Blueprint

Blueprint is a request-driven checklist generation engine that turns a high-level prompt plus an optional domain overlay into:

• a compiled run spec
• a living taxonomy
• validated checklist rules
• precedent memory
• resumable runtime state

The active engine now lives in one unversioned package under src/.

Layout

src/
  agents/
  core/
  court/
  domains/
  intelligence/
  main.py
  orchestrator.py
scripts/
  evaluate_output.py
  run_8h.sh
  run_supervised.py
  spawn_supervisor.py
tests/
main.py

Entrypoints

Run the engine from the repository root:

python3 main.py

Or directly as a module:

python3 -m src.main

Useful flags:

python3 -m src.main \
  --request "Frontend development checklist for highest quality possible" \
  --domain src/domains/frontend_engineering.yaml \
  --output outputs/frontend_run \
  --max-loops 0

Runtime Dependencies

Blueprint expects:

• a DeepSeek-compatible chat completion endpoint via DEEPSEEK_API_KEY
• a local Ollama embedding service at http://localhost:11434/api/embed

Core Python dependencies are listed in requirements.txt.

Outputs

A run writes its artifacts to the chosen output directory, including:

• run_spec.json
• run_metadata.json
• rules.json
• hierarchy.json
• runtime_hierarchy.json
• runtime_state.json
• caselaw.json
• checklist.md
• run.log
• supervisor.log

You can evaluate a finished run with:

python3 scripts/evaluate_output.py outputs/frontend_run

Tests

Run the test suite from the repository root:

venv/bin/pytest -q

Notes

• The repository exposes one canonical runtime under src/.
• Generated run outputs should live under outputs/, not under src/.