Capabilities

Core Rule

When asked what the kit offers, enumerate the actually installed capabilities from the filesystem — never recite a memorized list, it drifts as the kit evolves. Read what is present under .claude/ and the project root, group it, and present one scannable briefing.

When to Use

When you (or the user) need a fast map of what this kit-enabled project can do:

• Starting a session in a project that has the kit installed — "what do I have here?"
• The user asks "what can this kit do / which skills / agents / hooks are available / what's enabled".
• Onboarding a teammate, or a fresh session, to the project's setup.

Do NOT use for:

• Installation health, orphan hooks, or unfilled placeholders → run scripts/doctor.sh.
• Listing skills from a terminal, outside a session → npx @tansuasici/claude-code-kit skills.
• A portable capability file for other agents/tools → npx @tansuasici/claude-code-kit generate agents-md.

Process

Harvest live, then summarize. Don't dump full file bodies — names + one-liners only.

1. Skills. Enumerate .claude/skills/*/SKILL.md (skip _-prefixed infra dirs like _shared, _templates). For each, read name, the first sentence of description, and whether it is user-invocable: true. Include the wiki module (wiki-module/.claude/skills/*) when WIKI.md is present.

   for f in .claude/skills/*/SKILL.md; do
     name=$(awk '/^name:/{sub(/^name:[[:space:]]*/,""); print; exit}' "$f")
     desc=$(awk '/^description:/{sub(/^description:[[:space:]]*/,""); print; exit}' "$f" | sed 's/\. .*/./')
     inv=$(grep -q '^user-invocable:[[:space:]]*true' "$f" && echo "/" || echo "·")
     printf '%s %s — %s\n' "$inv" "$name" "$desc"
   done

2. Agents. List .claude/agents/*.md; read each name and its one-line role.

3. Active hooks. Read .claude/settings.json → hooks; report the wired events and the scripts under each (what runs automatically). Flag any hook file in .claude/hooks/ that is present but not wired — those are opt-in (see agent_docs/hooks.md → Hook Profiles).

4. Modules & overlay. Detect by marker file and report active vs available:

   for m in "WIKI.md:Knowledge wiki" "ARTIFACTS.md:HTML artifacts" \
            "docs/ARCHITECTURE.md:Harness docs" "DESIGN.md:Design system" \
            "CLAUDE.project.md:Project overlay"; do
     f="${m%%:*}"; label="${m#*:}"
     [ -e "$f" ] && echo "✓ $label" || echo "· $label (not enabled)"
   done

5. Version. Read VERSION (or .claude-plugin/plugin.json) for the installed kit version.

6. Present the briefing (see Output Format).

Output Format

A single scannable briefing:

• Kit — version + which optional modules are on.
• Skills — grouped by purpose (audit · review · workflow · docs · meta · wiki), each as /name — one-liner. Mark agent-facing (non-invocable) ones distinctly.
• Agents — name — role.
• Hooks — by event (what runs automatically), plus any opt-in hooks available but not wired.
• Modules & overlay — active vs available.
• Next steps — 1–2 concrete suggestions for this project (e.g. fill CODEBASE_MAP.md, then /office-hours to scope or /feature-cycle to build from a spec).

Notes

• Read from disk every time → never stale; new skills/agents/hooks show up automatically.
• Hooks run deterministically in the background. This lists them for awareness; it does not enable/disable them — edit .claude/settings.json (see agent_docs/hooks.md).
• Complements, not replaces, the Claude Code / slash menu and npx @tansuasici/claude-code-kit skills.
• Pairs with scripts/doctor.sh: capabilities = "what you have", doctor = "is it set up correctly".