toolkit-dashboard

Toolkit Dashboard

Generate a single, self-contained dashboard/index.html that gives a clear visual overview of the whole toolkit: every skill and agent (name, category, version, description, docs), plus a history of skill/agent runs and their findings. It runs on localhost with one command and has zero runtime dependencies.

When this runs

• First thing when someone opens the toolkit and wants to see what's inside.
• Whenever a skill or agent is added, updated, or run — regenerate so the dashboard stays current.
• When the user asks for an overview, a dashboard, run history, or "what do I have".

What it does

1. Scans skills/*/SKILL.md — parses YAML frontmatter (name, description, optional category, version), and notes whether each skill has a USAGE.md and a references/ folder.
2. Scans agents/*.md (excluding *.USAGE.md) — parses frontmatter (name, description, tools, optional category, version).
3. Merges catalog-meta.json (optional, at the toolkit root) — fills in category/version/tags for any item that doesn't declare them in frontmatter.
4. Loads runs/*.json — structured run records (see the schema below) and renders them as a searchable, filterable history with verdict badges and findings.
5. Writes dashboard/index.html — one file, all data embedded, no external requests. Works over file:// or a local server.

How Claude runs it (no terminal for the user)

When this skill triggers, Claude builds and opens the dashboard for the user — they should never have to run Python themselves.

1. Build it with the Bash tool. The generator auto-detects the toolkit root, so no arguments are needed:

   python3 "<toolkit>/skills/toolkit-dashboard/scripts/build_dashboard.py"
   

   (--root <toolkit> also works if you want to be explicit.)
2. Open it for the user:
   • Cowork: present dashboard/index.html with the file-presentation tool so they open it in one click.
   • Claude Code / terminal: give them the path and offer to serve it — cd dashboard && python3 -m http.server 8080 → http://localhost:8080.
3. (Optional) Also build the case-study showcase:

   python3 "<toolkit>/skills/toolkit-dashboard/scripts/build_showcase.py" --target "Project Atlas"
   

   → showcase/index.html (a branded, screenshot-ready results page).