メニュー
Reviews an agent skill's SKILL.md against authoring best practices and reports findings by severity. Use when the user asks to review, audit, or check the quality of an agent skill or a SKILL.md.
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
Creates agent skills from an OSS project's documentation. Splits the documentation other than README.md into per-topic skills/<oss>-<topic>/*.md files, and has README.md and SKILL.md reference them, so human-facing documentation and skills share the same body and avoid double maintenance. Use when turning an OSS project's documentation into skills, or restructuring existing documentation into a skill-based layout.
Writes commit messages following Conventional Commits in English. Use when the user asks to write, format, or improve a commit message.
Translates the document or text in context into English. Use when the user asks to translate something into English, or invokes /en.
Conventions and commands for Go projects — testing, linting, JSON Schema generation, and error handling. Use when developing, testing, or linting Go code in this repository.
Baseline rules for this repository — use English, never commit to default branches, never rewrite commit history. Use at the start of any coding or git task.
Creates a branch, commits the changes, and opens a pull request whose description summarizes the full diff between base and head. Use when the user asks to create a pull request, open a PR, or submit changes for review.
| name | review-skill |
| description | Reviews an agent skill's SKILL.md against authoring best practices and reports findings by severity. Use when the user asks to review, audit, or check the quality of an agent skill or a SKILL.md. |
Review the target agent skill against the best practices, then report findings.
The argument identifies the target skill. Resolve it to the skill directory:
SKILL.md file path (e.g. skills/pr/SKILL.md) → use its parent directory.skills/pr) → use it directly.pr) → locate its directory, e.g. find . -type d -name <name> or check skills/<name>. If it matches more than one or none, ask the user which skill to review.Then read every file under that directory, with SKILL.md as the primary subject.
Pull the repository:
test -d "$(ghq root)/github.com/agentskills/agentskills" || ghq get agentskills/agentskills
git -C "$(ghq root)/github.com/agentskills/agentskills" pull
If the command is denied, tell the user.
Read the first best practice in full: ~/repos/src/github.com/agentskills/agentskills/docs/skill-creation/best-practices.mdx
If the file doesn't exist, tell the user.
Download the second best practice, then read the downloaded file in full. Don't use the WebFetch tool — it doesn't necessarily read all the content.
tempdir=$(mktemp -d)
temppath=${tempdir}/best-practices.md
curl -Lq -o "$temppath" https://platform.claude.com/docs/en/agents-and-tools/agent-skills/best-practices.md
echo "$temppath"
If the URL is 404 or the download is denied, tell the user.
Compare the skill against the best practices. Pay particular attention to:
description: includes both what the skill does and when to use it, written in third person, with discovery keywords.name: valid (≤64 chars, lowercase/numbers/hyphens, no reserved words). A short, well-understood abbreviation is fine — do not force a long gerund name at the cost of invocation convenience.SKILL.md under ~500 lines; references one level deep; large content split into separate files.Group findings by severity and end with prioritized actions:
## Review: <skill path>
### 🔴 High
- <finding, with current vs. suggested>
### 🟡 Medium
- <finding>
### 🟢 Low
- <finding>
### Good points
- <what the skill already does well>
### Recommended actions (by priority)
1. <most impactful fix first>
After reporting, offer to create a revised SKILL.md.