AGENT SKILL

code-reviewer

Review completed implementation batches for spec compliance and code quality. Invoke after execution batches complete, before merging, or when a review gate is reached in the workflow.

适用职业开发工程师
License 许可
Apache-2.0
最近更新
2026-11-14
Stars
★ 1.6k
Forks
491

01 前言 Front Matter

name
code-reviewer
description
Review completed implementation batches for spec compliance and code quality. Invoke after execution batches complete, before merging, or when a review gate is reached in the workflow.

02 资源文件 1

03 SKILL.md 正文

Code Reviewer

Bundled runtime

Before executing a CLI line below, replace its leading SSF with node "<plugin-root>/scripts/spec-superflow.mjs"; <plugin-root> is the absolute directory two levels above this file. Never run SSF literally or call an ssf from PATH.

Two responsibilities: reviewing the recorded Git range and acting on findings with technical rigor. Review according to the persisted policy; verify feedback before implementing it.

Part 1: Requesting Review

Mandatory: one whole-range review for Native final; one review per planned wave for wave and legacy plans. Avoid redundant per-task or final reviews of unchanged evidence. Optional: when stuck, before refactoring, after fixing complex bugs.

Procedure

  1. Get the review range from recorded execution evidence. For final review resolve the base in order: the recorded state.review_base start anchor, then the review_base in the recorded isolation context, then git merge-base <target-branch> HEAD for a legacy/manual branch with an unambiguous target; head is git rev-parse HEAD. The CLI verifies this complete range. A failed final review retains the same base across fixes; wave repairs use the prior wave head. For a wave review use the recorded wave-start base. Never substitute HEAD~1; it misses earlier commits in multi-commit work.

  2. For Native final, the current executor reviews the complete diff locally and writes the report; do not dispatch a reviewer subagent. For user-authorized SDD with wave review, dispatch at most one reviewer for that wave using skills/code-reviewer/code-reviewer-prompt.md. If dispatch is unavailable, review locally.

  3. When dispatching, fill only [DESCRIPTION], [PLAN_OR_REQUIREMENTS], [BASE_SHA], [HEAD_SHA], [WAVE_ID], and [REVIEW_REPORT_FILE]. Do not send the whole planning bundle.

  4. Write a non-empty report at .superpowers/sdd/reviews/<wave-id>.md, then record that path with SSF execution review <change-dir> --wave <wave-id> --base <base-sha> --head <head-sha> --report .superpowers/sdd/reviews/<wave-id>.md --verdict <pass|fail>. For schema-2 plans, failed receipts also require --issue <stable-finding-id> for the blocking defect being repaired; reuse its ID across retries. Other findings keep their own IDs in the report. Identical failed input cannot consume another attempt.

  5. Critical/Important findings require a fail receipt, focused repair, one focused re-review, and replacement pass. Note Minor for later.

  6. At adjudication-required, wait for human authorization before another review.

Minimality And Scope

For unrequested complexity, cite the missing task requirement and diff line. Use Important for merge-blocking complexity and Minor for safe, behavior-neutral redundancy; never score by line count.

Part 2: Receiving Review Feedback

The Response Pattern

  1. READ feedback without reacting
  2. UNDERSTAND and restate requirement
  3. VERIFY against codebase reality
  4. EVALUATE: technically sound for THIS codebase?
  5. RESPOND: technical acknowledgment or reasoned pushback
  6. IMPLEMENT: one item at a time, test each

Severity Levels

Level Meaning Action
Critical Bugs, security, data loss, broken functionality Fix immediately
Important Architecture problems, missing features, poor error handling, test gaps Fix before next batch
Minor Code style, optimization, documentation polish Note for later

Forbidden Responses

Never: performative agreement ("You're right!", "Great point!"), blind implementation before verification, thanking the reviewer. Instead: restate the requirement, ask clarifying questions, push back with reasoning, or just fix it (actions > words).

Handling Unclear Feedback

Ask only about unclear findings that change a material decision; continue independent, already-understood repairs within scope. Do not turn an unclear optional comment into a global stop.

Source-Specific Rules

From user: Trusted — implement after understanding. Still ask if scope unclear. No performative agreement.

From external reviewer: Before implementing, check: technically correct for this codebase? breaks existing functionality? reason for current implementation? works on all platforms? reviewer understands full context? If suggestion seems wrong, push back with technical reasoning.

When to Push Back

Suggestion breaks existing functionality, reviewer lacks context, violates YAGNI, technically incorrect for this stack, legacy/compatibility reasons, conflicts with user's architectural decisions. Push back with technical reasoning, not defensiveness.

Implementation Order

  1. Clarify only the material uncertainty that blocks a repair
  2. Fix blocking issues (breaks, security)
  3. Fix simple issues (typos, imports)
  4. Fix complex issues (refactoring, logic)
  5. Test each fix individually, verify no regressions

Common Mistakes

Mistake Fix
Performative agreement State requirement or just act
Blind implementation Verify against codebase first
Batch without testing One at a time, test each
Proceeding without a wave receipt Record pass/fail via SSF execution review before the next dependent wave
Assuming reviewer is right Check if breaks things
Avoiding pushback Technical correctness > comfort
Unclear feedback Clarify the blocked finding while continuing independent repairs

Exception Handling

  • Parse failures: Report specific file, request regenerated review package
  • Missing files: Regenerate via scripts/review-package. Empty diff = nothing to review
  • User interruption: Re-read review report on resume, continue from next unreviewed batch

04 同仓库更多 Skills

hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-review-with-annotations

Create, inspect, reply to, update, or resolve situated Worktable annotations. Use when feedback belongs to a Doc section, block, text range, or HTML Doc, or when the user asks to address existing review feedback and coordinate a requested artifact edit.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-find-and-synthesize-context

Find, assess, and synthesize context from Worktable Docs and Records. Use when the user asks what a workspace contains, wants workspace-backed research, needs current facts reconciled across artifacts, or requests a cited synthesis without changing workspace state.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-create-or-update-docs

Create, edit, rename, or maintain narrative Worktable Docs. Use for plans, decisions, briefs, notes, research syntheses, and other durable prose when the user requests a new artifact or a safe update to an existing one.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-create-or-manage-records

Organize repeated structured information as durable Worktable Records. Use for research datasets, inventories, directories, pipelines, trackers, and other collections whose items need identity, fields, filtering, links, validation, or individual updates.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-create-or-manage-interactive-html

Create or improve durable visual and interactive artifacts in Worktable. Use for dashboards, trackers, calculators, mini apps, slide presentations, and other requests that benefit from a polished HTML Doc.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-collaborate-threads

Read, answer, and hand off messages through durable Worktable threads with users or other authenticated participants. Use when the user asks to check addressed messages, reply to a participant, continue a thread, request another agent's help, wait briefly for a response, or watch for comments and messages sent to the agent.

★ 1.6k