How to use the GitHub Project Triage agent skill

What the skill makes the agent do

When you type triage inside a GitHub project, the agent pulls the open issues and pull requests using the GitHub CLI (gh) and writes a maintainer-style review. Each item is shown with its GitHub URL first, followed by what it is about, why it matters, author trust, fit, risk, test or proof state, blockers and a suggested next action. The skill says never to return only queue numbers or opaque references.

The output is sorted into three groups: Autonomous candidates (items that look fixable without more product input), Needs owner decision (items blocked on a maintainer's call, missing access or security judgment) and Defer/close/supersede (stale, duplicate or overlapping items). The skill describes autonomous candidates as a selection for review, not permission to start work.

Install the github-project-triage skill

Download the ZIP for this skill from this page. You have two ways to add it to your agent.

Option one is to unzip it into a skills folder. For a personal install, unzip it into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. The skill folder contains SKILL.md, agents/openai.yaml and scripts/github-activity.sh.

Option two is to upload the ZIP in the Claude apps. Open Settings, go to Capabilities, then Skills, and upload the file there.

• Personal install: ~/.claude/skills/
• Project install: .claude/skills/ inside your repository
• Claude apps: Settings → Capabilities → Skills

Check the requirements before you trigger it

The skill needs git and the GitHub CLI (gh), and gh must be authenticated to GitHub. The skill runs scripts and uses the network, so both are required for it to work.

Optional extras are a repobar binary, or a local RepoBar SwiftPM checkout, for broad discovery across several repositories or owners. If the RepoBar fallback is used, a Swift toolchain is also needed. You do not need RepoBar for single-repository triage.

• git installed
• GitHub CLI (gh) installed and logged in to GitHub
• Optional: repobar binary or local RepoBar SwiftPM checkout for multi-repo discovery
• Optional: Swift toolchain if the RepoBar fallback is used

Trigger the skill with natural requests

The simplest trigger is typing triage. The skill is meant to run whenever the user types triage, unless the request clearly targets a non-GitHub domain. From inside a Git repository with a GitHub remote, it triages only that project by default.

To widen the scope, the skill's scope rule says to include words such as broad, all, everything, or to name several owners or organisations, or to ask for cross-repo triage. A request to work through safe items on your own should use wording like do work autonomously or keep going. Be aware that the skill's default owners for broad triage are steipete and openclaw, so broad requests will reach beyond your own repository.

• triage (inside a repo: reviews that project's open issues and PRs)
• broad triage (widens the scope across relevant owners)
• do work autonomously (processes eligible items one at a time; see safety notes)

Safety notes: scripts, network and autonomous work

The skill runs scripts and reads from GitHub over the network. Before the agent starts work in any local project, the skill requires the checkout to be on main, the pull with --ff-only to succeed and the worktree to be clean. If any of these fail, the agent is told to stop and ask rather than switch branches, stash, commit, reset or clean.

The skill also includes an autonomous work mode that can land code. Use it only with review and branch protection in place. The skill tells the agent to process items one at a time, to add proof to each pull request, and to ask first for new features, product choices, risky dependencies and changes that cannot be tested end to end.

The skill is written around one maintainer's workflow. It names a specific owner (Peter) whose comments override the agent's judgment, and it refers to RepoBar and the openclaw owners. Edit those references before using it in another team.

• Local repo gate: branch must be main, pull must succeed, worktree must be clean
• Autonomous mode: review and branch protection recommended
• Owner-specific wording must be edited for other teams

Tips for getting useful triage output

Add a VISION.md file to your repository if you have one. The skill says to read it first and use it as the source of truth for product fit before judging what can be handled autonomously. Without it, the agent falls back to its built-in autonomous-fit rules.

For large queues, the skill inspects only the top priority slice and says which items were not expanded. For small queues of about 10 open items or fewer, it inspects every item. Read the comments on each item yourself too, because the skill treats owner comments as authoritative.

• Keep a VISION.md in the repo so product-fit judgments are grounded
• Expect a detailed pass on small queues and a top-slice pass on large ones
• Check the URLs the skill prints before acting on any item