Fleet coordination

Fleet coordination

Cross-machine staging coordination (see the repo's claude/agent-hierarchy.md, § Cross-machine staging coordination) runs over a dedicated coordination branch of this wiki repository — not over this rendered wiki, and not over a pinned issue.

• STATE.md on that branch: durable current truth (machines, standing arrangements, acting-coordinator records). Edited in place; never grows.
• LOG.md: append-only event stream (red-CI claims, contested-issue claim-checks, unusual-batch announcements). Claims are commits: a successful push acquires the claim; a rejected push means someone else got there first — pull, read, defer.
• Routine pushes to origin/main are NOT announced (git's fetch-merge-before-push + non-fast-forward rejection already serializes them safely).
• The branch is periodically deleted and recreated (STATE.md carried forward) so its churn never bloats this wiki's history. It is deliberately unrendered.
• Why the wiki repo and not the main hydra repo or an issue: main-repo events (commits, issue comments) are echoed to the community Discord — coordination churn there would spam human developers. Wiki changes are not echoed. Do NOT migrate this channel into the main repo or onto a GitHub issue.

Agents: fetch the branch before any push to main, before claiming a red-CI fix or a contested issue, and every ~10-15 minutes inside long watcher loops.