check-skill-clis

name	check-skill-clis
description	Audit CLI dependencies referenced by active, installed, and disabled skills for Claude, Codex, and Copilot. Use when the user asks to check whether skill-required commands are installed, verify all skill CLIs, compare enabled versus disabled skill dependencies, or produce a report of missing, shadowed, or misresolved command-line tools used by local agent skills.

Check Skill CLIs

Overview

Use this skill to inventory skills across Claude, Codex, and Copilot surfaces, extract command-line tools referenced by those skills, and verify whether each CLI resolves on the current machine. Produce evidence: skill path, inferred agent ecosystem, status, command, resolved binary, version probe, and missing or suspicious entries.

Scope

Audit these ecosystems by default:

• claude: ~/.claude/**, plus Claude plugin-cache skill folders when present.
• codex: ~/.codex/skills/**, ~/.codex/plugins/cache/**/skills/**, and local authored skills such as /home/jmagar/.agents/src/skills/**.
• copilot: GitHub Copilot CLI and extension skill surfaces under ~/.config/github-copilot/**, ~/.config/gh/**, or plugin-cache paths when present.

Classify status conservatively:

• active: skill appears in the current session skill list, active config, symlink target, or enabled manifest.
• installed: skill exists on disk but no active evidence was found.
• disabled: skill exists under a disabled directory, disabled manifest entry, .disabled suffix, commented config entry, or explicit disable list.
• unknown: skill exists but status cannot be inferred without more runtime context.

Workflow

1. Inventory skill files

   • Run the bundled audit script from this skill:

     python3 <skill-dir>/scripts/audit_skill_clis.py --output docs/reports/skill-cli-audit.md
     

   • If the repo does not use docs/reports/, write to a clear local path such as /tmp/skill-cli-audit.md.

   • Include --json <path> when the user wants machine-readable output.

   • When the current session exposes an active skill list, pass it with repeated --active-skill <name> flags or --active-skills-file <path>.

2. Review extraction results

   • Treat script output as a starting point, not a final truth source.
   • Inspect any high-value skills whose CLI references look ambiguous, especially skills with shell snippets, allowed-tools, command examples, setup sections, or wrapper functions.
   • Add manually confirmed CLIs to the report if the script missed them.

3. Verify installed CLIs

   • Use command -v <cli> or which -a <cli> for each candidate.
   • Probe versions with low-risk commands only: --version, version, -V, or tool-specific documented version commands.
   • Do not run mutating commands while auditing availability.

4. Check resolution quality

   • Flag missing commands.
   • Flag suspicious resolution, such as a name collision where the resolved binary is clearly the wrong tool.
   • Flag multiple binaries when which -a shows conflicting candidates.
   • Flag skills that reference commands only available through shell functions or aliases, because non-interactive agents may not see them.

5. Report

   • Summarize totals by ecosystem and status.
   • List missing CLIs first, grouped by command and affected skills.
   • List suspicious or ambiguous CLIs next.
   • Include installed-good CLIs after the findings, not before.
   • Include exact commands run and output paths.

Expected Output Shape

# Skill CLI Audit

## Summary

| ecosystem | active | installed | disabled | unknown | cli refs | missing |
| --- | ---: | ---: | ---: | ---: | ---: | ---: |

## Findings

### Missing

- `rtk`: referenced by `<skill path>`; `command -v rtk` failed.

### Suspicious

- `lab`: multiple binaries found with `which -a`; verify the expected one comes first.

## Inventory

| ecosystem | status | skill | cli | resolution | version |
| --- | --- | --- | --- | --- | --- |

Script Notes

The bundled script intentionally favors recall over precision. It scans Markdown, YAML, JSON, TOML, and shell snippets for common CLI-reference patterns, filters obvious shell keywords, and checks each candidate with command -v, which -a, and safe version probes. Always review important findings before claiming a CLI is truly required by a skill.