/commands — Self-discovery of the toolkit

You are Claude. The user wants to see what personal skills they have installed and what each does.

This is the self-documentation command. Useful when:

• You've been away from the toolkit for weeks and forgot the names
• You just installed the toolkit on a new machine
• You're not sure which command to invoke for a task

Phase 1 — Scan installed skills

ls -dt ~/.claude/skills/*/SKILL.md 2>/dev/null

If nothing matches, output:

"No personal skills installed. Run:

curl -fsSL https://raw.githubusercontent.com/mauroepce/claude-workspace/main/bin/install-personal.sh | bash

to install Mauricio's toolkit."

If skills exist, proceed. Also check for leftover legacy command files (ls ~/.claude/commands/*.md 2>/dev/null) — if any match a skill name, flag them as stale duplicates the installer should have removed.

Phase 2 — Extract metadata

For each SKILL.md, read the frontmatter and get the description: field. The skill's name is its directory name. Note the file's modified date (the install date or last update).

Group skills by category:

If a skill doesn't fit any category, list it under "Other".

Phase 3 — Output

Present in this format (don't make this fancy — clarity over aesthetics):

PERSONAL SKILLS INSTALLED (~/.claude/skills/)

Task framework
  /work         — spec → generate → review → iterate framework
  /quick-work   — lite /work for small tasks
  /todo         — persistent task tracking (cross-session)
  /issue <N>    — pull a GitHub issue and structure context

Codebase understanding
  /conventions  — scan codebase for style/patterns (persists to .claude/conventions.md)
  /architecture — scan codebase structure (persists to .claude/architecture-map.md)
  /journeys     — Mermaid sequence diagrams of user flows
  /explain      — explain a system/bug/decision from the foundations up
  /onboard      — orchestrate conventions + architecture + journeys for day-one ramp-up

Debugging & decisions
  /debug        — hypothesis-driven debugging
  /decision     — capture a technical decision with confidence level

Code review & commit
  /code-review  — atomic quality review of staged/specified changes
  /safe-commit  — runs /code-review + /security-review + commit with confirm
  /safe-push    — full-branch review + push (refuses main without OK)

Scaffolding & environment
  /scaffold     — pick a template from ~/.claude/templates/ and insert it
  /isolate      — clean workspace for tests/interviews (prevents context bleed)

Meta
  /commands     — this command — list everything installed

Last update: <most recent file mtime>
Total: <count> skills installed

Phase 4 — Offer next step

End with:

"All skill source files live at ~/.claude/skills/<name>/SKILL.md. To see a skill's full prompt, just cat ~/.claude/skills/<name>/SKILL.md.

To update or reinstall: curl -fsSL https://raw.githubusercontent.com/mauroepce/claude-workspace/main/bin/install-personal.sh | bash"

What NOT to do

• Don't show stale skills. If a skill exists in ~/.claude/skills/ but is no longer in the installer's PERSONAL_SKILLS list, flag it: "⚠ /<name> is installed but not in the official toolkit anymore — consider running install-personal.sh to remove it."
• Don't fabricate descriptions. If the frontmatter is missing or malformed, show "(no description — file may be corrupt)" instead of inventing.
• Don't add explanation about how each skill works in detail — that's docs/FRAMEWORK.md. This is just the index.