Guide · Agent skill
How to use the DeerFlow Maintainer Orchestrator agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The deerflow-maintainer-orchestrator skill gives an AI agent a defined role as a maintainer assistant for the bytedance/deer-flow project on GitHub. It finds the issues or pull requests you mean, reads their content and existing comments, and drafts or posts comment-only responses. This guide explains what the skill does, how to install it, how to ask for work in plain language, and what to check before you let it post under your account.
DeerFlow Maintainer Orchestratorbytedance/deer-flow · MIT Download the skill Quick start
- 1Check prerequisites. Install the GitHub CLI (gh) and authenticate it with an account that has access to bytedance/deer-flow.
- 2Download the skill ZIP. Get the skill ZIP from this page.
- 3Unzip it into the right folder. Unzip into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder.
- 4Or upload it in the Claude apps. In the Claude apps, open Settings, go to Capabilities, then Skills, and upload the ZIP.
- 5Make a scoped request. Ask for a specific issue, PR, URL, or a count with a time window in plain language.
- 6Choose draft or post mode. Say analysis only if you want drafts to review before anything is posted.
- 7Check the run report. Review posted comment URLs, skipped items and failures in the maintainer report.
What does the deerflow-maintainer-orchestrator skill make the agent do?
An agent skill is a folder with a SKILL.md file that tells an AI agent how to handle a specific kind of task. This one is for DeerFlow maintainers who want help triaging GitHub work. It is scoped to commenting: the agent resolves which issues or PRs you mean, inspects the evidence, and prepares or posts issue comments and PR review comments with risk notes and fix suggestions.
The skill explicitly avoids coding, branch management and release work. It is about analysis and communication on GitHub, not changing the codebase.
- Resolve issue and PR scopes with the GitHub CLI (gh)
- Analyze issues and classify their surface, such as Frontend UI, Backend API or Sandbox
- Post or draft issue comments and PR review comments
- Review batches of recent issues or PRs
- Compare competing PRs that target the same issue
- Give fix strategy, risk classification and validation guidance
gh issue view <number> --repo <repo> --json number,title,url,state,body,labels,author,commentsHow do you install the skill?
Download the skill ZIP from this page. You can install it in one of two ways. For a personal setup, unzip it into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder.
If you use the Claude apps instead, open Settings, go to Capabilities, then Skills, and upload the ZIP there. Either way, the skill's entry point is its SKILL.md file.
- Personal: unzip into ~/.claude/skills/
- Project only: unzip into the project's .claude/skills/ folder
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
How do you trigger the skill with natural requests?
You do not need a special command. Describe the scope in plain language and name the artifacts. The default repository is bytedance/deer-flow, so you only need to specify another repo if you mean one. You can also paste a full GitHub URL: a /issues/ link goes to the issue flow and a /pull/ link goes to the PR review flow.
For a bare number such as 123, the skill tries it as a PR first and then as an issue, so you do not need to say which type it is. A count or time window is respected. If you say recent hours without a number, the default window is six hours.
- Review PR #123
- Triage the latest 10 issues
- Draft comments for issues from the last 6 hours, without posting
- Compare the competing PRs that target issue #456
What does the skill need, and what are the safety limits?
The skill requires the GitHub CLI (gh) installed and authenticated with access to bytedance/deer-flow, plus network access to GitHub. It does not run scripts, but it does use the network.
The main risk is public posting. Once you give a bounded scope, the skill is instructed to post without asking follow-up questions: one public comment per selected non-skipped issue, and one PR review comment per selected PR that has high-confidence findings. If you want to see the text first, ask for analysis only and it will return drafts without posting. RFC issues are skipped by default, and existing comments suppress duplicate posting, so re-runs should not stack repeated comments.
The skill is written for maintainers and trusted local agents. It would post under the maintainer's account, so it is not suited to general contributors.
- Ask for drafts only when you want to review before anything is public
- Check that your gh login is the account you intend to comment as
- Use it only where unreviewed public replies would be acceptable
What tips get better results?
Give a clear scope every time: a list of numbers, a URL, or a count with a time window. Broad, vague requests get a practical recent slice, and the report lists anything left unprocessed. Use analysis-only mode when a comment would be sensitive. When several PRs target one issue, ask for a comparison so the skill gathers all candidates before reviewing.
The skill writes comments in the language of the issue or PR body, so Chinese text gets Chinese output and English text gets English output.
- Name the repo explicitly if it is not bytedance/deer-flow
- Ask for competing PR comparisons when several PRs fix the same issue
- Expect a compact report with posted URLs, skipped items, failures or drafts
What people use it for
- Triaging the newest issues in a busy week, with each one classified by surface and actionability
- Reviewing a PR and posting high-confidence review comments with risk notes and validation steps
- Comparing two or more PRs that target the same issue before choosing which to discuss further
- Drafting comments for a batch of issues to read before anything is posted publicly
- Re-running a review on a PR so that only new findings are added to an existing review thread
Frequently asked questions
Is the deerflow-maintainer-orchestrator skill free?
The skill is published under the MIT licence, as listed on this page. You still need a GitHub account with access to bytedance/deer-flow and the GitHub CLI (gh) installed and authenticated.
What do I need to run the deerflow-maintainer-orchestrator skill?
You need the GitHub CLI (gh) installed and authenticated with access to bytedance/deer-flow, plus network access to GitHub. The skill does not run scripts, but it does use the network.
Where do I install a Claude agent skill like this one?
Unzip the skill into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps, you can upload the ZIP under Settings, then Capabilities, then Skills.
Can the skill post comments without me reviewing them first?
Yes, if you give it a bounded scope and do not ask for analysis only. It is instructed to post one comment per selected non-skipped issue and one review comment per PR with high-confidence findings, without follow-up questions. Ask for drafts first if you want to review.
Can I use it on a repository other than DeerFlow?
The defaults are hardcoded to bytedance/deer-flow. Using it on another project would need the repository and workflow rules adapted first.
Are there alternatives to this skill, and how does it compare?
The source does not compare it with other tools. If you want a fully manual workflow, or a skill built for another project, review those options yourself. This skill is best when you want an agent to resolve scopes and draft comment-only responses for DeerFlow.
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 are a DeerFlow maintainer who wants help triaging batches of issues or PRs, or comparing competing PRs, and who is comfortable with comments being posted under your account. Pick something else if you are a general contributor, work on a project other than DeerFlow without adapting it, or cannot accept an unreviewed public reply.
Download the skill