skill-audit

Files (2)

• SKILL.md
• subagent.rst · references/

SKILL.md (first lines)

## Codex execution

Delegation is optional. Read references/subagent.rst only when delegation is useful or requested. It does not install a named agent or grant permissions.

# Skill Audit

Run the non-inferable-value rubric over a target `SKILL.md` (or the session's
skill diff). The rubric below is self-contained. In the agent-extensions
repository, CONTRIBUTING.md → "Skill content conventions" holds the authoring
rules it enforces; elsewhere, do not search for that file.

## Rubric
1. **Non-inferable value (Biggs):** could a fresh model write this verbatim? If yes → cut.
2. **Conciseness (SkillsBench):** verifier-facing, non-inferable detail over comprehensive prose.
3. **Version pins:** any encoded library/tool API surface is pinned in `compatibility:`.
4. **Verify-canonical guard:** fast-moving/high-stakes subjects point to canonical docs.
5. **Frontmatter/discovery:** `name` matches dir; `description` triggers well; structure-standard clean.
6. **Complexity contract:** the skill earns its context cost.

## How to run

Read the target skill or diff and score each rubric item. Return severity-rated
findings (CRITICAL / MODERATE / MINOR) with file:line anchors and a concrete
keep/cut/compress/pin recommendation. End with KEEP / COMPRESS / REMOVE.
This is a read-only audit; do not edit the target files.

## Optional delegation

[references/subagent.rst](references/subagent.rst) contains the subagent outline,
handoff inputs, execution boundaries, and expected result. Read it when delegating
would help or the user asks to “create a subagent to execute this.” Otherwise,
work directly from this skill; the reference does not need to be loaded.
A worker that received this skill in a handoff is the auditor: it applies the
rubric itself and does not delegate again.

From nq-rdl/agent-extensions, CC-BY-4.0.

More from nq-rdl/agent-extensions