Guide · Agent skill
How to use the PR Labeling skill for skills CLI pull requests
Updated · written with AI from the project's own documentation and checked against our catalogue data
The pr-labeling skill is an instruction file for an AI agent that triages pull requests in the vercel-labs skills CLI repository. It tells the agent how to read a pull request, decide whether it is a documentation, bug, or enhancement change, and apply exactly one label. This guide explains what the skill does, how to install it, how to trigger it and what to check before relying on it.
PR Labeling for skills CLIvercel-labs/skills · MIT Download the skill Quick start
- 1Download the ZIP. Get the pr-labeling ZIP from the page where you found this skill.
- 2Unzip the skill. Extract it into ~/.claude/skills/ for personal use, or into .claude/skills/ inside a project.
- 3Or upload in the Claude apps. Open Settings, go to Capabilities, then Skills, and upload the ZIP.
- 4Check the tools. Confirm your automation has read_pull_request and add_label tools with GitHub access to the repository.
- 5Send a trigger request. Ask the agent to label a specific pull request by its number.
- 6Check the result. Confirm the chosen label on the pull request, and review mixed changes yourself.
What the pr-labeling skill makes the agent do
An agent skill is a set of written instructions that an AI agent loads so it knows how to handle a specific task. This one covers first-pass labeling of pull requests in the vercel-labs skills CLI repository.
The agent reads the pull request's title, description, current labels and changed file patches. It then chooses one label from three options: documentation, bug or enhancement.
- documentation: changes only documentation, examples or explanatory text, without changing CLI behavior.
- bug: corrects existing CLI behavior, including installation, update checks, source parsing, agent detection or security checks.
- enhancement: adds or intentionally expands CLI behavior, commands, agent support or integration support.
When the agent skips or stops without labeling
The skill is deliberately cautious. If the pull request already has bug, documentation or enhancement, the agent stops, so labels set by maintainers or earlier runs are kept.
If the patch is truncated, mixed or too thin to classify confidently, the agent adds no label. It does not infer a label from the title alone. If the head commit no longer matches the commit that triggered the run, it also stops so a newer run can handle the pull request.
Install the skill (ZIP download or Claude apps)
Download the ZIP file from the page where you found this skill. Unzip it into your personal skills folder at ~/.claude/skills/, or into a project's .claude/skills/ folder if you want the skill limited to one project.
If you use the Claude apps instead, open Settings, go to Capabilities, then Skills, and upload the ZIP there.
The package contains a single file, SKILL.md, which holds the instructions the agent follows.
- Personal install: ~/.claude/skills/
- Project install: .claude/skills/ inside the project folder
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
How to trigger the skill with a natural request
The skill expects a trigger message that includes a pull request number. The agent uses that number to call read_pull_request, so the number is the one thing your request must contain.
The skill also needs an automation environment that provides a read_pull_request tool and an add_label tool with GitHub access to the repository. Without those tools, the agent cannot read or label the pull request.
- Example request: "Label skills CLI pull request 123."
- Example request: "Classify PR 456 using the pr-labeling skill."
Requirements and safety notes
The skill needs an automation environment with GitHub access to the repository, plus the read_pull_request and add_label tools. It does not run scripts. It does use the network, because it reads from and writes to GitHub through those tools.
The agent applies labels without a human review step. The source notes that it may misclassify mixed changes, so a maintainer should check labels where priority or ownership matters.
- Runs scripts: no
- Uses network: yes
- Adds labels directly, with no review step in the skill itself
Tips for getting reliable labels
Give the agent a clean, current pull request. The skill checks that the head commit still matches the triggered commit, so push-and-trigger races are handled by stopping rather than labeling stale code.
Expect some pull requests to stay unlabeled. Truncated or mixed patches are left alone on purpose, so a human should label those.
- Trigger the skill once per pull request, because it adds one label per run.
- Treat existing maintainer labels as final; the skill will not change them.
- Review mixed pull requests yourself rather than relying on the rubric.
What people use it for
- Giving a skills CLI repository a consistent first-pass label on each incoming pull request.
- Using the labeling rubric as a starting point for an automated triage bot.
- Skipping pull requests that a maintainer has already labeled.
- Flagging unclear or incomplete patches for manual review by leaving them unlabeled.
Frequently asked questions
Is the pr-labeling skill free to use?
The source material for this skill does not state a price or licence terms on this page. The upstream repository is listed under the MIT licence, which you should read directly in the source repository before reuse.
What do I need to install the pr-labeling skill?
You need the SKILL.md file packaged as a ZIP, placed in ~/.claude/skills/ or a project's .claude/skills/ folder, or uploaded under Settings → Capabilities → Skills in the Claude apps. You also need an automation environment with read_pull_request and add_label tools that have GitHub access.
Can I use pr-labeling on repositories other than skills CLI?
The skill is not suited to other repositories. Its label rules and file paths are specific to the vercel-labs skills CLI project, so you would need to rewrite the rubric for your own codebase.
Does the skill run scripts or use the internet?
It does not run scripts. It does use the network, because it reads and labels pull requests on GitHub through the provided tools.
Why does the agent sometimes add no label?
By design, it stops without labeling when the patch is truncated, mixed or insufficient, and when the head commit no longer matches the triggered commit. It also will not guess from the title alone.
How does this compare with a general-purpose labeling bot?
This skill is a narrow, written rubric with three labels and explicit stop conditions. A general bot may offer more labels and rules, but this skill is built around the skills CLI repository's specific categories.
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 pr-labeling if you maintain the vercel-labs skills CLI repository and want consistent first-pass labels, or if you are building a triage bot and need a concrete rubric to adapt. Pick something else if you work on a different repository, or if priority and ownership decisions must be made by a person without automated review.
Download the skill