Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
28 lines (20 loc) · 1.84 KB

CONTRIBUTING.md

File metadata and controls

28 lines (20 loc) · 1.84 KB
Raw

Contributing

This is a living research system, published as it runs. Contributions are welcome — with the understanding that you are operating on an organism, not a library.

Ground rules

  1. The SOUL is immutable. PRs that weaken SOUL.md, the safety gates (SafetyManager, AsimovSafetyFilter, BarratSafetyProtocol, injection guards), or the leak filter will be closed. Hardening them is always welcome.
  2. No fake victories. The project's core law is grounding: a feature "works" when there is a verifiable artifact (test, log, reproducible behavior) — not a claim. PR descriptions should show the artifact.
  3. One file is the design. core.py stays a monolith on purpose (self-patching + AST watchdog). Don't split it into packages; improve it in place. Dashboard code lives in dashboard/ and follows normal Next.js conventions.
  4. Respect the lineage. Ideas borrowed from others are credited inline (see KarpathyAutoResearch, BabyAGI notes). Keep that habit.
  5. Russian prompts are not bugs. The agent's inner voice is Russian by birth. i18n of prompts is a welcome contribution — as configuration, not as deletion.

Good first territories

  • Prompt i18n (extract inline prompts into a swappable locale layer)
  • More providers in UniversalLLMRouter (the slot model makes this mechanical)
  • Grounding judge hardening (execution-based verification of forged skills)
  • Dashboard: new organs for the world, better mobile
  • Tests: TestSuite tasks that verify loop behavior

Dev loop

python -c "import ast; ast.parse(open('core.py').read())"   # the minimum bar — the watchdog's bar
./deploy.sh update                                           # run it; watch docker logs -f nagual

Open an issue before large changes — the architecture has laws (see ARCHITECTURE.md) that are not obvious from the code alone.