# Handoff Feedback Reviewer

> Validate Claude Code feedback packets for the handoff-system loop. Use when Codex receives Claude output and must classify untraced actions, requirement drift, risk gates, validation failures, loop limits, and completion readiness.

- Skill: `zengxiansheng007/handoff-feedback-reviewer` (Agent Skill, multi-file: 12 files)
- Install (CLI): `npx skillmds@latest add zengxiansheng007/handoff-feedback-reviewer`
- Raw SKILL.md: https://api.skillmd.com/api/skills/zengxiansheng007/handoff-feedback-reviewer/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: Zengxiansheng007 (https://skillmd.com/u/zengxiansheng007)
- Updated: 2026-09-21
- Page: https://skillmd.com/skills/zengxiansheng007/handoff-feedback-reviewer

---


# Handoff Feedback Reviewer

## Operating Rules

- Never accept a free-form completion claim as completed.
- Feedback must be valid JSON matching the feedback schema.
- Every changed file, completed item, validation command, problem, option, and recommendation must map to FR/AC/confirmed decision.
- Mark unmapped items as `untracedAction`.
- P0/P1 drift blocks auto-continue and completion.

## Workflow

1. Parse Claude output.
2. Validate against feedback schema.
3. Check evidenceIndex and requirementTraceDelta.
4. Classify drift severity.
5. Check riskAndRollback and nextStepRecommendation.
6. Return one Codex state.

## Completion Gate

Completion requires:

- valid feedback packet;
- non-empty evidenceIndex;
- no untracedAction;
- driftReview severity is `none` or `allowed`;
- no high-risk action pending;
- completion review generated by Codex.

## References

- Read `references/feedback-review-rules.md` when feedback is invalid, ambiguous, or claims completion.

## Validation

Use the feedback validator bundled in `handoff-claude-executor`, then apply the completion gate above.

## Safety And Escalation

Do not mark completed when feedback is invalid, missing evidence, missing traceability, or contains P0/P1 drift. Ask the user before accepting high-risk next steps or requirement changes.