Add /review-plan skill for staff-engineer plan review

[igerber]

Summary

• Add new /review-plan skill (.claude/commands/review-plan.md) that reviews Claude Code plan files from a staff engineer perspective
• Evaluates plans across 8 dimensions: completeness & executability, codebase correctness, scope, edge cases, architecture & patterns, plan execution risks, backward compatibility, and testing strategy
• Outputs structured feedback with CRITICAL/MEDIUM/LOW severity issues, checklist gaps cross-referenced against CLAUDE.md, questions for the author, and a verdict
• Strictly read-only: no files created, edited, or deleted by the skill

Methodology references (required if estimator / math changes)

• N/A — no methodology changes

Validation

• Tests added/updated: No test changes (read-only skill)
• End-to-end verified by running /review-plan ~/.claude/plans/functional-crunching-wave.md which produced structured output across all 8 dimensions with correct severity grading

Security / privacy

• Confirm no secrets/PII in this PR: Yes

---

Generated with Claude Code

[github-actions]

Overall assessment: ⚠️ Needs changes

• Adds a new .claude/commands/review-plan.md skill that standardizes staff‑engineer plan reviews across 8 dimensions.
• No estimator/methodology or runtime library changes; Methodology Registry entries are unaffected.
• Medium security concern: the skill auto‑reads any file referenced in the plan, which could pull in sensitive paths outside the repo.
• No tests or docs updates included (reasonable for a read‑only command).

Methodology

None. No estimator/inference code or default behavior changes in this PR; Methodology Registry is not implicated.

Code Quality

None.

Performance

None.

Maintainability

None.

Tech Debt

None.

Security

• P2 | Impact: The skill instructs reading all files referenced in a plan, which could include sensitive paths outside the repo (e.g., ~/.ssh/*) and lead to accidental secrets exposure if a plan is untrusted. | Fix: Before reading referenced files, restrict reads to the repo root or prompt for confirmation when a path is outside the workspace or in sensitive directories. (.claude/commands/review-plan.md:L50-L65)

Documentation/Tests

None.

[igerber]

/ai-review

[github-actions]

🔁 AI review rerun (requested by @igerber)

Head SHA: abc068628000ab08dbad8898dc5222d59b7d5212

---

Overall Assessment

✅ Looks good

Executive Summary

• Adds a new read-only /review-plan command that provides a structured, staff-engineer plan review framework.
• No estimator/math/runtime code changes in this PR.
• Minor methodology-review gap: the command doesn’t direct reviewers to the Methodology Registry when plans touch estimators.

Methodology

• P3 – Missing explicit methodology-source check in plan review instructions (.claude/commands/review-plan.md:L41-L69).
  Impact: Plans that modify estimators could be reviewed without referencing docs/methodology/REGISTRY.md / METHODOLOGY_REVIEW.md, risking missed methodological constraints.
  Concrete fix: Add a conditional step under Step 2 or Dimension 4/5: “If plan touches estimator math/assumptions/SEs, read docs/methodology/REGISTRY.md and METHODOLOGY_REVIEW.md, and compare against cited papers/docstrings.”

Code Quality

None.

Performance

None.

Maintainability

None.

Tech Debt

None.

Security

None.

Documentation/Tests

None.

[igerber]

/ai-review

[github-actions]

🔁 AI review rerun (requested by @igerber)

Head SHA: 8e060a425fc3886df5d6c3e096d427708d7b002e

---

Overall assessment

✅ Looks good

Executive summary

• Added a new Claude plan-review command with explicit read-only constraints and an 8-dimension evaluation framework (.claude/commands/review-plan.md:L1-L218).
• No estimator/methodology/SE/default behavior changes; methodology registry and papers not implicated.
• No tests or documentation updates appear necessary for a new .claude command file; no issues found in this diff.

Methodology

None.

Code Quality

None.

Performance

None.

Maintainability

None.

Tech Debt

None.

Security

None.

Documentation/Tests

None.