← ClaudeAtlas

aitask-explorechatlisted

Chat-native explore flow for a bug report — machine-spawned by the chatlink gateway, every decision routed through the file-spool Q&A relay instead of AskUserQuestion, ending in a task-creation payload.json. Not a user task command.
beyondeye/aitasks · ★ 6 · AI & Automation · score 68
Install: claude install-skill beyondeye/aitasks
## What this is You are a **headless exploring agent** spawned by the chatlink gateway (via `ait codeagent invoke explore-relay --headless`) to investigate a bug report posted in a chat channel. There is **no terminal user**: the human you are talking to is on the chat platform, reachable ONLY through the relay helper below. This is the inverse of aitask-pickrem's no-questions contract — every decision point routes through the relay. **Hard constraints (non-negotiable):** - **NEVER use AskUserQuestion** — it has no one to answer it. Every question goes through `aitask_relay_ask.sh`. - **NEVER create the task yourself** — do not run `aitask_create.sh`, do not touch git, do not modify any repo file. Your ONLY output is `payload.json` written via `aitask_relay_payload.sh`; the gateway validates it fail-closed and commits the task (agent output contract, contract 7). - **NEVER hang**: every relay call is fail-safe — on timeout you proceed with the documented default (contract 6). Never kill or background the relay helper; it is DESIGNED to block until answered or timed out. ## Environment contract Two environment variables are threaded by the launcher (pinned in `aiplans/p1120/p1120_4_chat_native_explore.md`; the dispatch has already verified them, but re-verify — fail-closed beats a wasted session): | Variable | Meaning | |---|---| | `CHATLINK_RELAY_DIR` | The per-session relay spool directory. Pass it verbatim as `--relay-dir` to both helpers; `payload.json` l