How to use the Address GitHub PR Comments agent skill

What the skill makes the agent do

An agent skill is a folder with a SKILL.md file that gives an AI agent step-by-step instructions for one kind of task. When a request matches the skill's description, the agent follows its workflow.

This skill follows four stages. First it inspects comments on the current branch's pull request. Then it categorizes the feedback and proposes a fix for each item. Next it applies the changes, reads the surrounding code, and runs the repository's relevant checks. Finally, when you authorize it, it replies to comments with the commit and test result that addressed them.

• Inspects conversation comments and inline review threads, including file and line anchors and resolution state.
• Asks you only when feedback conflicts, expands the scope, or needs a significant decision.
• Preserves unrelated edits in your code.
• Does not treat a conversation comment as closing an inline review thread.

Install the skill from the ZIP or the Claude apps

Option 1: download the ZIP from this page and unzip it into your personal skills folder, ~/.claude/skills/, or into a project's .claude/skills/ folder if you want the skill available only in that repository.

Option 2: in the Claude apps, open Settings, go to Capabilities, and upload the skill under Skills.

The skill's own files are a single SKILL.md. The folder you unzip should contain the address-github-comments skill folder with that file inside.

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

Set up GitHub CLI before you start

The skill uses the GitHub CLI (gh). Check that you are logged in before asking for any work. The skill itself tells the agent to run gh auth status first, and to run gh auth login if you are not logged in.

You also need a GitHub account that can access the repository and the pull request, and the repository's own build and test tooling so the agent can run the checks it needs.

• Install the GitHub CLI (gh) on your machine.
• Check login with gh auth status.
• If you are not logged in, run gh auth login.
• Confirm you have write access to the repository.

Trigger the skill with natural requests

You do not need a special command. Describe the job in plain language, mentioning the pull request and the kind of feedback. The skill's description covers review or issue comments on an open GitHub pull request using the gh CLI, so requests like these should match.

Be specific about what you want replies to do. If you want the agent to post replies or resolve threads, say so, because the skill asks for authorization before it posts or resolves anything.

• "Address the review comments on my open PR for this branch."
• "Go through the inline review threads on this PR, propose a fix for each, and wait for my approval before changing code."
• "Fix the feedback on PR 42 and reply to each comment with the commit SHA and the test result."
• "Look at the comments on the current branch's pull request and list which ones you think are unresolved."

What the workflow looks like in practice

The agent starts by fetching the conversation with gh pr view --comments. The skill notes that this flag shows conversation comments only and does not list every inline review thread. For inline comments, the agent uses the GitHub connector or the GraphQL reviewThreads connection with pagination, and records the repository, PR number, head SHA, thread IDs and requested changes.

After it proposes fixes, it applies them and runs the relevant repository checks. It reports a fix only after binding the result to the updated head commit. Replies follow the pattern in the skill: they cite the commit and the verification.

• Read the surrounding code of each comment before changing anything.
• Fix only the comments in scope, and keep unrelated edits intact.
• Reply with the commit SHA and the check that passed.
• Leave a proposed reply for you if resolving a thread is not authorized.

Safety notes: scripts, network and GitHub writes

This skill does not run scripts of its own, but it does use the network, because it talks to GitHub. Once you authorize it, it can write to GitHub under your account: posting replies and resolving review threads. Use it only on pull requests where you are allowed to make those changes.

Its risk level is marked critical in the skill file, so review the proposed replies before you grant permission to post. Do not use it on other people's PRs without their review, and do not use it on repositories where you have not authenticated gh or lack write access.

• Runs scripts: no. Uses network: yes.
• Writes to GitHub only after you authorize posting or resolving.
• Resolve threads only after the request is addressed and you have authorized resolution.
• Keep your gh credentials under your control.

Tips for better results

Give the agent the PR number or make sure the right branch is checked out, so it inspects the correct pull request. Ask it to list the comments and its planned fixes first if you want to review before any code changes.

Keep requests scoped. The skill is designed to address feedback that falls within your requested scope and to ask when feedback conflicts with it. Check the verification it reports, since the skill itself says it is not a substitute for testing or expert review.

• Start with a planning request before a fix-and-reply request.
• Name the PR number to avoid ambiguity.
• Read each proposed reply before authorizing posting.
• Make sure your repository's checks run locally so the evidence is meaningful.