Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
48 lines (37 loc) · 2.04 KB

CONTRIBUTING.md

File metadata and controls

48 lines (37 loc) · 2.04 KB
Raw

Contributing to command-code-mods

Thanks for helping make this suite better. It's small by design, so the bar is simple and mechanical.

The changelog is part of the change

Every user-visible or behavior-affecting change updates CHANGELOG.md in the same commit — no separate "docs" commits. Entries are organized per mod under the current Unreleased heading. Version headers bump only at milestone releases.

Commit style

  • Conventional prefixes: fix(mod): …, perf(mod): …, feat(mod): …, chore: ….
  • One cohesive change per commit — each mod's changes land in their own diffable commit.

Architecture rules

  • Each mod is standalone-installable. Duplicated regex tables and helpers across mods are intentional — the cross-mod contract is the event schema (see README), never shared code.
  • Namespaced events only. Emit and consume through cmd.events with namespaced names (self-repair/verdict, memory-bank/graduate).
  • Cache discipline. The system prompt must stay byte-identical across turns: appendSystemPrompt content is static or one-shot; dynamic content rides transformContext (tail append), onStop reasons, or tool additionalContext. Injected messages use array content blocks — content: [{type: 'text', text}] — never raw strings.
  • Ownership markers. A mod that writes to shared locations must stamp an owner marker and refuse to clobber artifacts it doesn't own.

Verification before opening a PR

  1. node scripts/check-contracts.mjs — must pass (CI also runs this).
  2. The changed mods load cleanly (commandcode mods list shows zero load warnings).
  3. A headless smoke run completes: commandcode -p "Reply with exactly SMOKE_OK and nothing else".

What makes a good mod change

  • Fixes the root cause, not the symptom — and documents the root cause.
  • Guards the guardrails: budget caps, min/max flag validation, and refuse-and-suggest on conflicts.
  • Receipts for autonomous actions: what ran, why it belonged, how it was verified, how to undo it.