Launch and monitor detached headless pipeline orchestrators for GitHub issues.
npx skills add https://github.com/jpmoya/claude-agents --skill orchestrate-jpmoya
Please help me install this Agent Skill. Skill: orchestrate Source: https://github.com/jpmoya/claude-agents/tree/main/skills/orchestrate Command: npx skills add https://github.com/jpmoya/claude-agents --skill orchestrate-jpmoya
claude -p process so it outlives the session, and provides status, restart, and stop controls. ## Core Features & Use Cases - Detached headless launch: Starts the orchestrator via orchestrate.sh with nohup, enforcing one orchestrator per issue and a max of 3 concurrent runs with a memory floor; excess launches are queued automatically. - Supervisor-driven recovery: A cron-ticked supervisor restarts crashed orchestrators with backoff, drains the queue, manages agent-go/agent-in-progress labels, reconciles orphaned labels, and dispatches shared work across machines with a claim protocol. - Status and control: status, tail, stop, and queue subcommands report run state (running, restarting, held, done) from local pid files and issue markers, plus a heartbeat push to a shared status board. - Use Case: Ask to "run the pipeline on issue 547" and the Skill launches a detached orchestrator for that repo and issue, reports the PID and log path, and lets you check progress later without keeping the session alive. ## Quick Start Ask the assistant to orchestrate a specific GitHub issue by number, for example "orchestrate #547 in the scheduler repo", and it will launch the detached pipeline run and report its status.High-intent search queries and answers about installing and using this skill.
Run orchestrate.sh with the repo path and issue number, for example `orchestrate.sh ~/dev/scheduler 547`. The script launches a detached headless claude process, prints the PID and log path, and queues the run automatically if all concurrency slots are full.
Run `orchestrate.sh status` to list every recorded orchestrator with its state and latest issue marker, or `orchestrate.sh status 547` for one issue. Use `orchestrate.sh tail 547 40` to view the last 40 lines of its log.
The supervisor cron tick detects the dead process, checks the issue's latest marker, and requeues it with exponential backoff. After 3 restarts without marker progress or 6 total, it posts a note on the issue and parks the run as held for manual intervention.
No. Launching sets the agent-in-progress label, and the launcher refuses issues carrying that label that it does not own locally. Shared dispatch uses a claim comment with a 15-second settle window so only the earliest claiming machine launches the issue.
Subagents run inside the parent session and die when it compacts, restarts, or exits, killing the pipeline mid-run. The orchestrate.sh script instead starts the orchestrator as the main agent of its own detached claude -p process, and a hook blocks Agent-tool launches.
Run `orchestrate.sh stop <issue>`, which kills the orchestrator process and its claude child, then writes a tombstone file. The supervisor sees the tombstone and will not auto-restart that issue until you relaunch it manually.
OutlineDrivenApply configurable label-based state machines to triage GitHub issues.
astralform-aiDrives an open GitHub pull request to merge by classifying review findings and fixing blockers.
Miosa-osaCoordinate agent wake-up cycles and health monitoring via heartbeat protocol.
anxiety135790Delegate coding, refactoring, and code review tasks to OpenCode CLI.