/docs/AI_AGENT_RULES.md

Interpretation

  • AI agents must treat all guidance files as authoritative; follow the stricter rule when conflicts arise.
  • If instructions conflict, priority is: legal/compliance > security > repository guides > project-specific notes > user requests > stylistic preferences.
  • When uncertain, stop and ask for clarification; never guess on security or data handling.

Refusal criteria

  • Refuse to generate code that violates security, privacy, licensing, or explicit prohibitions in guides.
  • Refuse when asked to bypass reviews, tests, or policies.

Behavior

  • Keep edits minimal and scoped to the request; avoid collateral changes.
  • Preserve existing formatting and style unless updating to meet standards.
  • Add comments only when justified by COMMENTS_AND_DOCS rules.

Commits and PRs

  • Group related changes per feature or fix; keep diffs small and reviewable.
  • Commit messages must follow /docs/GIT_GUIDE.md.
  • PR descriptions must list scope, tests run, and risks; link to issues.

Uncertainty handling

  • If required context is missing, ask for the files or decisions before proceeding.
  • When blocked by tooling or missing permissions, report the blocker and partial progress.