Audit Tools

---
name: audit-tools
description: "Audit tools/ for structure, tests, CLI quality, and docs. Use when checking tool health, validating tools, or after adding a tool. Also 'check the tools', 'which tools need work'."
allowed-tools: Read, Grep, Glob, Bash
disable-model-invocation: true
argument-hint: "[tool-name] [--fix]"
---

# Tools Audit

**Goal:** surface every tool in `tools/` that is missing files, under-documented, untested, or has a broken CLI surface, by running the 10 checks in [CHECKS.md](CHECKS.md) against each tool and reporting a severity-grouped result per tool plus a fleet summary. Report only — findings go to a human. Also the right skill for reviewing tool quality after modifying a tool, and for questions like "are our tools documented".

## Repo Context Probe

If `.claude/skill-context/audit-tools.md` exists, read it and honor its declarations; otherwise use the generic defaults described below.

The context file is where a repo declares its CLI-naming convention (e.g. a `valor-*` entry-point prefix) and its skill-scaffolding command, so the `[cli-registered]` check and the "missing structure" next-step name the repo's real conventions. When the file is absent, the generic baseline applies: a tool is "cli-registered" when it has a console-script entry in `pyproject.toml [project.scripts]` under any name, and missing structure is scaffolded by hand.

## Quick start

If `$ARGUMENTS` names a specific tool (e.g., `/audit-tools sms_reader`), audit only that tool. Otherwise audit all tools.

### Step 1: Discover tools

```bash
ls -d tools/*/ | grep -v __pycache__ | grep -v _template | sort
```

### Step 2: Run checks

Read [CHECKS.md](CHECKS.md) for the full check definitions. For each tool, run all 10 checks and record PASS/WARN/FAIL per check.

### Step 3: Report

Present findings using this format:

```
## Tools Audit Report

### tool-name (6/10 PASS, 3 WARN, 1 FAIL)

#### FAIL
- [test-coverage] 3 capabilities in manifest, only 1 has test coverage (search, analyze untested)

#### WARN
- [cli-help] --help output is 2 lines — should describe arguments, options, and examples
- [error-docs] README has no error handling section
- [output-types] Functions return untyped dicts — return types not documented

#### PASS
- [manifest-exists] manifest.json present and valid
- [readme-exists] README.md present with required sections
- [tests-exist] tests/ directory with test_sms_reader.py
- [inputs-documented] All public functions have typed parameters
- [examples] README has working code examples
- [cli-registered] Registered in pyproject.toml [project.scripts] (under the repo's CLI-naming convention; see the repo context file if present)

---

### Summary

| Status | Count |
|--------|-------|
| Fully compliant | 12 |
| Has warnings | 5 |
| Has failures | 3 |

Total: 20 tools, 164/200 checks passing
```

### Step 4: Disposition

This audit is **report only** — findings are presented for human review. If `--fix` was passed, create a GitHub issue with the full report for tracking.

## After the audit

This skill produces findings only. Next steps are decided by the human:
- Fix individual tools based on findings
- Create GitHub issues for tools that need significant work
- Scaffold missing structure for incomplete tools using the repo's skill/tool scaffolding command if its context file declares one (otherwise scaffold by hand)
- Delete tools that are abandoned (e.g., empty placeholders)