Agent skill
skill-reviewer
Grades a SKILL.md file on eight short dimensions (trigger description, when not to use it, inputs, steps with named outputs, quality bar, mistakes to avoid, standalone, length).
Filed under Content and SEO.
From matteotitta/gtm-skill-template · 3 skill entries · 0 · pushed 2026-10-14
What it does when it runs
Grades a SKILL.md file on eight short dimensions (trigger description, when not to use it, inputs, steps with named outputs, quality bar, mistakes to avoid, standalone, length). Scores each 0 to 2, quotes the evidence, and lists specific fixes in priority order. Use it after you write or change a skill, or before you share one. Triggers: "review this skill", "grade my SKILL.md", "check skill quality", "audit this skill", "is this skill ready". Not for reviewing what a skill produced, such as a post or a landing page; review that output on its own terms.
Automated analysis of the skill and the 0 files bundled beside it. A skill’s own description is written to be selected by an agent, so it describes the job and not the dependencies.
- Keys and connectors you must supply
- None found.
- Hosts it reaches
- No third-party host appears in the skill or its bundled files.
- Tool permissions it declares
- No
allowed-toolsin the frontmatter. It only issues instructions, so there is nothing to bound. - Actions present in the files
- None. Instructions only.
Ask about skill-reviewer
Opens your assistant with this page's recorded context. Check the linked sources before relying on it.
Install it
View source on GitHub ↗git clone --depth 1 --filter=blob:none --sparse https://github.com/matteotitta/gtm-skill-template.git /tmp/gtm-skill-template git -C /tmp/gtm-skill-template sparse-checkout set "skills/skill-reviewer" mkdir -p ~/.claude/skills/skill-reviewer cp -R "/tmp/gtm-skill-template/skills/skill-reviewer/." ~/.claude/skills/skill-reviewer/
Picked up without a restart. A project skill of the same name is shadowed by your personal one. For one repository only, swap ~/.claude/skills for .claude/skills. Claude Code docs ↗
Or take the whole library
This repo ships a .claude-plugin manifest, so Claude Code can install all 3 skills at once. Plugin skills are invoked as /<plugin>:<skill>, so they never collide with your own.
/plugin marketplace add matteotitta/gtm-skill-template /plugin
The folder is the same in every client that implements the format — 46 of them — so if yours is not above, only the destination changes.
The skill
Source on GitHub ↗Reproduced in full from matteotitta/gtm-skill-template/blob/9c084e1df6745cdc625d1b1e3ae453b873fda701/skills/skill-reviewer/SKILL.md, which is licensed MIT (skill frontmatter). 789 words, 9 headings.
Skill reviewer
Grade a skill definition, not a skill's output. A good skill runs the same way for a stranger as it does for its author. This review finds the gaps that make it guess.
When to run
- You just wrote a skill from the template at the root of this plugin.
- You changed an existing skill and want to know if it got better or worse.
- You adapted someone else's skill with
/stealand want to check it before using it. - A skill gives uneven results from run to run, and you want to know why.
When not to
- To judge a piece of content the skill wrote. That is a content review, not a skill review.
- To design a new skill from nothing. Start from the template, then come back here.
Inputs
Required: the path to one SKILL.md, or the text pasted in. If the path doesn't exist, say so and ask for the right one.
Optional: the folder the skill lives in, so you can check that every file it mentions is really there. The template at the root of this plugin, as the reference shape.
Steps
- Read the file. Parse the frontmatter. List every heading. Produces: a section inventory.
- Score the eight dimensions below, 0 to 2 each. For every score, quote the line or name the missing section that earned it. Produces: the score table.
- Write the fixes. One fix per point lost, specific enough to paste in. Order them by how much they change the skill's behaviour, not by dimension number. Produces: the fix list.
- Run the simple test. Ask: would a careful reader say this skill is overbuilt for its job? If yes, say which sections to cut. Produces: one line, cut or keep.
The eight dimensions
Score 2 = clearly there and specific. 1 = there but vague or partial. 0 = missing.
| # | Dimension | What a 2 looks like |
|---|---|---|
| 1 | Trigger description | description says what it does, gives 3+ phrases a user would say, and is under 1,000 characters |
| 2 | When not to use | Names 1+ nearby request it should not handle, and where that request should go |
| 3 | Inputs | Required vs optional is clear; says where to read them and what to do if one is missing |
| 4 | Steps with named outputs | Numbered steps, and each one says what it produces |
| 5 | Quality bar | "What good looks like" items are checkable, not adjectives |
| 6 | Mistakes to avoid | 2+ mistakes specific to this task, each with what to do instead |
| 7 | Standalone | Mentions no skill, file, tool, rule or API key that isn't shipped with it or marked optional |
| 8 | Length | Says each thing once; no padding, no long examples that belong in a separate file. Most skills fit in 60 to 150 lines |
Total out of 16. 14 to 16: ready to use. 10 to 13: usable, fix the 0s first. Under 10: rewrite before you rely on it. Any 0 on dimension 7 blocks sharing, whatever the total.
Output
Reply in chat with this shape:
- Skill: name and path
- Score: X / 16, and the verdict line from above
- Table: dimension, score, one-line finding with the quoted evidence
- Fixes: numbered, highest impact first. Each says where in the file and what to write
- Cut or keep: the one-line answer from step 4
If the user asks for a file, save it next to the skill as review-MMYY.md.
What good looks like
- Every score points at a line in the file, or at a section that isn't there.
- Every fix can be applied without asking a follow-up question.
- The verdict matches the numbers.
- Counts are real. If you say "3 trigger phrases", list them.
Mistakes to avoid
- Grading on looks. A skill with every heading and vague content inside scores 1s, not 2s. Read what's under the heading.
- Missing hidden dependencies. Look for names of other skills, scripts, folders, connected tools and saved rules. Each one the user may not have costs points on dimension 7 unless it's marked optional.
- Fixes that add length. Prefer cutting or tightening. A longer skill is not a better one.
- Rewriting the whole skill unasked. Give the fixes. Offer the rewrite in one line.
- Inventing a standard. Grade against the eight dimensions here and the template in this plugin, not your own taste.
Other skills for the same job
Different authors, same problem. Matched on the words in the skill name, across every library in the catalogue except this one.
- tweet-draft-reviewer by BrianRWagner · 450
- pipeline-reviewer by janskuba · 206
- content-reviewer by taizen-ai · 8
- qa-reviewer by V00nikos · 1
Need help setting it up?
This page tells you what skill-reviewer does and what it needs. Cheetah builds the agent setup it runs inside: data, CRM, sequencing and the guardrails.
Book a call →The directory stays free. There is nothing gated behind this.