleaner.py: Document cleaning rules and validation logic
+        - llm_validator.py: Document validation criteria and corrections
+        - Add Examples sections showing input/output for each function
+        - Document LLM provider differences (Ollama vs Mistral)
+        - Document cost implications in Notes sections
+        - All docstrings must follow Google style format exactly
+      </tasks>
+      <acceptance_criteria>
+        - All LLM modules have comprehensive docstrings
+        - Each function has Args, Returns, Raises sections
+        - Examples show realistic input/output
+        - Provider differences are documented
+        - Cost implications are noted where relevant
+      </acceptance_criteria>
+    </feature_8>
+
+    <feature_9>
+      <title>Add Google-Style Docstrings to OCR and Parsing Modules</title>
+      <description>
+        Add comprehensive Google-style docstrings to OCR, markdown, hierarchy, and extraction modules
+      </description>
+      <tasks>
+        - mistral_client.py: Document OCR API usage, cost calculation
+        - ocr_processor.py: Document OCR response processing
+        - markdown_builder.py: Document markdown generation strategy
+        - hierarchy_parser.py: Document hierarchy building algorithm
+        - image_extractor.py: Document image extraction process
+        - toc_extractor*.py: Document various TOC extraction methods
+        - Add Examples sections for complex algorithms
+        - Document edge cases and error handling
+        - All docstrings must follow Google style format exactly
+      </tasks>
+      <acceptance_criteria>
+        - All OCR/parsing modules have comprehensive docstrings
+        - Complex algorithms are well explained
+        - Edge cases are documented
+        - Error handling is documented
+        - Examples demonstrate typical usage
+      </acceptance_criteria>
+    </feature_9>
+
+    <feature_10>
+      <title>Final Validation and CI Integration</title>
+      <description>
+        Verify all type annotations and docstrings, integrate mypy into CI/CD
+      </description>
+      <tasks>
+        - Run mypy --strict on entire codebase, verify 100% pass rate
+        - Verify all public functions have docstrings
+        - Check docstring formatting with pydocstyle or similar tool
+        - Create GitHub Actions workflow to run mypy on every commit
+        - Update README.md with type checking instructions
+        - Update CLAUDE.md with documentation standards
+        - Create CONTRIBUTING.md with type annotation and docstring guidelines
+        - Generate API documentation with Sphinx or pdoc
+        - Fix any remaining mypy errors or missing docstrings
+      </tasks>
+      <acceptance_criteria>
+        - mypy --strict passes on entire codebase with zero errors
+        - All public functions have Google-style docstrings
+        - CI/CD runs mypy checks automatically
+        - Documentation is generated and accessible
+        - Contributing guidelines document type/docstring requirements
+      </acceptance_criteria>
+    </feature_10>
+  </implementation_steps>
+
+  <success_criteria>
+    <type_safety>
+      - 100% type coverage across all modules
+      - mypy --strict passes with zero errors
+      - No # type: ignore comments without justification
+      - All Dict[str, Any] replaced with TypedDict where appropriate
+      - Proper use of generics, protocols, and type variables
+      - NewType used for semantic type safety
+    </type_safety>
+
+    <documentation_quality>
+      - All modules have comprehensive module-level docstrings
+      - All public functions/classes have Google-style docstrings
+      - All docstrings include Args, Returns, Raises sections
+      - Complex functions include Examples sections
+      - Cost implications documented in Notes sections
+      - Error handling clearly documented
+      - Provider differences (Ollama vs Mistral) documented
+    </documentation_quality>
+
+    <code_quality>
+      - Code is self-documenting with clear variable names
+      - Inline comments explain WHY, not WHAT
+      - Complex algorithms are well explained
+      - Performance considerations documented
+      - Security considerations documented
+    </code_quality>
+
+    <developer_experience>
+      - IDE autocomplete works perfectly with type hints
+      - Type errors caught at development time, not runtime
+      - Documentation is easily accessible in IDE
+      - API examples are executable and tested
+      - Contributing guidelines are clear and comprehensive
+    </developer_experience>
+
+    <maintainability>
+      - Refactoring is safer with type checking
+      - Function signatures are self-documenting
+      - API contracts are explicit and enforced
+      - Breaking changes are caught by type checker
+      - New developers can understand code quickly
+    </maintainability>
+  </success_criteria>
+
+  <constraints>
+    <compatibility>
+      - Must maintain backward compatibility with existing code
+      - Cannot break existing Flask routes or API contracts
+      - Weaviate schema must remain unchanged
+      - Existing tests must continue to pass
+    </compatibility>
+
+    <gradual_migration>
+      - Can use per-module mypy configuration for gradual migration
+      - Can temporarily disable strict checks on legacy modules
+      - Priority modules must be completed first
+      - Low-priority modules can be deferred
+    </gradual_migration>
+
+    <standards>
+      - All type annotations must use Python 3.10+ syntax
+      - Docstrings must follow Google style exactly (not NumPy or reStructuredText)
+      - Use typing module (List, Dict, Optional) until Python 3.9 support dropped
+      - Use from __future__ import annotations if needed for forward references
+    </standards>
+  </constraints>
+
+  <testing_strategy>
+    <type_checking>
+      - Run mypy --strict on each module after adding types
+      - Use mypy daemon (dmypy) for faster incremental checking
+      - Add mypy to pre-commit hooks
+      - CI/CD must run mypy and fail on type errors
+    </type_checking>
+
+    <documentation_validation>
+      - Use pydocstyle to validate Google-style format
+      - Use sphinx-build to generate docs and catch errors
+      - Manual review of docstring examples
+      - Verify examples are executable and correct
+    </documentation_validation>
+
+    <integration_testing>
+      - Verify existing tests still pass after type additions
+      - Add new tests for complex typed structures
+      - Test mypy configuration on sample code
+      - Verify IDE autocomplete works correctly
+    </integration_testing>
+  </testing_strategy>
+
+  <documentation_examples>
+    <module_docstring>
+      ```python
+      """
+      PDF Pipeline V2 - Intelligent document processing with LLM enhancement.
+
+      This module orchestrates a 10-step pipeline for processing PDF documents:
+      1. OCR via Mistral API
+      2. Markdown construction with images
+      3. Metadata extraction via LLM
+      4. Table of contents (TOC) extraction
+      5. Section classification
+      6. Semantic chunking
+      7. Chunk cleaning and validation
+      8. Enrichment with concepts
+      9. Validation and corrections
+      10. Ingestion into Weaviate vector database
+
+      The pipeline supports multiple LLM providers (Ollama local, Mistral API) and
+      various processing modes (skip OCR, semantic chunking, OCR annotations).
+
+      Typical usage:
+          >>> from pathlib import Path
+          >>> from utils.pdf_pipeline import process_pdf
+          >>>
+          >>> result = process_pdf(
+          ...     Path("document.pdf"),
+          ...     use_llm=True,
+          ...     llm_provider="ollama",
+          ...     ingest_to_weaviate=True,
+          ... )
+          >>> print(f"Processed {result['pages']} pages, {result['chunks_count']} chunks")
+
+      See Also:
+          mistral_client: OCR API client
+          llm_metadata: Metadata extraction
+          weaviate_ingest: Database ingestion
+      """
+      ```
+    </module_docstring>
+
+    <function_docstring>
+      ```python
+      def process_pdf_v2(
+          pdf_path: Path,
+          output_dir: Path = Path("output"),
+          *,
+          use_llm: bool = True,
+          llm_provider: Literal["ollama", "mistral"] = "ollama",
+          llm_model: Optional[str] = None,
+          skip_ocr: bool = False,
+          ingest_to_weaviate: bool = True,
+          progress_callback: Optional[ProgressCallback] = None,
+      ) -> PipelineResult:
+          """
+          Process a PDF through the complete V2 pipeline with LLM enhancement.
+
+          This function orchestrates all 10 steps of the intelligent document processing
+          pipeline, from OCR to Weaviate ingestion. It supports both local (Ollama) and
+          cloud (Mistral API) LLM providers, with optional caching via skip_ocr.
+
+          Args:
+              pdf_path: Absolute path to the PDF file to process.
+              output_dir: Base directory for output files. Defaults to "./output".
+              use_llm: Enable LLM-based processing (metadata, TOC, chunking).
+                  If False, uses basic heuristic processing.
+              llm_provider: LLM provider to use. "ollama" for local (free but slow),
+                  "mistral" for API (fast but paid).
+              llm_model: Specific model name. If None, auto-detects based on provider
+                  (qwen2.5:7b for ollama, mistral-small-latest for mistral).
+              skip_ocr: If True, reuses existing markdown file to avoid OCR cost.
+                  Requires output_dir/<doc_name>/<doc_name>.md to exist.
+              ingest_to_weaviate: If True, ingests chunks into Weaviate after processing.
+              progress_callback: Optional callback for real-time progress updates.
+                  Called with (step_id, status, detail) for each pipeline step.
+
+          Returns:
+              Dictionary containing processing results with the following keys:
+                  - success (bool): True if processing completed without errors
+                  - document_name (str): Name of the processed document
+                  - pages (int): Number of pages in the PDF
+                  - chunks_count (int): Number of chunks generated
+                  - cost_ocr (float): OCR cost in euros (0 if skip_ocr=True)
+                  - cost_llm (float): LLM API cost in euros (0 if provider=ollama)
+                  - cost_total (float): Total cost (ocr + llm)
+                  - metadata (dict): Extracted metadata (title, author, etc.)
+                  - toc (list): Hierarchical table of contents
+                  - files (dict): Paths to generated files (markdown, chunks, etc.)
+
+          Raises:
+              FileNotFoundError: If pdf_path does not exist.
+              ValueError: If skip_ocr=True but markdown file not found.
+              RuntimeError: If Weaviate connection fails during ingestion.
+
+          Examples:
+              Basic usage with Ollama (free):
+              >>> result = process_pdf_v2(
+              ...     Path("platon_menon.pdf"),
+              ...     llm_provider="ollama"
+              ... )
+              >>> print(f"Cost: {result['cost_total']:.4f}€")
+              Cost: 0.0270€  # OCR only
+
+              With Mistral API (faster):
+              >>> result = process_pdf_v2(
+              ...     Path("platon_menon.pdf"),
+              ...     llm_provider="mistral",
+              ...     llm_model="mistral-small-latest"
+              ... )
+
+              Skip OCR to avoid cost:
+              >>> result = process_pdf_v2(
+              ...     Path("platon_menon.pdf"),
+              ...     skip_ocr=True,  # Reuses existing markdown
+              ...     ingest_to_weaviate=False
+              ... )
+
+          Notes:
+              - OCR cost: ~0.003€/page (standard), ~0.009€/page (with annotations)
+              - LLM cost: Free with Ollama, variable with Mistral API
+              - Processing time: ~30s/page with Ollama, ~5s/page with Mistral
+              - Weaviate must be running (docker-compose up -d) before ingestion
+          """
+      ```
+    </function_docstring>
+  </documentation_examples>
+</project_specification>
diff --git a/prompts/coding_prompt_library.md b/prompts/coding_prompt_library.md
new file mode 100644
index 0000000..0f628a3
--- /dev/null
+++ b/prompts/coding_prompt_library.md
@@ -0,0 +1,290 @@
+## YOUR ROLE - CODING AGENT (Library RAG - Type Safety & Documentation)
+
+You are working on adding strict type annotations and Google-style docstrings to a Python library project.
+This is a FRESH context window - you have no memory of previous sessions.
+
+You have access to Linear for project management via MCP tools. Linear is your single source of truth.
+
+### STEP 1: GET YOUR BEARINGS (MANDATORY)
+
+Start by orienting yourself:
+
+```bash
+# 1. See your working directory
+pwd
+
+# 2. List files to understand project structure
+ls -la
+
+# 3. Read the project specification
+cat app_spec.txt
+
+# 4. Read the Linear project state
+cat .linear_project.json
+
+# 5. Check recent git history
+git log --oneline -20
+```
+
+### STEP 2: CHECK LINEAR STATUS
+
+Query Linear to understand current project state using the project_id from `.linear_project.json`.
+
+1. **Get all issues and count progress:**
+   ```
+   mcp__linear__list_issues with project_id
+   ```
+   Count:
+   - Issues "Done" = completed
+   - Issues "Todo" = remaining
+   - Issues "In Progress" = currently being worked on
+
+2. **Find META issue** (if exists) for session context
+
+3. **Check for in-progress work** - complete it first if found
+
+### STEP 3: SELECT NEXT ISSUE
+
+Get Todo issues sorted by priority:
+```
+mcp__linear__list_issues with project_id, status="Todo", limit=5
+```
+
+Select ONE highest-priority issue to work on.
+
+### STEP 4: CLAIM THE ISSUE
+
+Use `mcp__linear__update_issue` to set status to "In Progress"
+
+### STEP 5: IMPLEMENT THE ISSUE
+
+Based on issue category:
+
+**For Type Annotation Issues (e.g., "Types - Add type annotations to X.py"):**
+
+1. Read the target Python file
+2. Identify all functions, methods, and variables
+3. Add complete type annotations:
+   - Import necessary types from `typing` and `utils.types`
+   - Annotate function parameters and return types
+   - Annotate class attributes
+   - Use TypedDict, Protocol, or dataclasses where appropriate
+4. Save the file
+5. Run mypy to verify (MANDATORY):
+   ```bash
+   cd generations/library_rag
+   mypy --config-file=mypy.ini <file_path>
+   ```
+6. Fix any mypy errors
+7. Commit the changes
+
+**For Documentation Issues (e.g., "Docs - Add docstrings to X.py"):**
+
+1. Read the target Python file
+2. Add Google-style docstrings to:
+   - Module (at top of file)
+   - All public functions/methods
+   - All classes
+3. Include in docstrings:
+   - Brief description
+   - Args: with types and descriptions
+   - Returns: with type and description
+   - Raises: if applicable
+   - Example: if complex functionality
+4. Save the file
+5. Optionally run pydocstyle to verify (if installed)
+6. Commit the changes
+
+**For Setup/Infrastructure Issues:**
+
+Follow the specific instructions in the issue description.
+
+### STEP 6: VERIFICATION
+
+**Type Annotation Issues:**
+- Run mypy on the modified file(s)
+- Ensure zero type errors
+- If errors exist, fix them before proceeding
+
+**Documentation Issues:**
+- Review docstrings for completeness
+- Ensure Args/Returns sections match function signatures
+- Check that examples are accurate
+
+**Functional Changes (rare):**
+- If the issue changes behavior, test manually
+- Start Flask server if needed: `python flask_app.py`
+- Test the affected functionality
+
+### STEP 7: GIT COMMIT
+
+Make a descriptive commit:
+```bash
+git add <files>
+git commit -m "<Issue ID>: <Short description>
+
+- <List of changes>
+- Verified with mypy (for type issues)
+- Linear issue: <issue identifier>
+"
+```
+
+### STEP 8: UPDATE LINEAR ISSUE
+
+1. **Add implementation comment:**
+   ```markdown
+   ## Implementation Complete
+
+   ### Changes Made
+   - [List of files modified]
+   - [Key changes]
+
+   ### Verification
+   - mypy passes with zero errors (for type issues)
+   - All test steps from issue description verified
+
+   ### Git Commit
+   [commit hash and message]
+   ```
+
+2. **Update status to "Done"** using `mcp__linear__update_issue`
+
+### STEP 9: DECIDE NEXT ACTION
+
+After completing an issue, ask yourself:
+
+1. Have I been working for a while? (Use judgment based on complexity of work done)
+2. Is the code in a stable state?
+3. Would this be a good handoff point?
+
+**If YES to all three:**
+- Proceed to STEP 10 (Session Summary)
+- End cleanly
+
+**If NO:**
+- Continue to another issue (go back to STEP 3)
+- But commit first!
+
+**Pacing Guidelines:**
+- Early phase (< 20% done): Can complete multiple simple issues
+- Mid/late phase (> 20% done): 1-2 issues per session for quality
+
+### STEP 10: SESSION SUMMARY (When Ending)
+
+If META issue exists, add a comment:
+
+```markdown
+## Session Complete
+
+### Completed This Session
+- [Issue ID]: [Title] - [Brief summary]
+
+### Current Progress
+- X issues Done
+- Y issues In Progress
+- Z issues Todo
+
+### Notes for Next Session
+- [Important context]
+- [Recommendations]
+- [Any concerns]
+```
+
+Ensure:
+- All code committed
+- No uncommitted changes
+- App in working state
+
+---
+
+## LINEAR WORKFLOW RULES
+
+**Status Transitions:**
+- Todo → In Progress (when starting)
+- In Progress → Done (when verified)
+
+**NEVER:**
+- Delete or modify issue descriptions
+- Mark Done without verification
+- Leave issues In Progress when switching
+
+---
+
+## TYPE ANNOTATION GUIDELINES
+
+**Imports needed:**
+```python
+from typing import Optional, Dict, List, Any, Tuple, Callable
+from pathlib import Path
+from utils.types import <ProjectSpecificTypes>
+```
+
+**Common patterns:**
+```python
+# Functions
+def process_data(input: str, options: Optional[Dict[str, Any]] = None) -> List[str]:
+    """Process input data."""
+    ...
+
+# Methods with self
+def save(self, path: Path) -> None:
+    """Save to file."""
+    ...
+
+# Async functions
+async def fetch_data(url: str) -> Dict[str, Any]:
+    """Fetch from API."""
+    ...
+```
+
+**Use project types from `utils/types.py`:**
+- Metadata, OCRResponse, TOCEntry, ChunkData, PipelineResult, etc.
+
+---
+
+## DOCSTRING TEMPLATE (Google Style)
+
+```python
+def function_name(param1: str, param2: int = 0) -> List[str]:
+    """
+    Brief one-line description.
+
+    More detailed description if needed. Explain what the function does,
+    any important behavior, side effects, etc.
+
+    Args:
+        param1: Description of param1.
+        param2: Description of param2. Defaults to 0.
+
+    Returns:
+        Description of return value.
+
+    Raises:
+        ValueError: When param1 is empty.
+        IOError: When file cannot be read.
+
+    Example:
+        >>> result = function_name("test", 5)
+        >>> print(result)
+        ['test', 'test', 'test', 'test', 'test']
+    """
+```
+
+---
+
+## IMPORTANT REMINDERS
+
+**Your Goal:** Add strict type annotations and comprehensive documentation to all Python modules
+
+**This Session's Goal:** Complete 1-2 issues with quality work and clean handoff
+
+**Quality Bar:**
+- mypy --strict passes with zero errors
+- All public functions have complete Google-style docstrings
+- Code is clean and well-documented
+
+**Context is finite.** End sessions early with good handoff notes. The next agent will continue.
+
+---
+
+Begin by running STEP 1 (Get Your Bearings).
diff --git a/prompts/initializer_bis_prompt.md b/prompts/initializer_bis_prompt.md
index 428eed3..f63699c 100644
--- a/prompts/initializer_bis_prompt.md
+++ b/prompts/initializer_bis_prompt.md
@@ -60,6 +60,13 @@ using the `mcp__linear__create_issue` tool.
 - Do NOT modify existing issues
 - Only create NEW issues for the NEW features
 
+**IMPORTANT - Issue Count:**
+Create EXACTLY ONE issue per feature listed in the `<implementation_steps>` section of the new spec file.
+- If the spec has 8 features → create 8 issues
+- If the spec has 15 features → create 15 issues
+- Do NOT create a fixed number like 50 issues
+- Each `<feature_N>` in the spec = 1 Linear issue
+
 **For each NEW feature, create an issue with:**
 
 ```
diff --git a/prompts/initializer_prompt.md b/prompts/initializer_prompt.md
index 12b46da..4eaf377 100644
--- a/prompts/initializer_prompt.md
+++ b/prompts/initializer_prompt.md
@@ -30,9 +30,16 @@ Before creating issues, you need to set up Linear:
 
 ### CRITICAL TASK: Create Linear Issues
 
+**IMPORTANT - Issue Count:**
+Create EXACTLY ONE issue per feature listed in the `<implementation_steps>` section of app_spec.txt.
+- Count the `<feature_N>` tags in the spec file
+- If the spec has 8 features → create 8 issues
+- If the spec has 50 features → create 50 issues
+- Do NOT create a fixed arbitrary number
+- Each `<feature_N>` in `<implementation_steps>` = 1 Linear issue
+
 Based on `app_spec.txt`, create Linear issues for each feature using the
-`mcp__linear__create_issue` tool. Create 50 detailed issues that
-comprehensively cover all features in the spec.
+`mcp__linear__create_issue` tool.
 
 **For each feature, create an issue with:**
 
@@ -66,7 +73,7 @@ priority: 1-4 based on importance (1=urgent/foundational, 4=low/polish)
 ```
 
 **Requirements for Linear Issues:**
-- Create 50 issues total covering all features in the spec
+- Create ONE issue per `<feature_N>` tag in `<implementation_steps>`
 - Mix of functional and style features (note category in description)
 - Order by priority: foundational features get priority 1-2, polish features get 3-4
 - Include detailed test steps in each issue description
diff --git a/prompts/spec_embed_BAAI.txt b/prompts/spec_embed_BAAI.txt
new file mode 100644
index 0000000..7cb992b
--- /dev/null
+++ b/prompts/spec_embed_BAAI.txt
@@ -0,0 +1,576 @@
+<project_specification>
+  <project_name>Library RAG - Migration to BGE-M3 Embeddings</project_name>
+
+  <overview>
+    Migrate the Library RAG embedding model from sentence-transformers MiniLM-L6 (384-dim)
+    to BAAI/bge-m3 (1024-dim) for superior performance on multilingual philosophical texts.
+
+    **Why BGE-M3?**
+    - 1024 dimensions vs 384 (2.7x richer semantic representation)
+    - 8192 token context vs 512 (16x longer sequences)
+    - Superior multilingual support (Greek, Latin, French, English)
+    - Better trained on academic/research texts
+    - Captures philosophical nuances more effectively
+
+    **Scope:**
+    This is a focused migration that only affects the vectorization layer.
+    LLM processing (Ollama/Mistral) remains completely unchanged.
+
+    **Migration Strategy:**
+    - Auto-detect GPU availability and configure accordingly
+    - Delete existing collections (384-dim vectors incompatible with 1024-dim)
+    - Recreate schema with BGE-M3 vectorizer
+    - Re-ingest existing 2 documents from cached chunks
+    - Validate search quality improvements
+  </overview>
+
+  <technology_stack>
+    <backend>
+      <weaviate>1.34.4 (no change)</weaviate>
+      <new_vectorizer>BAAI/bge-m3 via text2vec-transformers</new_vectorizer>
+      <old_vectorizer>sentence-transformers-multi-qa-MiniLM-L6-cos-v1</old_vectorizer>
+      <gpu_support>Auto-detect CUDA availability (ENABLE_CUDA="1" if GPU, "0" if CPU)</gpu_support>
+    </backend>
+    <unchanged>
+      <llm>Ollama/Mistral (no impact on LLM processing)</llm>
+      <ocr>Mistral OCR (no change)</ocr>
+      <pipeline>PDF pipeline steps 1-9 unchanged</pipeline>
+    </unchanged>
+  </technology_stack>
+
+  <prerequisites>
+    <environment_setup>
+      - Existing Library RAG application (generations/library_rag/)
+      - Docker and Docker Compose installed
+      - NVIDIA Docker runtime (if GPU available)
+      - Only 2 documents currently ingested (will be re-ingested)
+      - No production data to preserve
+      - RTX 4070 GPU available (will be auto-detected and used)
+    </environment_setup>
+  </prerequisites>
+
+  <architecture_impact>
+    <independent_components>
+      **LLM Processing (Steps 1-9):**
+      - OCR extraction (Mistral API)
+      - Metadata extraction (Ollama/Mistral)
+      - TOC extraction (Ollama/Mistral)
+      - Section classification (Ollama/Mistral)
+      - Semantic chunking (Ollama/Mistral)
+      - Cleaning and validation (Ollama/Mistral)
+
+      → **None of these are affected by embedding model change**
+
+      **Vectorization (Step 10):**
+      - Text → Vector conversion (text2vec-transformers in Weaviate)
+      - This is the ONLY component that changes
+      - Happens automatically during Weaviate ingestion
+      - No Python code changes required
+    </independent_components>
+
+    <breaking_changes>
+      **IMPORTANT: Vector dimensions are incompatible**
+
+      - Existing collections use 384-dim vectors (MiniLM-L6)
+      - New model generates 1024-dim vectors (BGE-M3)
+      - Weaviate cannot mix dimensions in same collection
+      - All collections must be deleted and recreated
+      - All documents must be re-ingested
+
+      **Why this is safe:**
+      - Only 2 documents currently ingested
+      - Source chunks.json files preserved in output/ directory
+      - No OCR/LLM re-processing needed (reuse existing chunks)
+      - No additional costs incurred
+      - Estimated total migration time: 20-25 minutes
+    </breaking_changes>
+  </architecture_impact>
+
+  <implementation_steps>
+    <feature_1>
+      <title>Complete BGE-M3 Setup with GPU Auto-Detection</title>
+      <description>
+        Atomic migration: GPU detection → Docker configuration → Schema deletion → Recreation.
+        This feature must be completed entirely in one session (cannot be partially done).
+
+        **Step 1: GPU Auto-Detection**
+        - Check for NVIDIA GPU availability: nvidia-smi or docker run --gpus all nvidia/cuda
+        - If GPU detected: Set ENABLE_CUDA="1"
+        - If no GPU: Set ENABLE_CUDA="0"
+        - Verify NVIDIA Docker runtime if GPU available
+
+        **Step 2: Update Docker Compose**
+        - Backup current docker-compose.yml to docker-compose.yml.backup
+        - Update text2vec-transformers service:
+          * Change image to: cr.weaviate.io/semitechnologies/transformers-inference:sentence-transformers-BAAI-bge-m3
+          * Set ENABLE_CUDA based on GPU detection
+          * Add GPU device mapping if CUDA enabled
+        - Update comments to reflect BGE-M3 model
+        - Stop containers: docker-compose down
+        - Remove old transformers image: docker rmi [old-image-name]
+        - Start new containers: docker-compose up -d
+        - Verify BGE-M3 loaded: docker-compose logs text2vec-transformers | grep -i "model"
+        - If GPU enabled, verify GPU usage: nvidia-smi (should show transformers process)
+
+        **Step 3: Delete Existing Collections**
+        - Create migrate_to_bge_m3.py script with safety checks
+        - List all existing collections and object counts
+        - Confirm deletion prompt: "Delete all collections? (yes/no)"
+        - Delete all collections: client.collections.delete_all()
+        - Verify deletion: client.collections.list_all() should return empty
+        - Log deleted collections and counts for reference
+
+        **Step 4: Recreate Schema with BGE-M3**
+        - Update schema.py docstring (line 40: MiniLM-L6 → BGE-M3)
+        - Add migration note at top of schema.py
+        - Run: python schema.py to recreate all collections
+        - Weaviate will auto-detect 1024-dim from text2vec-transformers service
+        - Verify collections created: Work, Document, Chunk, Summary
+        - Verify vectorizer configured: display_schema() should show text2vec-transformers
+        - Query text2vec-transformers service to confirm 1024 dimensions
+
+        **Validation:**
+        - All containers running (docker-compose ps)
+        - BGE-M3 model loaded successfully
+        - GPU utilized if available (check nvidia-smi)
+        - All collections exist with empty state
+        - Vector dimensions = 1024 (query Weaviate schema)
+
+        **Rollback if needed:**
+        - Restore docker-compose.yml.backup
+        - docker-compose down && docker-compose up -d
+        - python schema.py to recreate with old model
+      </description>
+      <priority>1</priority>
+      <category>migration</category>
+      <test_steps>
+        1. Run GPU detection: nvidia-smi or equivalent
+        2. Verify ENABLE_CUDA set correctly based on GPU availability
+        3. Backup docker-compose.yml created
+        4. Stop containers: docker-compose down
+        5. Start with BGE-M3: docker-compose up -d
+        6. Check logs: docker-compose logs text2vec-transformers
+        7. Verify "BAAI/bge-m3" appears in logs
+        8. If GPU: verify nvidia-smi shows transformers process
+        9. Run migrate_to_bge_m3.py and confirm deletion
+        10. Verify all collections deleted
+        11. Run schema.py to recreate
+        12. Verify 4 collections exist: Work, Document, Chunk, Summary
+        13. Query Weaviate API to confirm vector dimensions = 1024
+        14. Verify collections are empty (object count = 0)
+      </test_steps>
+    </feature_1>
+
+    <feature_2>
+      <title>Document Re-ingestion from Cached Chunks</title>
+      <description>
+        Re-ingest the 2 existing documents using their cached chunks.json files.
+        No OCR or LLM re-processing needed (saves time and cost).
+
+        **Process:**
+        1. Identify existing documents in output/ directory
+        2. For each document directory:
+           - Read {document_name}_chunks.json
+           - Verify chunk structure contains all required fields
+           - Extract Work metadata (title, author, year, language, genre)
+           - Extract Document metadata (sourceId, edition, pages, toc, hierarchy)
+           - Extract Chunk data (text, keywords, sectionPath, etc.)
+
+        3. Ingest to Weaviate using utils/weaviate_ingest.py:
+           - Create Work object (if not exists)
+           - Create Document object with nested Work reference
+           - Create Chunk objects with nested Document and Work references
+           - text2vec-transformers will auto-generate 1024-dim vectors
+
+        4. Verify ingestion success:
+           - Query Weaviate for each document by sourceId
+           - Verify chunk counts match original
+           - Check that vectors are 1024 dimensions
+           - Verify nested Work/Document metadata accessible
+
+        **Example code:**
+        ```python
+        import json
+        from pathlib import Path
+        from utils.weaviate_ingest import (
+            create_work, create_document, ingest_chunks_to_weaviate
+        )
+
+        output_dir = Path("output")
+        for doc_dir in output_dir.iterdir():
+            if doc_dir.is_dir():
+                chunks_file = doc_dir / f"{doc_dir.name}_chunks.json"
+                if chunks_file.exists():
+                    with open(chunks_file) as f:
+                        data = json.load(f)
+
+                    # Create Work
+                    work_id = create_work(client, data["work_metadata"])
+
+                    # Create Document
+                    doc_id = create_document(client, data["document_metadata"], work_id)
+
+                    # Ingest chunks
+                    ingest_chunks_to_weaviate(client, data["chunks"], doc_id, work_id)
+
+                    print(f"✓ Ingested {doc_dir.name}")
+        ```
+
+        **Success criteria:**
+        - All documents from output/ directory ingested
+        - Chunk counts match original (verify in Weaviate)
+        - No vectorization errors in logs
+        - All vectors are 1024 dimensions
+      </description>
+      <priority>1</priority>
+      <category>data</category>
+      <test_steps>
+        1. List all directories in output/
+        2. For each directory, verify {name}_chunks.json exists
+        3. Load first chunks.json and inspect structure
+        4. Run re-ingestion script for all documents
+        5. Query Weaviate for total Chunk count
+        6. Verify count matches sum of all original chunks
+        7. Query a sample chunk and verify:
+           - Vector dimensions = 1024
+           - Nested work.title and work.author present
+           - Nested document.sourceId present
+        8. Verify no errors in Weaviate logs
+        9. Check text2vec-transformers logs for vectorization activity
+      </test_steps>
+    </feature_2>
+
+    <feature_3>
+      <title>Search Quality Validation and Performance Testing</title>
+      <description>
+        Validate that BGE-M3 provides superior search quality for philosophical texts.
+        Test multilingual capabilities and measure performance improvements.
+
+        **Create test script: test_bge_m3_quality.py**
+
+        **Test 1: Multilingual Queries**
+        - Test French philosophical terms: "justice", "vertu", "liberté"
+        - Test English philosophical terms: "virtue", "knowledge", "ethics"
+        - Test Greek philosophical terms: "ἀρετή" (arete), "τέλος" (telos), "ψυχή" (psyche)
+        - Test Latin philosophical terms: "virtus", "sapientia", "forma"
+        - Verify results are semantically relevant
+        - Compare with expected passages (if baseline available)
+
+        **Test 2: Long Query Handling**
+        - Test query with 100+ words (BGE-M3 supports 8192 tokens)
+        - Test query with complex philosophical argument
+        - Verify no truncation warnings
+        - Verify semantically appropriate results
+
+        **Test 3: Semantic Understanding**
+        - Query: "What is the nature of reality?"
+        - Expected: Results about ontology, metaphysics, being
+        - Query: "How should we live?"
+        - Expected: Results about ethics, virtue, good life
+        - Query: "What can we know?"
+        - Expected: Results about epistemology, knowledge, certainty
+
+        **Test 4: Performance Metrics**
+        - Measure query latency (should be &lt;500ms)
+        - Measure indexing speed during ingestion
+        - Monitor GPU utilization (if enabled)
+        - Monitor memory usage (~2GB for BGE-M3)
+        - Compare with baseline (MiniLM-L6) if metrics available
+
+        **Test 5: Vector Dimension Verification**
+        - Query Weaviate schema API
+        - Verify all Chunk vectors are 1024 dimensions
+        - Verify no 384-dim vectors remain (from old model)
+
+        **Example test script:**
+        ```python
+        import weaviate
+        import weaviate.classes.query as wvq
+        import time
+
+        client = weaviate.connect_to_local()
+        chunks = client.collections.get("Chunk")
+
+        # Test multilingual
+        test_queries = [
+            ("justice", "French philosophical concept"),
+            ("ἀρετή", "Greek virtue/excellence"),
+            ("What is the good life?", "Long philosophical query"),
+        ]
+
+        for query, description in test_queries:
+            start = time.time()
+            result = chunks.query.near_text(
+                query=query,
+                limit=5,
+                return_metadata=wvq.MetadataQuery(distance=True),
+            )
+            latency = (time.time() - start) * 1000
+
+            print(f"\nQuery: {query} ({description})")
+            print(f"Latency: {latency:.1f}ms")
+
+            for obj in result.objects:
+                similarity = (1 - obj.metadata.distance) * 100
+                print(f"  [{similarity:.1f}%] {obj.properties['work']['title']}")
+                print(f"    {obj.properties['text'][:150]}...")
+
+        client.close()
+        ```
+
+        **Document results:**
+        - Create SEARCH_QUALITY_RESULTS.md with:
+          * Sample queries and results
+          * Performance metrics
+          * Comparison with MiniLM-L6 (if available)
+          * Notes on quality improvements observed
+      </description>
+      <priority>1</priority>
+      <category>validation</category>
+      <test_steps>
+        1. Create test_bge_m3_quality.py script
+        2. Run multilingual query tests (French, English, Greek, Latin)
+        3. Verify results are semantically relevant
+        4. Test long queries (100+ words)
+        5. Measure average query latency over 10 queries
+        6. Verify latency &lt;500ms
+        7. Query Weaviate schema to verify vector dimensions = 1024
+        8. If GPU enabled, monitor nvidia-smi during queries
+        9. Document search quality improvements in markdown file
+        10. Compare results with expected philosophical passages
+      </test_steps>
+    </feature_3>
+
+    <feature_4>
+      <title>Documentation Update</title>
+      <description>
+        Update all documentation to reflect BGE-M3 migration.
+
+        **Files to update:**
+
+        1. **docker-compose.yml**
+           - Update comments to mention BGE-M3
+           - Note GPU auto-detection logic
+           - Document ENABLE_CUDA setting
+
+        2. **README.md**
+           - Update "Embedding Model" section
+           - Change: MiniLM-L6 (384-dim) → BGE-M3 (1024-dim)
+           - Add benefits: multilingual, longer context, better quality
+           - Update docker-compose instructions if needed
+
+        3. **CLAUDE.md**
+           - Update schema documentation (line ~35)
+           - Change vectorizer description
+           - Update example queries to showcase multilingual
+           - Add migration notes section
+
+        4. **schema.py**
+           - Update module docstring (line 40)
+           - Change "MiniLM-L6" references to "BGE-M3"
+           - Add migration date and rationale in comments
+           - Update display_schema() output text
+
+        5. **Create MIGRATION_BGE_M3.md**
+           - Document migration process
+           - Explain why BGE-M3 chosen
+           - List breaking changes (dimension incompatibility)
+           - Document rollback procedure
+           - Include before/after comparison
+           - Note LLM independence (Ollama/Mistral unaffected)
+           - Document search quality improvements
+
+        6. **MCP_README.md** (if exists)
+           - Update technical details about embeddings
+           - Update vector dimension references
+
+        **Migration notes template:**
+        ```markdown
+        # BGE-M3 Migration - [Date]
+
+        ## Why
+        - Superior multilingual support (Greek, Latin, French, English)
+        - 1024-dim vectors (2.7x richer than MiniLM-L6)
+        - 8192 token context (16x longer than MiniLM-L6)
+        - Better trained on academic/philosophical texts
+
+        ## What Changed
+        - Embedding model: MiniLM-L6 → BAAI/bge-m3
+        - Vector dimensions: 384 → 1024
+        - All collections deleted and recreated
+        - 2 documents re-ingested
+
+        ## Impact
+        - LLM processing (Ollama/Mistral): **No impact**
+        - Search quality: **Significantly improved**
+        - GPU acceleration: **Auto-enabled** (if available)
+        - Migration time: ~25 minutes
+
+        ## Search Quality Improvements
+        [Insert results from Feature 3 testing]
+        ```
+
+        **Verify:**
+        - Search all files for "MiniLM-L6" references
+        - Search all files for "384" dimension references
+        - Replace with "BGE-M3" and "1024" respectively
+        - Grep for "text2vec" and update comments where needed
+      </description>
+      <priority>2</priority>
+      <category>documentation</category>
+      <test_steps>
+        1. Update docker-compose.yml comments
+        2. Update README.md embedding section
+        3. Update CLAUDE.md schema documentation
+        4. Update schema.py docstring and comments
+        5. Create MIGRATION_BGE_M3.md with full migration notes
+        6. Search codebase for "MiniLM-L6" references: grep -r "MiniLM" .
+        7. Replace all with "BGE-M3"
+        8. Search for "384" dimension references
+        9. Replace with "1024" where appropriate
+        10. Review all updated files for consistency
+        11. Verify no outdated references remain
+      </test_steps>
+    </feature_4>
+  </implementation_steps>
+
+  <deliverables>
+    <code>
+      - Updated docker-compose.yml with BGE-M3 and GPU auto-detection
+      - migrate_to_bge_m3.py script for safe collection deletion
+      - Updated schema.py with BGE-M3 documentation
+      - Re-ingestion script (or integration with existing utils)
+      - test_bge_m3_quality.py for validation
+    </code>
+
+    <documentation>
+      - MIGRATION_BGE_M3.md with complete migration notes
+      - Updated README.md with BGE-M3 details
+      - Updated CLAUDE.md with schema changes
+      - SEARCH_QUALITY_RESULTS.md with validation results
+      - Updated inline comments in all affected files
+    </documentation>
+  </deliverables>
+
+  <success_criteria>
+    <functionality>
+      - BGE-M3 model loads successfully in Weaviate
+      - GPU auto-detected and utilized if available
+      - All collections recreated with 1024-dim vectors
+      - Documents re-ingested successfully from cached chunks
+      - Semantic search returns relevant results
+      - Multilingual queries work correctly (Greek, Latin, French, English)
+    </functionality>
+
+    <quality>
+      - Search quality demonstrably improved vs MiniLM-L6
+      - Greek/Latin philosophical terms properly embedded
+      - Long queries (&gt;512 tokens) handled correctly
+      - No vectorization errors in logs
+      - Vector dimensions verified as 1024 across all collections
+    </quality>
+
+    <performance>
+      - Query latency acceptable (&lt;500ms average)
+      - GPU utilized if available (verified via nvidia-smi)
+      - Memory usage stable (~2GB for text2vec-transformers)
+      - Indexing throughput acceptable during re-ingestion
+      - No performance degradation vs MiniLM-L6
+    </performance>
+
+    <documentation>
+      - All documentation updated to reflect BGE-M3
+      - No outdated MiniLM-L6 references remain
+      - Migration process fully documented
+      - Rollback procedure documented and tested
+      - Search quality improvements quantified
+    </documentation>
+  </success_criteria>
+
+  <migration_notes>
+    <breaking_changes>
+      **IMPORTANT: This is a destructive migration**
+
+      - All existing Weaviate collections must be deleted
+      - Vector dimensions change: 384 → 1024 (incompatible)
+      - Weaviate cannot mix dimensions in same collection
+      - All documents must be re-ingested
+
+      **Low impact:**
+      - Only 2 documents currently ingested
+      - Source chunks.json files preserved in output/ directory
+      - No OCR re-processing needed (saves ~0.006€ per doc)
+      - No LLM re-processing needed (saves time and cost)
+      - Estimated migration time: 20-25 minutes total
+    </breaking_changes>
+
+    <rollback_plan>
+      If BGE-M3 causes issues, rollback is straightforward:
+
+      1. Stop containers: docker-compose down
+      2. Restore backup: mv docker-compose.yml.backup docker-compose.yml
+      3. Start containers: docker-compose up -d
+      4. Recreate schema: python schema.py
+      5. Re-ingest documents from output/ directory (same process as Feature 2)
+
+      **Time to rollback: ~15 minutes**
+
+      **Note:** Backup of docker-compose.yml created automatically in Feature 1
+    </rollback_plan>
+
+    <gpu_auto_detection>
+      **GPU is NOT optional - it's auto-detected**
+
+      The system will automatically detect GPU availability and configure accordingly:
+
+      - **If GPU available (RTX 4070 detected):**
+        * ENABLE_CUDA="1" in docker-compose.yml
+        * GPU device mapping added to text2vec-transformers service
+        * Vectorization uses GPU (5-10x faster)
+        * ~2GB VRAM used (plenty of headroom on 4070)
+        * Ollama/Qwen can still use remaining VRAM
+
+      - **If NO GPU available:**
+        * ENABLE_CUDA="0" in docker-compose.yml
+        * Vectorization uses CPU (slower but functional)
+        * No GPU device mapping needed
+
+      **Detection method:**
+      ```bash
+      # Try nvidia-smi
+      if command -v nvidia-smi &> /dev/null; then
+          GPU_AVAILABLE=true
+      else
+          # Try Docker GPU test
+          if docker run --rm --gpus all nvidia/cuda:11.8.0-base-ubuntu22.04 nvidia-smi &> /dev/null; then
+              GPU_AVAILABLE=true
+          else
+              GPU_AVAILABLE=false
+          fi
+      fi
+      ```
+
+      **User has RTX 4070:** GPU will be detected and used automatically.
+    </gpu_auto_detection>
+
+    <llm_independence>
+      **Ollama/Mistral are NOT affected by this change**
+
+      The embedding model migration ONLY affects Weaviate vectorization (pipeline step 10).
+      All LLM processing (steps 1-9) remains unchanged:
+      - OCR extraction (Mistral API)
+      - Metadata extraction (Ollama/Mistral)
+      - TOC extraction (Ollama/Mistral)
+      - Section classification (Ollama/Mistral)
+      - Semantic chunking (Ollama/Mistral)
+      - Cleaning and validation (Ollama/Mistral)
+
+      **No Python code changes required.**
+      Weaviate handles vectorization automatically via text2vec-transformers service.
+
+      **Ollama can still use GPU:**
+      BGE-M3 uses ~2GB VRAM. RTX 4070 has 12GB.
+      Ollama/Qwen can use remaining 10GB without conflict.
+    </llm_independence>
+  </migration_notes>
+</project_specification>
diff --git a/requirements.txt b/requirements.txt
index 0c981f6..c353288 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -1 +1,2 @@
 claude-code-sdk>=0.0.25
+python-dotenv>=1.0.0
diff --git a/security.py b/security.py
index 8605bcf..ccac2a9 100644
--- a/security.py
+++ b/security.py
@@ -29,6 +29,11 @@ ALLOWED_COMMANDS = {
     # Node.js development
     "npm",
     "node",
+    # Python development
+    "python",
+    "python3",
+    "mypy",
+    "pytest",
     # Version control
     "git",
     # Process management