ワンクリックで
doc-review
Review documentation quality — AI notes, Slite notes, plans, and specifications
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
メニュー
Review documentation quality — AI notes, Slite notes, plans, and specifications
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Create a Claude Code rule, skill, or agent — global (~/.claude) or project-local (.claude)
Code review using OpenAI Codex CLI with project context (rules, diff, conventions)
Multi-perspective code review from base branch with 3 independent reviewer agents
Analyze working tree changes, map to existing commits, and create fixup commits for autosquash
Analyze working tree changes, plan logically minimal NEW atomic commits per hunk (no fixup), and execute them
Review every commit message since base branch and add reword fixups where the message needs improvement (non-interactive)
| name | doc-review |
| allowed-tools | Bash(deno run:*), Bash(ls:*), Read, Glob, Grep, Agent |
| argument-hint | <path-or-keyword> |
| description | Review documentation quality — AI notes, Slite notes, plans, and specifications |
$ARGUMENTS = {path-or-keyword}
~/Compost/AI-Notes/2026-03/04-1200-design.md)/doc-check.Based on the argument:
deno run -A ~/.claude/skills/ai-notes/notes.ts list --limit 1 to get the most recent note, then read itdeno run -A ~/.claude/skills/ai-notes/notes.ts list --limit 20 and filter, or use Grep across ~/Compost/AI-Notes/If the document cannot be found, inform the user and STOP.
Determine the document type from content:
If the document references specific code, APIs, or repositories:
Use the Agent tool to spawn 1 agent.
subagent_type: "general-purpose"
Prompt (adapt review criteria based on document type):
You are a technical document reviewer. Review the following {document_type} for quality and completeness.
## Methodology
1. Read the document carefully
2. If the document references code or implementations, use Read/Glob/Grep to verify those references are accurate
3. Check the document's internal consistency (do later sections contradict earlier ones?)
## Review criteria for {document_type}
### For 仕様書 (Specification):
- Are requirements clear and unambiguous?
- Are edge cases and error scenarios covered?
- Are interfaces and data formats fully defined?
- Is the scope clearly bounded (what is NOT included)?
- Are acceptance criteria defined?
### For 設計書 (Design document):
- Is the architectural approach sound? Are there obvious alternatives that weren't considered?
- Are component interactions and data flows clearly described?
- Are failure modes and error handling strategies addressed?
- Does the design align with referenced code/existing architecture?
- Are assumptions explicitly stated?
- Are trade-offs acknowledged?
### For 計画書 (Plan):
- Are implementation steps in a logical order?
- Are dependencies between steps identified?
- Are risks and mitigation strategies realistic?
- Is the testing strategy sufficient for the scope of changes?
- Are there missing steps that would be needed in practice?
### For all types:
- **Technical accuracy**: Do code examples, API references, and technical claims match the actual implementation?
- **Logical completeness**: Are there gaps in reasoning or missing considerations?
- **Feasibility**: Are the proposed approaches practically achievable?
- **Consistency**: Does the document contradict itself?
- **Assumptions**: Are unstated assumptions that could invalidate the plan?
## What to IGNORE
- Formatting, typos, Markdown syntax
- Writing style preferences
- Minor wording improvements
Document content:
{document content}
{referenced code context if any}
For each issue: section/heading where the issue is, Severity (Critical/Warning/Notice), what is missing or wrong, suggested improvement.
Display-only. Do NOT modify the document.
Severity:
(★★★): Technical inaccuracy, logical gap that invalidates the plan(★★☆): Missing consideration, unstated assumption, feasibility concern(★☆☆): Improvement suggestion that would strengthen the document## ドキュメントレビュー結果
**対象**: `{path}` | **種別**: {document_type}
---
### 1. タイトル (セクション名) (★★★)
> 問題の要約。
具体的な改善案。
---
### 2. タイトル (セクション名) (★★☆)
> 問題の要約。
...
Parse $ARGUMENTS and execute from Step 1.