catch-me

Entry point for the catch-me-if-you-can onboarding flow. The user wants to rapidly pass as — and think like — a practitioner in some domain, so they can direct Claude better on this specific project.

What happens when this skill fires

1. The role-onboarder subagent (bundled at ${CLAUDE_PLUGIN_ROOT}/agents/role-onboarder.md) takes over in a forked context.
2. It asks up to three short questions (role + flavor, depth, project context) — skipping any already supplied in the invocation args.
3. It generates 6 role artifacts + meta.json inline, following ${CLAUDE_PLUGIN_ROOT}/skills/catch-me/templates/role-generation-prompt.md.
4. It pipes the output through ${CLAUDE_PLUGIN_ROOT}/skills/catch-me/scripts/write_role_files.sh <slug>, which writes atomically to the user's project at $(pwd)/.claude/catch-me/roles/<slug>/ and updates $(pwd)/.claude/catch-me/active-role.json.
5. It returns a compact cover-story summary (≈15 lines) to the user.

Invocation patterns

• /catch-me — ask everything from scratch.
• /catch-me <role> — role given, ask depth + project context.
• /catch-me "<free-text description>" — parse what you can from the description, ask the rest.

Notes

• User input may be in any language, but generated artifacts are always English (v1 scope).
• All role state lives inside the user's current project at .claude/catch-me/ — nothing is written outside the project. Different projects get independent role state.
• If the requested role already exists under .claude/catch-me/roles/, the subagent offers resume / regenerate / replace instead of blindly overwriting.
• After activation, the passive active-role skill auto-attaches the persona on role-relevant prompts — no further action needed from the user.