Methodology

How Agent Skill Picks filters public AI agent skills

Rule-curated, source-first, no fake testing claims.

Agent Skill Picks is a rule-curated directory. A directory profile means a skill passed public-data checks; it does not mean fisher has run a hands-on test. Human-tested reviews are reserved for future `/skills/` pages.

Current calibration

The current local calibration scanned 87 entries, found 37 index-eligible candidates, kept 40 reference entries, and recorded 5 rejections.

Publishing status: approved.

Score dimensions

  • Maintenance: recent verification and source freshness.
  • Provenance: official, vendor, or community source signal.
  • Popularity: public repository signal, capped for monorepos.
  • Completeness: license, description, platform, and install hints.
  • Hygiene: obvious risky install or secret-handling patterns.

Index gate

A skill profile can become indexable only when it has strong value and confidence scores, a source URL, a usable description, a creator, a category, platform evidence, and no stale-source flag. If any required signal is missing, the profile remains visible as a reference entry but uses `noindex,follow`.

List page gate

Category, use-case, platform, creator, and best-of pages are also gated. They need at least five index-eligible entries plus enough explanatory context before they are allowed into the sitemap.

Recheck policy

Listed and reference entries are designed to be rechecked weekly. If a source disappears, becomes archived, or goes stale, the entry is downgraded or rejected and the public data feed is regenerated.

Public data

The public structured feed is available at /api/skills.json. It exposes source, score, status, and indexing fields for public skill entries.