agentleFS
Sign inStart free
RifteoRifteo/skills/skill-benchmark/SKILL.md

skill-benchmark

Skill37 starsChanged 5 months agoMITView on GitHub

Scores any SKILL.md across 5 quality dimensions, runs a compatibility check against 50+ AI agents, and returns a ranked fix list. Trigger when the user asks to benchmark, score, or audit a skill, wants to check cross-agent compatibility (Cursor, Windsurf, Gemini, etc.), needs a ranked fix list before publishing a SKILL.md to the community repo, is comparing two skills before a pull request, or wants to audit a SKILL.md for compatibility before opening a PR.

3 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

8 rules. 0 are common, 5 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If the user provided a path, read that file directly.

    line 19
  2. Distinctive:

    If not, search in this priority order:

    line 20
  3. Distinctive:

    If multiple are found, list them and ask the user to confirm which one.

    line 24
  4. Very specific:

    If no SKILL.md is found after exhausting all search paths, stop and output: [skill-benchmark error] No SKILL.md found — provide a file path …

    line 25
  5. Distinctive:

    Read the full file into context before proceeding.

    line 27
  6. Very specific:

    Read references/agent-profiles.md.

    line 63
  7. Very specific:

    Use the template in references/report-template.md.

    line 79
  8. Distinctive:

    After presenting the report, offer exactly these three options:

    line 91

“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Skill Benchmark

Reads a target SKILL.md, scores it across 5 weighted dimensions (100 pts total), runs a static compatibility analysis against 50+ agent profiles, and returns a structured report with a ranked fix list.

No agents need to be installed. Everything is static analysis — you read the skill file, apply the rubric below, and produce the report.

Protocol

Step 0 — Locate the target skill

  1. If the user provided a path, read that file directly.
  2. If not, search in this priority order:
    • Current directory: SKILL.md, */SKILL.md
    • Project skill paths: .claude/skills/, .agents/skills/, .cline/skills/
    • Global paths: ~/.claude/skills/, ~/.config/agents/skills/
  3. If multiple are found, list them and ask the user to confirm which one.
  4. If no SKILL.md is found after exhausting all search paths, stop and output: [skill-benchmark error] No SKILL.md found — provide a file path to continue.
  5. Read the full file into context before proceeding.

Step 1 — Intake

Extract and note:

  • Skill name (frontmatter name: field)
  • Description (frontmatter description: field, full text)
  • Body length (line count)
  • Has bundled resources? (scripts/, references/, assets/ subdirs)
  • Explicit agent targets (any --agent flag mentions or agent-specific sections)

Step 2 — Score each dimension

Apply the rubric in references/rubric.md. For each dimension, note the score, a one-sentence rationale, and the single most impactful fix.

#DimensionMax pts
1Trigger quality25
2Instruction clarity25
3Agent-agnostic design20
4Self-containment15
5Output definition15
Total100

Scoring is strict. A score of 70+ means the skill is publishable. Below 50 means it will silently fail on most agents.

Step 3 — Run compatibility matrix

Read references/agent-profiles.md. For each of the 7 agent profiles, check the target skill for the compatibility signals listed. Assign one of:

  • ✅ Compatible — no blocking signals found
  • ⚠️ Partial — one or more soft-fail signals (skill may work but will degrade or skip steps)
  • ❌ Likely broken — one or more hard-fail signals

Then map each named agent to its profile result. When an agent appears in multiple profiles (e.g., Claude Code in A and C), assign the worst result across all applicable profiles: ❌ overrides ⚠️ overrides ✅.

Step 4 — Generate the report

Use the template in references/report-template.md.

Fill in every section. Do not skip the "Top fixes" section — it is the most actionable part and the reason the skill exists.

Sort the 54 agents into three groups (Compatible / Partial / Broken) and list them comma-separated. Do not list each agent on its own line.

Step 5 — Offer next steps

After presenting the report, offer exactly these three options:

  1. Auto-fix — apply the top-ranked fix and re-score immediately
  2. Deep dive — expand any dimension score with full evidence from the skill text
  3. Export — save the report as skill-benchmark-report.md in the same directory as the evaluated skill

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free