What does this skill do, and when should you use it?
cc-inspect is a Claude Code skill that runs a Bash script to scan user- and project-level `.claude/` directories, collecting skills, plugins, MCP servers, commands, hooks, and memory folders, then generates a self-contained HTML dashboard and opens it in your browser. Its distinguishing feature is scope-aware display (User / Project / Local layers) and grouped cleanup recommendations for auto-memory directories — stale, orphaned, and duplicated entries each get a copy-to-clipboard archive command, and nothing is ever deleted. It has zero dependencies beyond Bash and Python 3.
Runs inspect.sh, which scans ~/.claude/ and the current project's .claude/ directory; reads mcpServers and hooks from settings. at all scopes; walks skills, plugins/marketplaces, and commands paths; decodes memory-folder ownership from the ground-truth cwd field in session .l files; MD5-hashes memory files for duplicate detection; then writes a self-contained HTML file to /tmp/cc-inspector.html and opens it with the open command. File contents are never read into the dashboard — only locations, sizes, and timestamps are reported.
- A heavy Claude Code user wants a visual inventory of every skill, plugin, and MCP server they've installed
- After months of use, `~/.claude/projects/` has grown to 20+ memory directories including leftovers from deleted projects — the user needs safe cleanup guidance
- A developer suspects leftover hooks or MCP config in a project or local settings and wants to audit by scope
- Before migrating machines or reinstalling, a user wants a complete snapshot of their Claude Code environment
- A collaborator wants to see which commands and hooks a project repository defines
How do you install this skill?
- The skill is macOS-only; Linux users must change open to xdg-open manually, and Windows is unsupported.
- The dashboard displays MCP server and hook command lines, which may expose API keys or tokens — beware of information leakage when sharing screenshots.
- Orphan detection samples only the first 100 lines of session l files and may produce false positives; verify paths before running any copied rm dedupe commands.
- Errors are silently suppressed (2>/dev/null); missing dashboard content gives no failure reason — run the script manually to debug.
- No versioning or changelog; track the repository yourself for updates. Publisher is not verified by the FollowSkills registry.
- Shell / CLI
- Local filesystem
BashPython 3macOS (uses `open`; Linux needs `xdg-open` swap)Claude Code CLI
git clone https://github.com/howardpen9/cc-inspect.git && cd cc-inspect && mkdir -p ~/.claude/skills/cc-inspector && cp inspect.sh SKILL.md ~/.claude/skills/cc-inspector/ && chmod +x ~/.claude/skills/cc-inspector/inspect.sh. Prerequisites: macOS (the script uses open; Linux users must swap in xdg-open), Python 3, and the Claude Code CLI.
git clone --depth 1 https://github.com/howardpen9/cc-inspect.git ~/.claude/skills/cc-inspectGenerated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Dashboard opened in browser.
In any Claude Code conversation, type /inspect. The script runs and a browser tab opens with the full dashboard; the assistant replies only with "Dashboard opened in browser." Per SKILL.md, do not ask the model to scan directories itself — the script handles everything.
What are this skill's strengths and limitations?
- Zero dependencies: pure Bash + Python 3, output is a single self-contained HTML file
- Scope-aware: distinguishes User / Project / Local layers with dynamic badge-count filtering
- All memory cleanup suggestions use reversible archive commands (moving to ~/.claude-archive/) — nothing is deleted
- Decodes folder ownership from the `cwd` field in session .l files, so paths with real dashes or unusual characters resolve correctly
- Privacy-conservative: file contents are only hashed for duplicate detection, never rendered
- Uses macOS's `open` command — works out of the box on macOS, needs a manual `xdg-open` swap on Linux; no Windows guidance in the source
- Orphan/stale detection depends on .l session files existing and a 30-day threshold, which won't cover every edge case
- No test suite in the repo; no version history or maintenance commitment stated in the README
- Read-only inspection plus advisory archive commands — it never performs actual cleanup, so you must copy and run commands yourself
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| cc-inspect Ecosystem Inspector this page | 49 · Use with care | ★ 17 | 5mo ago | — |
| Claude Automation Advisor ✓ Anthropic · Official | 50 · Use with care | ★ 38k | 11d ago | Apache-2.0 |
| cmux Diagnostics | 51 · Use with care | ★ 28k | 11d ago | NOASSERTION |
| Engram Persistent Memory Protocol | 68 · Recommended | ★ 7.2k | 11d ago | MIT |
| Engram Architecture Guardrails | 50 · Use with care | ★ 7.2k | 11d ago | MIT |
How did FollowSkills review this skill?
Script is primarily read-only scanning, outputs self-contained HTML to /tmp, and archival uses reversible mv into ~/.claude-archive; destructive commands require the user to copy and run them; data-flow disclosure is decent (memory files only hashed, never rendered). Deductions: dedupe generates rm commands, full ~/.claude/ contents (including MCP/hook command lines that may contain tokens) are scanned and displayed without explicit confirmation, misjudgment risk of orphan detection (sampled l) is under-disclosed, publisher identity unverified.
Script is well-structured with set -euo pipefail and embedded Python; the main path (scan, generate HTML, open) is statically self-consistent. Deductions: pervasive 2>/dev/null || true swallows errors so failures are silent; python -c string interpolation breaks on paths with quotes/special characters; hardcoded macOS open; no tests or failure feedback.
Clear audience (Claude Code users auditing their local setup), bilingual English/Traditional-Chinese docs, explicit scan-scope tables, complete install/usage instructions. Deductions: macOS-only (Linux requires manual xdg-open edit), broad trigger description, no declared non-fit boundaries (Windows, remote environments).
MIT license, bilingual README, documented scan scope and memory-hygiene grouping, good layering with screenshots. Deductions: no version number or changelog, no maintenance-ownership or update-path statement; SKILL.md references ~/.claude/skills/cc-inspector which mismatches the repo name (hidden assumption); no FAQ or known-limitations list.
Concrete value proposition (three-scope dashboard plus reversible memory cleanup suggestions), better than manual directory browsing, zero-dependency and lightweight. Deductions: static review cannot confirm output is directly usable; screenshots and claimed behavior are unverified; the dedupe keep-newest assumption may not always match user intent.
README provides screenshots and detailed mechanism descriptions; script source is auditable; English and Chinese docs are consistent and cross-checkable. Deductions: no test suite, no CI evidence, no third-party execution corroboration — all runtime behavior is author-claimed and unreproduced in this static review.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →