pm

Orchestrates autonomous agent fleets by triaging GitHub issues, PRs, and running workers.

Updated Sep 2, 2026
One-click install
npx skills add https://github.com/volter-ai/open-autonomy-compiler --skill pm-volter-ai
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: pm
Source: https://github.com/volter-ai/open-autonomy-compiler/tree/main/.claude/skills/pm
Command: npx skills add https://github.com/volter-ai/open-autonomy-compiler --skill pm-volter-ai

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Coordinating a fleet of autonomous coding agents across GitHub issues and pull requests requires constant situational awareness; without an orchestrator, work duplicates, failed PRs loop forever, and human-blocked items silently stall. ## Core Features & Use Cases - Full-board triage: Reads every open issue's full history, every open agent PR's checks and merge state, and every in-flight worker session before deciding one action per item. - Duplicate and loop prevention: Guards against launching a second develop run for an issue already in flight, enforces a rework cap via oa-rework marker comments, and cancels stuck or runaway runs through the Runner. - Dangling PR reaping and human escalation: Closes stale PRs whose issues are already closed, assigns and @mentions maintainers or requesters on blocked items, and re-pings them on a configurable SLA. - Use Case: On a scheduled sweep, the orchestrator finds an issue whose PR has green checks but a merge conflict, relaunches develop to rebuild it on fresh main, and escalates a 26-hour-old human-required item to the maintainer. ## Quick Start Ask the agent to sweep the repository board: review every open issue, PR, and running worker, then take the appropriate orchestration action on each.

Frequently Asked Questions about pm

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I orchestrate autonomous coding agents on GitHub issues?▼

Use an orchestrator sweep that reads every open issue's full history, every open PR's checks and merge state, and every in-flight worker via the Runner, then takes exactly one action per item such as launching develop, commenting status, or escalating to a human.

How do I prevent duplicate agent runs for the same GitHub issue?▼

Check the Runner's in-flight list for a develop session whose ref matches the issue number, including proposing states, and verify no PR exists on the agent/issue-<n> branch in any state before launching a new run.

What happens when an agent PR has green checks but a merge conflict?▼

A PR with mergeStateStatus DIRTY will never auto-merge even with all checks green. The orchestrator relaunches develop to rebuild the change onto fresh main, or closes the issue if the work is obsolete, while respecting the rework cap.

How does the rework cap stop failed agent runs from looping?▼

The orchestrator counts its own oa-rework marker comments on the issue against max_develop_attempts from autonomy.yml, defaulting to 2. At or above the cap it stops relaunching, labels the issue human-required, and escalates to a maintainer.

Can the orchestrator merge pull requests or edit code?▼

No. The permission split gives it no contents:write access, so it cannot merge or edit code. It can only comment, label, assign, launch or cancel runs, and close dangling PRs whose issues are already closed.

How are human-blocked issues escalated without spamming?▼

Blocked items get assigned to the maintainer or issue author with one @mention. If no human replies within policy.human.sla_minutes, default 24 hours, a single escalation re-ping is posted per SLA window, never every sweep.