yaas-quest-dispatch

Read the quest's context.md first, then select the exact watch entries named in the dispatch prompt and re-query their sources. Act per the quest's decision rules, append follow-up watches for anything you touched, log every action with log-event.py, and ack each item before exiting.

Query the thread with slack_read_thread using the watch's watermark truncated to 6 decimals, then send through the slack-send.py surface helper so the message body is logged automatically. Append a slack_thread watch with last_checked_ts set to your reply's response_ts.

No. The triage orchestrator is the sole owner of watermark state and advances last_checked_ts itself. You may only read watch.json or append new entries via the add-watch helper; editing existing entries corrupts the termination-safety guarantee.

Slack returns zero messages when the watermark timestamp has more than 6 decimal places. Truncate last_checked_ts to 6 decimals before using it as oldest, and if the read still comes back empty, ack blocked rather than nothing_to_do so the watermark is not burned.

Queue an action when meta.json sets allow_send to false, when a watch reason says DRAFT ONLY, or when you judge the action needs user review, such as first contact with an external party. Write it with sq approval write, which also arms the approval watch.

Yes. GitHub PR and issue watches are handled with gh commands, and Jira watches use the packaged jira-call.sh REST bridge since the interactive Atlassian MCP is unavailable in headless dispatch. Out-of-scope changes should be ignored without logging.