| name | evaluate-legibility |
| description | Cold-read a SKILL.md with a zero-context agent reader to check its intent is legible. Use when validating whether a skill says clearly when to invoke it and how to start. |
| args | <plugin/skill-name> |
| allowed-tools | Task, Read, Glob, Bash(bash *), TodoWrite |
| argument-hint | git-plugin/git-commit |
| model | opus |
| agent | general-purpose |
| created | "2026-06-13T00:00:00.000Z" |
| modified | "2026-07-18T00:00:00.000Z" |
| compatibility | claude-code |
| reviewed | "2026-06-13T00:00:00.000Z" |
/evaluate:legibility
Dispatch an isolated, zero-context agent reader at a single SKILL.md and
ask it the only two questions that matter for reuse: from this file alone, can
you tell WHEN to invoke the skill, and can you carry out its FIRST concrete
action? What confuses a cold agent reader is what will silently degrade
auto-invocation and first-step execution for every future agent that loads the
skill.
This is the comprehension gate — read and critique, never execute. It
reuses the dispatch contract of agent-patterns-plugin:cold-read-gate (isolated
haiku reader, QUESTIONS / HESITATIONS / verdict schema, triage before
acting). The difference: the reader here is an agent loading a skill, not a
human triaging an outward artifact — so the persona and the triage table are
skill-doc-specific. See that skill for the dispatch rationale; this skill does
not restate it.
The verdict is advisory, not a CI gate: haiku is non-deterministic, so a
needs-revision is a prompt to look, never a merge block.
When to Use This Skill
| Use this skill when... | Use alternative when... |
|---|
| Checking a SKILL.md reads clearly to a fresh agent | Testing whether the skill executes correctly -> /evaluate:matrix |
| Triaging why a skill never auto-invokes | Need structural/lint validation -> scripts/plugin-compliance-check.sh |
| Reviewing a newly-authored or just-edited skill | Gating outward issues/docs for a human reader -> agent-patterns-plugin:cold-read-gate |
| Want a cheap, no-execution legibility read | Benchmarking effectiveness vs a baseline -> /evaluate:skill |
Context
- Available skills: !
find . -path '*/skills/*' -name 'SKILL.md' -not -path '*/.claude/worktrees/*'
Parameters
| Parameter | Default | Description |
|---|
<plugin/skill-name> | required | Target skill as plugin-name/skill-name |
Execution
Execute this legibility gate:
