Skills catalog

Catalog of skills shipped with this repository, with live clauditor badge status next to each entry. Each badge reflects the skill's most recent iteration data on the dev branch; the URL is constant, so the image updates automatically when CI re-runs clauditor grade + clauditor badge and commits the refreshed .clauditor/badges/<skill>.json artifact.

Returning from the root README. This doc is the secondary placement for badges — per the placement hierarchy, a catalog page is the canonical one-glance view when the repo ships multiple skills. For the full placement guide, the color-logic table, and the embedding recipe, see docs/badges.md.

How this catalog is populated

1. User-facing skills live under src/clauditor/skills/<name>/SKILL.md and ship in the wheel. Maintainer-only skills live at the repo root under .claude/skills/<name>/SKILL.md and are excluded from the wheel and from clauditor setup.
2. clauditor grade <path-to>/SKILL.md produces the iteration sidecars under .clauditor/iteration-N/<name>/.
3. clauditor badge <path-to>/SKILL.md aggregates those sidecars into .clauditor/badges/<name>.json, which is committed so shields.io can fetch it via raw.githubusercontent.com.
4. This page's badges point at those JSON files via the shields.io endpoint pattern — one Markdown image per skill.

Adding a new skill? Run clauditor badge <path-to>/SKILL.md --url-only --repo wjduenow/clauditor --branch dev to get a ready-to-paste image line, then add a row below.

User-facing skills

/clauditor

The bundled Claude Code skill for the clauditor workflow itself — activates when a user's prompt mentions quality-testing or grading a Claude Code skill. Walks the author through clauditor validate (L1) and clauditor grade (L3), and can propose an eval spec via clauditor propose-eval when one is missing.

Source: src/clauditor/skills/clauditor/SKILL.md · Eval (maintainer-only, not in user installs): clauditor.eval.json

The eval spec is maintainer-only — test_args references .claude/commands/chunk.md in this repo's dev-local tree — so the badge's live metadata comes from CI runs against that fixture rather than from a user's project. The badge currently shows lightgrey · no data because no iteration exists on the dev branch yet; the first CI run of clauditor grade on the bundled skill will populate real L1/L3 scores.

Internal skills (maintainer-only)

These skills are bundled with the repo but are not exposed via clauditor setup and are not intended for end-user invocation. They exist for clauditor's own release workflow and development dogfooding. Listed here so maintainers have a one-glance quality view alongside the user-facing catalog above.

/release-manager

Cuts a clauditor-eval release to TestPyPI (from dev) or PyPI (from main). Runs pre-flight checks (branch, clean working tree, remote sync, test suite), determines the release version, builds with uv build, verifies with twine, tags, creates a GitHub release, monitors the publish workflow, and confirms the package appears on the target index. Stops and asks for confirmation before any side-effectful step.

Source: .claude/skills/release-manager/SKILL.md · Eval: .claude/skills/release-manager/SKILL.eval.json

Eval note: the eval runs with test_args: "test" and allow_hang_heuristic: false. The skill stops at the confirmation prompt in non-interactive mode — assertions cover the pre-flight summary and version output only. L3 grading is declared but excluded from the badge until the CLI-transport truncation issue is resolved (see issue #93 context on interactive-skill limitations).

/review-agentskills-spec

Live audit of the agentskills.io specification — fetches the current spec text, diffs it against clauditor's internal conformance assumptions (encoded in src/clauditor/conformance.py), and reports deltas. Run this when the spec is suspected to have shifted so clauditor lint can be updated in lockstep.

Source: .claude/skills/review-agentskills-spec/SKILL.md · Eval: .claude/skills/review-agentskills-spec/SKILL.eval.json

Excluded from clauditor setup's install symlinks so it does not appear in a user's / slash-command menu. Maintainers invoke it via the direct-path live-runner pattern — see .claude/rules/internal-skill-live-test-tmp-symlink.md for the testing contract.

Interpreting a badge

Color	Meaning
brightgreen	L1 assertions all pass; L3 grade met thresholds (or L3 not declared).
yellow	L1 all pass but L3 fell below the declared pass-rate / mean-score threshold.
red	Any L1 assertion failed, or L3 grading produced no scorable results.
lightgrey	No iteration has been recorded yet for this skill, or the eval spec declares zero L1 assertions. Run clauditor grade to populate.

See docs/badges.md#color-logic for the full decision table, including the DEC-007 "zero L1 assertions" edge case and the DEC-009 "L3 all parse-failed → red" branch.

Regenerating a badge locally

# Produce the sidecars (spends Anthropic tokens via `claude -p`).
clauditor grade src/clauditor/skills/clauditor/SKILL.md

# Aggregate the latest iteration into the badge JSON.
clauditor badge src/clauditor/skills/clauditor/SKILL.md

# Commit the regenerated artifact so shields.io re-fetches.
git add .clauditor/badges/clauditor.json
git commit -m "Refresh clauditor badge"

Or bundle the two steps into a CI workflow — see docs/badges.md#ci-integration for the pattern.