Plugin eval

Listed by its author as plugin-eval

Static lint for Claude Code plugins and skills, with experimental LLM scoring for skills

by wshobson · MIT · 42k GitHub stars

DoesPrompts, agents and LLMs Validation and QA

Source on GitHub ↗

into

curl -fsSL https://api.cheatcodehq.com/install.sh | sh -s -- wshobson-plugin-eval --to all

Copies 1 skill into each app's skills folder (Claude Code: ~/.claude/skills), after checking every file against wshobson/agents at commit 9b15b34. Nothing else on your computer changes. To undo it, run the terminal command with --uninstall at the end; that also removes what the app added.

Then start a new session in your AI app and ask for the work in plain words; it picks the skill from its description. In Claude Code you can also type /wshobson-plugin-eval:evaluation-methodology.

See it in use

One real session, replayed step by step: the request, each step at the moment it happened, and the answer, unedited. Long waits are shown faster, with the real time in the corner. Sample skill draft, score report and notes invented for this example: Acme Notes is a fictional note-taking app, and the report was not produced by the real scorer.

The result

score-fix.md

A plain explanation of the score, which of your plans help or hurt, and a rewritten skill file to score again.

score-fix.md, the file the session saved.

Open the whole result

Try it

After installing, start a new session and say: My skill in SKILL.md scored low in report.json. Explain why in plain words and give me a fixed version.

What else it can do, and its limits

1 skill that explains how the PluginEval quality score for Claude Code skills and plugins is built: the lint checks, six flags that cut the score, dimension weights and badges
Reads a score report and says which changes move the number and which do not
The scoring tool itself is not in this package: running plugin-eval needs a terminal and a separate install
The model judge and repeated run layers are experimental by the tool's own account, so it treats their numbers as rough signals
Without the tool it estimates a new score and asks you to confirm it, as in this run

Sample input Recorded Oct 24, 2026 in Claude Code 2.1.308 with Sonnet 5.5. The video replays that session; nothing in it was acted or edited.

What's inside 1 skill

evaluation-methodologyPluginEval quality methodology, covering dimensions, rubrics, and scoring formulas. Use this skill when understanding how plugin quality is measured, when interpreting a low score on a specific dimension, when deciding how to improve a skill's triggering accuracy or orchestration fitness, when setting score thresholds for your marketplace, or when explaining quality badges to external partners like Neon.
PluginEval quality methodology, covering dimensions, rubrics, and scoring formulas. Use this skill when understanding how plugin quality is measured, when interpreting a low score on a specific dimension, when deciding how to improve a skill's triggering accuracy or orchestration fitness, when setting score thresholds for your marketplace, or when explaining quality badges to external partners like Neon.

About

Multi-harness agentic plugin marketplace for Claude Code, Codex, Cursor, OpenCode, GitHub Copilot, Google Antigravity, and Pi

Where this comes from. The public repository wshobson/agents at commit 9b15b34, under its MIT license, copied without changes. wshobson is not affiliated with Cheatcode and has not claimed this listing yet. New releases are repackaged after review, never synced automatically. If this is your project, write to admin@cheatcodehq.com to claim or remove it.

Something wrong with this listing, or it did not install? Tell us

Versions

v0.0.0-g9b15b34Listed from wshobson/agents at 9b15b34 · 2026-09-28