Add --pr flag to /review-plan for PR feedback coverage checking

[igerber]

Summary

• Extend /review-plan skill with --pr <comment-url> flag that fetches a specific PR review comment, extracts discrete feedback items, and verifies the plan covers each one
• Add new Dimension 9 (PR Feedback Coverage) with addressed/partially addressed/not addressed/dismissed tracking
• Add branch verification against PR head branch with mismatch warnings
• Support all three GitHub comment URL types: issue comments, inline review comments, and PR review bodies (with inline comments)
• Extend output template with PR Context header, Feedback Coverage section, and summary table row
• Add PR Feedback Coverage Delta subsection for --updated + --pr combo

Methodology references (required if estimator / math changes)

• N/A - no methodology changes (skill/prompt file only)

Validation

• Tests added/updated: No test changes (prompt-only skill file)
• Manual verification required: see Verification section in the implementation plan

Security / privacy

• Confirm no secrets/PII in this PR: Yes

---

Generated with Claude Code

[github-actions]

Overall assessment

✅ Looks good

Executive summary

• No estimator, inference, or methodology changes; Registry not implicated.
• --pr workflow is clearly scoped with URL parsing, gh API fetches, and error handling.
• Minor spec mismatches around stdout suppression, detached HEAD handling, dimension count wording, and created_at field expectations.

Methodology

• None — only .claude/commands/review-plan.md updated; no estimator/math/SE/assumption changes.

Code Quality

• Severity: P3; Impact: Branch verification can misreport a mismatch when the repo is in detached HEAD (empty output from git branch --show-current), leading to misleading warnings. Concrete fix: detect empty/HEAD and report “detached HEAD; branch check skipped,” or use git rev-parse --abbrev-ref HEAD and handle HEAD specially. Location: .claude/commands/review-plan.md:L122.
• Severity: P3; Impact: gh auth status 2>/dev/null doesn’t suppress stdout, so auth status text can leak into the review output. Concrete fix: redirect stdout too (e.g., gh auth status >/dev/null 2>&1) or explicitly instruct to ignore the output. Location: .claude/commands/review-plan.md:L110.

Performance

• None.

Maintainability

• Severity: P3; Impact: Step 4 heading still says “Evaluate Across 8 Dimensions,” even though Dimension 9 is now part of the flow when --pr is set; this can confuse reviewers. Concrete fix: rename to “Evaluate Across 8 Dimensions (+ optional 9)” or “Evaluate Across 9 Dimensions when --pr.” Location: .claude/commands/review-plan.md:L195.

Tech Debt

• None.

Security

• None.

Documentation/Tests

• Severity: P3; Impact: The statement “The response includes … created_at” is inaccurate for review and inline comment fetches because those --jq filters omit created_at, which can mislead later steps. Concrete fix: add created_at to those --jq outputs or scope the statement to issue comments only. Location: .claude/commands/review-plan.md:L162.