一键导入
requesting-review
Use when a diff is large or high-risk and Tier 0 inline self-review isn't enough — dispatches an independent reviewer subagent
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Use when a diff is large or high-risk and Tier 0 inline self-review isn't enough — dispatches an independent reviewer subagent
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when changing cogitation project settings (test/build/lint commands, branching conventions, graphify/codex toggles), or the user says "change config", "update settings", or "set test command"
Use when setting up a project for cogitation for the first time, or the user says "set up cogitation", "init", or "configure this project"
Use at the start of any conversation — establishes the cogitation workflow and requires invoking the right skill before responding, including before clarifying questions
Use for MEDIUM-sized work — new but contained behavior (one component, no architecture/data/API change) that deserves rigor without the design-doc + plan-doc ceremony of brainstorming
Use when the user wants to tailor the cogitation workflow — says "customise", "recustomise", "change the workflow", "make TDD less strict", "turn off finishing-branch", or wants a skill looser or disabled
Use when EC memories need cleanup — deduping, pruning stale entries, hygiene — or the user says "audit memories", "clean up EC", or "review what's stored"
| name | requesting-review |
| description | Use when a diff is large or high-risk and Tier 0 inline self-review isn't enough — dispatches an independent reviewer subagent |
Dispatch an independent reviewer for changes too large or risky for inline self-review. The reviewer gets precisely crafted context — never your session history — so it judges the work product, not your reasoning, and your own context stays clean.
Announce: "I'm using the requesting-review skill to dispatch an independent review."
Where this sits: Tier 1 of the review ladder (see @executing-plans Step 5). Default to Tier 0 inline self-review for small/medium changes. Escalate to Tier 2 external/adversarial review when you want a second model — only if codex is enabled in .cogitation/config.json:
/codex:adversarial-review --background (official codex-plugin-cc)@codex-review{test_command}
{lint_command}
{build_command}
If failures: Stop. Fix first using @debugging. Never request review on red.
Search EC so the reviewer gets codebase-specific signal:
ec_search:
query: code review feedback
type: learning
ec_search:
query: review pattern
type: pattern
Note prior feedback patterns to fold into the dispatch context.
Get the range:
BASE_SHA=$(git rev-parse origin/main) # or HEAD~N for a single task
HEAD_SHA=$(git rev-parse HEAD)
Dispatch the general-purpose agent (no named agent — one source of truth) with this template:
Task tool (general-purpose):
description: "Review code changes"
prompt: |
You are a Senior Code Reviewer. Review completed work against its plan/requirements
and identify issues before they cascade. Read the actual diff — don't trust summaries.
## What Was Implemented
{DESCRIPTION}
## Requirements / Plan
{PLAN_OR_REQUIREMENTS} (e.g. docs/plans/YYYY-MM-DD-<topic>.md)
## Git Range
Base: {BASE_SHA}
Head: {HEAD_SHA}
```bash
git diff --stat {BASE_SHA}..{HEAD_SHA}
git diff {BASE_SHA}..{HEAD_SHA}
```
## EC Context (codebase conventions / prior decisions)
{EC_CONTEXT}
## What to Check
- Plan alignment: all planned functionality present; deviations justified, not accidental
- Code quality: separation of concerns, error handling, types, DRY without premature abstraction, edge cases
- Architecture: sound decisions, security, integrates cleanly with surrounding code
- Testing: tests assert real behavior (not mocks); edge cases; all passing
## Calibration
Categorize by ACTUAL severity — not everything is Critical. Only flag issues that
would cause real problems during implementation or in production. Skip style,
wording, and formatting preferences. Acknowledge what was done well first so the
rest of the feedback is trusted.
## Output
### Strengths
### Issues
#### Critical (Must Fix) — bugs, security, data loss, broken behavior
#### Important (Should Fix) — architecture, missing features, test gaps
#### Minor (Nice to Have) — style, optimization, docs
(each: file:line · what's wrong · why it matters · how to fix)
### Assessment — Ready to merge? [Yes | No | With fixes] + 1-2 sentence reason
Placeholders: {DESCRIPTION} what you built · {PLAN_OR_REQUIREMENTS} plan path/task text · {BASE_SHA}/{HEAD_SHA} · {EC_CONTEXT} relevant decisions/patterns from Step 2.
@receiving-review.If approved:
"Review passed. Ready to finish the branch?"
If yes → Use @finishing-branch
If review reveals a durable pattern worth remembering:
ec_add:
type: learning
area: code-review
content: [What reviewers commonly catch in this codebase]
rationale: Recurring review feedback
| Do | Don't |
|---|---|
| Verify green before requesting | Request review on failing tests |
| Give the reviewer the diff range + plan | Dump code without context |
| Pass EC conventions/decisions | Assume the reviewer knows the codebase |
| Keep scope focused | Request review of WIP code |