Why
Multiple maintainers using multiple AI tools means inconsistent PR shapes, commit styles, comment density, and attribution if left undefined. Worth landing a short RFC/ADR that defines, tool-agnostically, what AI-assisted contributions should look like — for any contributor using any AI tool, not Claude- or Copilot-specific.
Framed as a discussion/RFC, not a deliverable. The conventions are opinionated and need maintainer agreement before they land as enforced rules.
Scope to cover
- Attribution: when and how to credit AI involvement.
Co-Authored-By lines are already used for Claude-driven commits — should Copilot and others follow the same shape? Should the AI's role (drafted / reviewed / executed verbatim) be distinguishable, or is "any AI involvement = one Co-Authored-By line" enough?
- PR description style: how much "explain my reasoning" detail an AI-assisted PR should include. Real tension between useful reviewer context and a wall of AI-generated text that obscures the actual change.
- Comments in code: AI tools love adding comments. CLAUDE.md already says "default to writing no comments". Promote to a repo-wide rule for humans too, or keep AI-specific?
- Review expectations: do AI-authored PRs go through the same review path as human ones, a faster one (rubber-stamp if CI green?), or a stricter one (assume hallucinations until proven otherwise)?
- What's NOT acceptable: speculative refactors, doc-only PRs that add no value, hallucinated API references, changelog entries not grounded in the diff, etc.
- Commit hygiene: AI tools produce many small commits. Squash-on-merge default, or guidance on commit granularity?
Action
Coordinate with
- AGENTS.md adoption issue — outcome lives there
- Data-handling policy issue — adjacent concern
Done when
- ADR drafted and merged
CONTRIBUTING.md / AGENTS.md reference the conventions
- (Optional) machine-enforced bits wired into hooks or PR templates
Why
Multiple maintainers using multiple AI tools means inconsistent PR shapes, commit styles, comment density, and attribution if left undefined. Worth landing a short RFC/ADR that defines, tool-agnostically, what AI-assisted contributions should look like — for any contributor using any AI tool, not Claude- or Copilot-specific.
Framed as a discussion/RFC, not a deliverable. The conventions are opinionated and need maintainer agreement before they land as enforced rules.
Scope to cover
Co-Authored-Bylines are already used for Claude-driven commits — should Copilot and others follow the same shape? Should the AI's role (drafted / reviewed / executed verbatim) be distinguishable, or is "any AI involvement = one Co-Authored-By line" enough?Action
docs/adr/CONTRIBUTING.mdand/orAGENTS.mdCoordinate with
Done when
CONTRIBUTING.md/AGENTS.mdreference the conventions