async

# Async Use this skill when the user wants to: - start a Codex subagent task now and continue the main thread - optionally wait for the result when the task becomes blocking - manage subagent work through a user-facing `task_ref` instead of raw `agent_id` - get a strong completion announcement when async work finishes - query, collect, summarize, or redirect a previously launched async task Do not use this skill for: - generic background job systems outside Codex subagents - cross-session durable task storage or reliable wake-up guarantees - external queues, webhooks, cron, or service orchestration unless the runtime explicitly provides them ## Resource Map Read these references only when needed: - Runtime assumptions and fallback behavior: [references/runtime-contract.md](references/runtime-contract.md) - `task_ref` naming and lifecycle rules: [references/task-ref.md](references/task-ref.md) - Completion announcement policy and boxed alert templates: [references/alerts.md](references/alerts.md) ## Core Model Treat each async task as a tracked subagent lifecycle: 1. frame a bounded task 2. launch one subagent 3. assign a stable `task_ref` 4. keep the main thread moving unless waiting is required 5. announce completion strongly when the runtime makes it observable 6. collect, summarize, or inspect the result later Expose `task_ref` to the user. Keep `agent_id` internal unless debugging requires it. ## Defaults - Default execution mode is `launch-and-continue`. - I...