One command, one HTML file — see everything Claude Code knows, can do, and should improve.
Zero dependencies. Runs locally. All secrets auto-masked.
curl -fsSL https://raw.githubusercontent.com/pxx-design/claude-code-inspector/main/install.sh | bashOr run directly:
node path/to/generate-dashboard.jsBoth generate claude-code-inspector.html in your current directory. Open it in any browser.
10 capabilities in a Bento grid — instantly see what Claude can and can't do in your project:
- Core: read/write files, run terminal commands
- Knowledge: project rules (CLAUDE.md), cross-session memory
- Connect: web search, browser automation, team tools (Feishu/Slack/Notion)
- Extend: plugins, hooks, auto-commit
A 0–100 score with findings sorted by severity. Each finding is expandable with a one-line fix suggestion.
Collapsible sections for all 9 dimensions:
| Section | What it shows |
|---|---|
| Skills | Project and user-level skills with metadata |
| Plugins | Installed plugins with enabled/blocked status |
| MCP Servers | Configured servers, commands, env keys |
| Hooks | Event-based hooks with matchers |
| Agents | Expert roles from plugins, project, and user level |
| Commands | Slash commands with descriptions |
| Memory | CLAUDE.md, memory files with previews |
| Settings | Git config, permissions, environment variables |
- Claude brand palette (Pampas background, terra cotta accents)
- Serif display type, system sans-serif body
- WCAG AA compliant contrast ratios
All secrets (API keys, tokens, passwords) are automatically masked. The generated HTML is safe to share.
- Node.js >= 16
A single Node.js script (~2000 lines, zero dependencies) that:
- Scans
~/.claude/and.claude/for all config files - Detects 10 capabilities from MCP servers, plugins, permissions, and settings
- Runs 30+ health checks across 9 dimensions
- Generates a self-contained HTML dashboard
No data leaves your machine.
rm -rf ~/.claude/skills/inspector/MIT