Skip to content

feat(review): read ## Review section from CLAUDE.md as repo-specific context#452

Open
evb87-tech wants to merge 1 commit intogarrytan:mainfrom
evb87-tech:feat/review-context
Open

feat(review): read ## Review section from CLAUDE.md as repo-specific context#452
evb87-tech wants to merge 1 commit intogarrytan:mainfrom
evb87-tech:feat/review-context

Conversation

@evb87-tech
Copy link

feat(review): read ## Review section from CLAUDE.md as repo-specific context

/review already reads several repo-local inputs: the mandatory checklist, the optional design checklist, DESIGN.md for design calibration, and CLAUDE.md for testing config.

What it doesn't have is a clean place for the other half: scope calibration (where intent lives, ticketing conventions), risk calibration (high-risk paths, trust boundaries), escalation rules (who to involve for auth/billing changes), and auto-fix boundaries (what should never be AUTO-FIXed without explicit approval).

Today that context usually ends up crammed into checklist.md (wrong shape), maintained in a fork of review/SKILL.md (drifts), or repeated manually in every review prompt (forgotten).

This PR adds a single optional step before scope drift detection: if CLAUDE.md has a ## Review section, read it as additive context. If not, skip silently.

What changes

Step 1.25 is added to /review, between branch check and scope drift detection:

  1. Check CLAUDE.md for a ## Review section.
  2. If present, read it as additive repo-specific calibration.
  3. If absent, skip silently.

The mandatory .claude/skills/review/checklist.md remains the source of truth for the rubric, severity, output format, and fix-vs-ask behavior. The ## Review section is for everything else.

Why CLAUDE.md

  • /review already reads CLAUDE.md for testing guidance — this extends an existing pattern, not a new one.
  • CLAUDE.md is already project-owned instruction space — teams version it with their code.
  • Host-agnostic — works the same from Claude Code or .agents hosts.
  • No new files, no new config surface, no new commands.
  • The same convention naturally extends to other skills (## Ship, ## QA) without new machinery.

Files changed

Area Files Existing code modified
Review skill review/SKILL.md.tmpl, review/SKILL.md +10 lines prompt text
Docs docs/skills.md +8 lines
Example examples/CLAUDE.md.review.example New

…c context

/review now checks CLAUDE.md for a `## Review` section before scope drift
detection. This gives teams a clean place for repo-specific calibration —
high-risk paths, auto-fix boundaries, escalation rules, external consumers —
without forking the skill or cramming non-checklist content into checklist.md.

Additive only. Skips silently if the section doesn't exist. The mandatory
checklist.md remains the source of truth for the rubric and output format.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant