Guide · Agent skill
How to use the Maintainer Orchestrator agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The maintainer-orchestrator skill teaches an AI coding agent how to manage a maintainer's queue when several independent issues, pull requests or repositories need handling at once. It is aimed at open-source maintainers and small teams running structured triage or batch sweeps. After this guide you will know how to install the skill, how to phrase requests so the agent uses it correctly, and what safety limits to expect.
Maintainer Orchestratorsteipete/agent-scripts · MIT Download the skill Quick start
- 1Download and unzip the skill. Download the ZIP from this page and unzip it so the maintainer-orchestrator folder is intact.
- 2Place it in the skills folder. Move the folder to ~/.claude/skills/ for all projects, or to .claude/skills/ inside one project. Alternatively, upload it under Settings → Capabilities → Skills in the Claude apps.
- 3Confirm prerequisites. Make sure Git is installed and the agent can reach the target repositories and GitHub issues, PRs and CI status.
- 4Phrase a batch request. Name the repositories and items, and state whether you want one-shot triage or ongoing maintenance.
- 5Review the written scope. Before work proceeds, check the explicit queue, repositories, and authorized public actions the agent records.
- 6Make owner decisions. Answer the items the agent marks as needing owner input, such as product, security or destructive-handling choices.
What the skill makes the agent do
The skill is a control-plane process for maintainer work. Before doing anything, the agent checks whether the request is really one item. If it names or implies a single issue, a single PR, or one coherent implementation, the agent skips orchestration and works directly in the current session, without creating extra helpers.
For genuine multi-item batches, the agent writes down an explicit scope listing the repositories, issues and PRs involved and which public actions are authorized. It assigns one worker per repository when two or more independent items are coordinated, protects any unsaved local work, and serializes public actions such as pushes, PR updates, merges and releases so they do not conflict.
- Single items stay direct: fix one issue, repair one contributor PR, or make one release.
- Batches get a written scope before any workers start.
- Local dirty or unmerged work is preserved, not stashed, reset or overwritten to begin.
- Public actions are queued one at a time per repository or branch.
Install the skill (ZIP download or Claude apps)
Download the skill ZIP from this page and unzip it. To make it available to Claude Code for your user account, place the unzipped folder in ~/.claude/skills/. To limit it to one project, place it in that project's .claude/skills/ folder instead.
If you use the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.
The skill folder contains SKILL.md, agents/openai.yaml and references/non-majority-repositories.md. Keep these together so the references file is found.
- User-wide: ~/.claude/skills/
- Single project: .claude/skills/ inside the project
- Claude apps: Settings → Capabilities → Skills
Check what you need before you start
The skill needs Git installed and access to the target repositories. It also relies on GitHub access to read issues, PRs and CI status. The source files do not specify how GitHub credentials should be provided, so confirm your agent environment can reach GitHub with the permissions you intend to grant.
Worker threads use the Codex app project-thread mechanism when it is available. This is optional: if it is not available, the agent falls back to the current session rather than simulating a worker hierarchy.
The skill does not run scripts, but it does use the network.
- Git installed, with access to target repositories
- GitHub access for issues, PRs and CI status
- Codex app project threads for workers (optional; falls back to current session)
Trigger it with natural requests
Say what the batch is and where it lives. The skill activates for requests that ask to handle multiple independent issues or PRs, coordinate several repositories, or run a queue, sweep, batch or ongoing triage. Name the repositories and items where you can, because the agent turns your wording into the explicit scope it writes down.
Do not invoke it for one item. A request such as fixing a single bug will be handled directly, and the agent will say that orchestration is unnecessary.
- Good: "Triage the open issues and PRs in repo-a and repo-b, and tell me what needs my decision."
- Good: "Sweep these five named PRs across two repositories and prepare them for review."
- Not needed: "Fix issue #42 in repo-a."
Understand scope, persistence and safety limits
Persistent monitoring is off by default. Heartbeats, recurring monitoring, automatic queue refill and broad owner scans are enabled only when you explicitly ask for ongoing or autonomous maintenance, a portfolio sweep or a maintained queue. Wording like that can switch them on, so read your request carefully.
The invocation authorizes only the scoped maintainer work and the public sequence you requested. It does not authorize releases, version bumps, tags, package publication, destructive handling of unique work, or unrelated external changes unless you request them separately.
For broad portfolio discovery, the skill excludes archived repositories and the list in references/non-majority-repositories.md unless you name them, and it does not refill the queue from openclaw/openclaw or openclaw/clawhub without a solicited request.
- Review the written scope before letting the agent proceed.
- Ongoing monitoring and queue refill only run when you ask for them.
- Releases, tags and package publication are never implied by a batch request.
What people use it for
- Triaging a backlog of open issues across several maintained repositories, sorting them into autonomous, needs-owner and not-planned groups.
- Reviewing and reconstructing a set of contributor PRs across repositories, preserving contributor credit where a rewrite is needed.
- Running a one-shot batch sweep of named PRs, with pushes and merges serialized per repository.
- Coordinating CI watching for several independent changes, with one watcher per run or head SHA.
Frequently asked questions
Is the maintainer orchestrator skill free?
The skill is published under the MIT licence, as stated on this page. Check the licence file in the repository for the full terms before reuse.
What do I need to use the maintainer-orchestrator skill?
You need Git installed and access to the target repositories, plus GitHub access for issues, PRs and CI status. Codex app project threads are optional for worker threads; without them the agent works in the current session.
Where do I install the skill?
Unzip the download into ~/.claude/skills/ for all projects, or into a project's .claude/skills/ folder for one project. In the Claude apps, you can upload it under Settings → Capabilities → Skills.
Does the skill run scripts?
No, according to the item details, the skill does not run scripts, but it does use the network to reach GitHub.
Will it work on a single issue or PR?
Not as an orchestrator. The skill is designed to skip orchestration for single items and continue directly, because coordinating one item adds overhead without value.
Can it make changes to my repositories unattended?
Only within the scope you authorize. Persistent monitoring and automatic queue refill are enabled only by explicit wording, so review the scope first if you do not want ongoing changes to public repositories.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you maintain many issues or PRs across several repositories and want an agent to triage them with written scope, protected local work and serialized public actions. Skip it for single items, where it adds overhead, and avoid it for unattended ongoing changes unless you are comfortable reviewing the scope first.
Download the skill