gh-fix-ci_skill

Overview

This skill inspects failing GitHub PR checks using the gh CLI, fetches GitHub Actions logs, summarizes the failure context, then drafts a concrete fix plan and implements approved changes. It focuses on GitHub Actions runs; external CI providers are reported but marked out of scope. Authentication via gh with workflow/repo scopes is required before running the skill.

How this skill works

The skill runs a script that locates the PR (current branch or provided PR number/URL), lists checks, and filters for failing GitHub Actions runs. For each failing run it pulls run and job logs, extracts concise failure snippets, and assembles a summary. It then uses a plan workflow to propose changes; after explicit user approval it applies code changes, summarizes diffs and test steps, and suggests rechecking the PR checks.

When to use it

• You have a pull request with failing GitHub Actions checks and want an actionable diagnosis.
• You need a concise failure snippet and the likely root cause from CI logs.
• You want a concrete fix plan drafted and implemented after you approve it.
• You need machine-friendly output (JSON) for automation or further processing.
• You want to avoid handling non-GitHub CI providers like Buildkite; those should be only reported.

Best practices

• Ensure gh is authenticated with workflow and repo scopes (run gh auth login then gh auth status).
• Prefer running the bundled inspect_pr_checks.py script which handles field drift and log fallbacks.
• If sandboxed, rerun gh auth status with sandbox_permissions=require_escalated to allow network/keyring access.
• Do not attempt to modify or debug external CI providers; report their details URL and mark them out of scope.
• Require explicit user approval before applying any code changes or opening a PR.

Example use cases

• Diagnose a failing unit test that fails only in CI due to missing environment variables and propose a fix to mock or set the env in workflows.
• Fetch failing integration test logs, summarize the error stack, propose a targeted code or workflow change, and apply it after approval.
• Automate triage for a PR: collect failing job logs, produce a short remediation plan, and implement small fixes (lint, dependency pin, or test adjustment) when approved.
• Run as part of a developer workflow: python scripts/inspect_pr_checks.py --repo . --pr 123 --json to feed CI failure details into other automation.