agentsclimarketplace

Skill reviewer

Skill justinramos101/agent-skill-kit/.agents/skills/skill-reviewer

Battle-tested Agent Skills for coding agents — source-grounded, failure-driven heuristics that audit and design real surfaces. Install with npx skills.

Install
npx -y skills add justinramos101/agent-skill-kit --skill skill-reviewer

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use when reviewing or editing proposed or existing skills in this catalog (public or repo-local) for structure, metadata, source safety, usefulness, validation, and release readiness. Do not use for building a new skill from source material; use `skill-curator` for that.

SKILL.md

2.8 KB, 567 tokens by cl100k_base, as published. Nobody here has run it

Skill Reviewer

Review skills for release readiness and report findings, ranked by severity, with a suggested fix per finding. Run the review to completion autonomously; edit files directly only when the request implies edit authority. Findings go in the reply — no ledgers, no saved workflow state.

Workflow

  1. Scope: the named skill, or every skill under skills/*/ and .agents/skills/*/ when asked for a catalog review.
  2. Read SKILL.md, skill.json, references, and evals for each target.
  3. Grade against the gates below plus docs/skill-authoring-principles.md.
  4. For source-inspired skills, apply references/source-safety-review.md.
  5. For load-bearing workflow skills, apply references/contract-drift-review.md.
  6. Report findings (SR-<area>-NNN), edits made, and validation output; run just check after any edit.

Review gates

Block or fix when:

  • skill.json is missing required fields, disagrees with the directory name, or its description differs from the SKILL.md frontmatter description
  • SKILL.md exceeds the house budget (~500-word target, 1500 hard ceiling — the binding number is the skill's own --word-max) or carries source provenance, marketing, or summaries — provenance belongs in skill.json
  • HITL ceremony is present: mode menus, bare-invocation menus, wait-for-user gates (the flat static-check shape must reject the skill)
  • the skill would not activate on realistic prompts, or its description omits anti-triggers for adjacent work
  • the content is generic prompt advice rather than gotchas, opinionated defaults, and named heuristics a model gets wrong unprompted
  • a must-not-miss rule (safety boundary, scoring invariant, hard never-do-X) lives only in a reference instead of SKILL.md
  • volatile external facts (vendor flags, API surfaces, versions) are pasted as static prose instead of pointing at the live source of truth
  • a looping workflow lacks a stop condition and iteration cap
  • content fails the deletion test — duplication, sediment, no-op instructions
  • copyrighted source material is reproduced or closely paraphrased
  • sensitive domains overclaim expertise or omit escalation boundaries
  • release-contract behavior changed without an equivalent deterministic gate (static check or CI), or runtime text drifted from what gates enforce
  • generated files (per-skill README.md, CATALOG.md) disagree with their source metadata (skill.json capabilities, catalog/catalog.json)

What ships with it: 4 files

6.8 KB alongside SKILL.md, 1 of them executable

evals/

references/

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.