getting-started

If you were dispatched as a subagent to execute a specific task, skip this skill. Subagents work from the context given to them, not from the full knowledge base. Do NOT respond to the user's first message until you have called `briefing` and reviewed the result. Do NOT skip this because the user's request seems simple or urgent. A 3-second briefing prevents 30 minutes of contradicting past decisions.

Getting Started

You're entering a task and need context. Do this quickly and silently:

1. Call briefing to get the session briefing (open threads, recent decisions, last checkpoint)
2. If the task involves a specific domain or topic, invoke orchestrator:consult-concierge with a query like "What should I know about [topic]?" - the concierge searches conventions, decisions, anti-patterns, and architecture notes, then returns a curated summary. For simple lookups, call lookup directly instead.
3. For a complete picture of tracked work, call list_work_items (all statuses) and list_open_threads - these return everything without keyword search, so nothing is missed.
4. Scan the briefing for anything relevant to your current task
5. If the briefing shows a recovery checkpoint, honor it - that's where the last session left off

Do NOT dump the full briefing to the user. Internalize it and proceed with the task. Only mention relevant items (e.g., "I see there's an open thread about X that relates to this").