gh-fix-ci Skill (OpenAI) for Claude Code: Fix Failing GitHub Checks
OpenAI's skill that makes your agent find failing GitHub Actions checks on a pull request, read the logs and propose a fix before changing anything.
- Installs
- 11.6K
- on skills.sh
- Stars
- 28.4K
- on GitHub
- Files
- 5
- besides SKILL.md
- Updated
- Oct 2026
- last repo commit
gh-fix-ci handles one job: a pull request whose GitHub Actions checks fail. The agent finds the pull request, lists the failing checks, pulls their logs and shows you the failing snippet. It then drafts a fix plan and waits for your approval before it changes code.
It needs the GitHub CLI (gh) signed in, with repo and workflow scopes, and Python to run the bundled inspect_pr_checks.py. Checks from other CI providers, such as Buildkite, are out of scope: it only reports their link.
Good for
- Developers whose pull request has a red GitHub Actions check
- Teams that want a written plan and approval before an agent edits code to fix CI
- Anyone who already uses the GitHub CLI (gh) from the terminal
Install gh-fix-ci
Claude Code
- 1
Download the skill folder from https://github.com/openai/skills/tree/HEAD/skills/.curated/gh-fix-ci.
- 2
Put it at ~/.claude/skills/gh-fix-ci/ so that SKILL.md sits directly inside.
- 3
Start a new session; the skill loads when its description matches the task.
Requires GitHub CLI (gh), Python.
The skill runs a Python script and the GitHub CLI on your machine, so a Claude app upload does not fit it. Sign in first with gh auth login.
Built from the package data we hold. Not yet tested by us.
What's in the skill
When it loads
Use when a user asks to debug or fix failing GitHub PR checks that run in GitHub Actions; use `gh` to inspect checks and logs, summarize failure context, draft a fix plan, and implement only after explicit approval. Treat external providers (for example Buildkite) as out of scope and report only the details URL.
Sections in SKILL.md
- Overview
- Inputs
- Quick start
- Workflow
- Bundled Resources
- scripts/inspect_pr_checks.py
Key facts
- Author
- OpenAI (openai/skills)
- Installs on skills.sh
- 11,634 as of 2026-11-02
- Last repo commit
- 2026-10-28
- What is inside
- 6 files including SKILL.md: 1 Python script, LICENSE.txt, agents/openai.yaml and 2 icon files
What it includes
Runs gh auth status and asks you to run gh auth login if needed.
It stops until you are signed in with repo and workflow scopes.
Runs code ·
Check gh sign-inUses the PR you name, or the current branch's PR.
It runs gh pr view to get the number and URL.
Runs code ·
Resolve the pull requestLists failing checks, fetches Actions logs and extracts a failure snippet.
The script calls the gh command for pr view, pr checks, run view and api, and git to find the repo root. It exits non-zero while failures remain, and --json gives machine-readable output.
Runs code ·
scripts/inspect_pr_checks.pyRuns the gh commands by hand when the script cannot.
gh pr checks, gh run view with --log, and gh api for job logs when a run is still in progress.
Runs code ·
Manual fallbackReports the link for non-GitHub-Actions checks and does nothing else.
Buildkite and other providers are labeled external.
Read ·
External checksDrafts a plan, waits for approval, then applies it.
It uses a create-plan skill if you have one, otherwise a short inline plan. After approval it applies the plan, summarizes diffs and tests, and asks about opening a PR.
Write ·
Plan, approve, implementSuggests re-running tests and gh pr checks afterwards.
This confirms whether the checks now pass.
Runs code ·
Recheck
From the official docs
Permissions
Low risk
The agent runs the GitHub CLI with your sign-in and can edit code after you approve a plan. Check the plan, and the diff, before you say yes.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only mode
Watch out for
It uses your GitHub sign-in
Everything goes through gh as you. Use a sign-in with only the access you want the agent to have. github.com
It edits code once you approve
Approval covers the plan. Read the diff and run your tests, which the skill also suggests. github.com
Logs enter the conversation
Failing job logs are shown to your AI provider. Logs can contain secrets that were printed by mistake, so check them if your repo is private. github.com
Technical details
| Turning on read-only | No switch. SKILL.md has the agent implement changes only after explicit approval of its plan. |
|---|---|
| Talks to | GitHub, through the gh CLI (pr view, pr checks, run view, api) |
| Sign-in and keys | Uses whatever sign-in the gh CLI already has. SKILL.md says repo and workflow scopes are typically required. |
| Scopes |
|
Questions
What does gh-fix-ci do?
It helps with failing GitHub Actions checks on a pull request. The agent reads the failing logs, summarizes the failure, proposes a fix plan and implements it only after you approve.
How do I install it?
Use the install box at the top of this page. It needs the GitHub CLI and Python on your machine, so there is no Claude app upload. The openai/skills README shows the Codex installer but says the repository is deprecated.
What do I need before using it?
The GitHub CLI (gh) signed in with repo and workflow scopes, plus Python for the bundled script. SKILL.md has the agent check gh auth status first.
Does it work with Buildkite or other CI?
No. SKILL.md treats other providers as out of scope and only reports the details URL.
Will it change my code without asking?
No. SKILL.md has it draft a plan and implement only after your explicit approval.
Is it safe to use?
It runs gh and git commands on your machine with your GitHub sign-in, and edits files once you approve a plan. The bundled script calls gh for pr view, pr checks, run view and api, and git rev-parse. Review the plan before you approve it.
Information
updated Nov 2, 2026
- Repository
- openai/skills
- Folder
skills/.curated/gh-fix-ci
Sources
Docs, issue trackers and discussions this page is built from.