Expert assistant for the Docling document conversion library. Helps with development, testing, and Docker workflows while enforcing critical project rules.
Expert assistant for working with the Docling document conversion library. This skill enforces critical project rules and provides guidance on development workflows, testing, and Docker usage.
When working with the Docling project, you MUST follow these rules:
1. **NEVER run Python directly** - Python is NOT installed locally. Use Docker for ALL Python tasks.
2. **NEVER modify files in `/home/corey/projects/docling/docling-github/`** - This is protected source from GitHub.
3. **Working directory**: Use `/home/corey/projects/docling/scout-docs/` for all new files and modifications.
These rules were strengthened after being violated 7-10 times. Breaking them demonstrates inability to follow explicit instructions.
1. **Check if Docker is running**: Use `docker compose ps` to verify containers are up
2. **If containers are not running**: Execute `docker compose up -d` to start them
3. **Run Python commands in Docker**: Use `docker compose exec backend python -c "..."` or `docker compose exec backend bash` for interactive sessions
4. **Never attempt**: Running `python`, `pytest`, or any Python commands directly in the host shell
1. **Verify the file path**: Check if the file is in `/home/corey/projects/docling/docling-github/`
2. **If in protected directory**: Inform the user that modifications cannot be made there, and ask if they want to create/modify in `/home/corey/projects/docling/scout-docs/` instead
3. **If in scout-docs**: Proceed with modifications
4. **After code changes**: Suggest running code quality checks using Docker:
- `docker compose exec backend ruff format`
- `docker compose exec backend ruff check --fix`
- `docker compose exec backend mypy docling`
1. **Use uv for package management**: Execute `uv add <package>` to update pyproject.toml and uv.lock
2. **Rebuild Docker containers**: After dependency changes, run `docker compose build` to ensure containers have the new dependencies
3. **Restart if needed**: Use `docker compose restart` to apply changes
1. **Ensure Docker is running**: Verify with `docker compose ps`
2. **Run pytest in Docker**: Use `docker compose exec backend pytest`
3. **For test data regeneration**: Use `docker compose exec backend bash -c "DOCLING_GEN_TEST_DATA=1 pytest"`
4. **Note about reference data**: All PRs modifying reference test data require double review
1. **Explain the conversion flow**:
- Input Processing → Backend Selection → Conversion → Pipeline Processing → Output Generation
2. **Key entry points**:
- `DocumentConverter` in `docling/document_converter.py`
- Format-specific backends in `docling/backend/`
3. **CLI usage** (inside Docker):
- Basic: `docling <file_or_url>`
- VLM pipeline: `docling --pipeline vlm --vlm-model smoldocling <file_or_url>`
- Note: vLLM automatically used on GPU for maximum performance
1. **For all services**: Use `docker compose logs -f`
2. **For specific service**: Use `docker compose logs -f backend`
3. **Explain what to look for**: Conversion errors, pipeline issues, backend selection
1. **Local preview**: Use `mkdocs serve` (runs on localhost:8000)
2. **Deployment**: Use `mkdocs gh-deploy` for GitHub Pages
3. **Documentation location**: Typically in `docs/` directory
4. **Tech stack**: MkDocs with Material theme
```bash
docker compose up -d
docker compose exec backend python -c "print('Hello')"
docker compose exec backend bash
docker compose logs -f backend
docker compose exec backend pytest
docker compose exec backend ruff format
docker compose exec backend ruff check --fix
docker compose exec backend mypy docling
docker compose down
```
1. Always check if operations would violate the protected directory rule
2. Always use Docker for Python execution - no exceptions
3. Always verify Docker containers are running before attempting Python operations
4. Always suggest code quality checks after modifications
5. Always remind about double review requirement for test data changes
Leave a review
No reviews yet. Be the first to review this skill!
# Download SKILL.md from killerskills.ai/api/skills/docling-project-assistant/raw