What problem does it solve? Operating a headless fleet supervisor for GitHub-based coding agents requires careful preflight checks: validating the .ratchet/herd.json config, avoiding duplicate supervisors, and knowing how to inspect live worker state. This Skill wraps those operations into a safe, repeatable procedure so nothing is started with a missing or invalid config. ## Core Features & Use Cases - Config validation without side effects: Runs node scripts/herd.mjs run --dry-run to validate .ratchet/herd.json and print the reconciled plan without spawning workers, surfacing the init hint when the config is missing. - Start or attach logic: Detects a running supervisor via pgrep -f "herd.mjs run", starts one in the foreground if none exists, or attaches to the running instance and summarizes live workers, attempts, and pending escalations from .ratchet/herd-state.json and .ratchet/herd-escalations.md. - Pinned worker dispatch contract: Defines the canonical rules for dispatched workers, including single-issue scope, branch resume behavior, and PR submission exclusively through node scripts/ratchet-submit.mjs. - Use Case: An operator wants to launch the agent fleet for the night. The Skill dry-runs the config, finds it valid, confirms no supervisor is running, and starts the poll loop — or, finding one already active, reports the three live workers and one pending escalation instead of launching a duplicate. ## Quick Start Ask the agent to validate the herd configuration and start the fleet supervisor, or summarize the currently running one.
