gh-fix-ci

The gh-fix-ci skill provides an automated workflow for developers to diagnose and resolve failures in GitHub Actions CI pipelines directly from the terminal. By integrating with the GitHub CLI (gh), this agent systematically inspects pull request checks, identifies failing jobs, and extracts relevant log snippets for root cause analysis. It streamlines the debugging process by summarizing the error context and providing a actionable plan for code adjustments, which can then be implemented once the user grants approval.

This tool is specifically designed for engineers working in repository-heavy environments who need to maintain high velocity in their PR lifecycle. It helps bridge the gap between noticing a pipeline failure and understanding the underlying code issue, allowing for rapid iteration without needing to switch contexts or manually parse large CI logs.

• Automatically lists failing PR checks and retrieves status information for GitHub Actions workflows.

• Fetches and parses stdout/stderr from individual job logs to extract essential failure markers like exceptions, stack traces, and test timeouts.

• Generates concise, developer-focused failure summaries that highlight the most critical log segments.

• Facilitates a structured fix lifecycle by proposing a remediation plan before applying any modifications to the codebase.

• Leverages the plan skill for drafting and verifying changes, ensuring that all fixes are deliberate and reviewed.

• The skill requires a properly authenticated GitHub CLI environment with escalated repository and workflow scopes.

• Input parameters include the optional repository path and PR number or URL; it defaults to the current branch PR when no specific target is provided.

• It strictly focuses on GitHub Actions; external services like Buildkite are detected and reported solely via their URL references.

• Use the provided inspect_pr_checks.py script directly or through the agent for machine-readable JSON output, which is useful for further analysis or integration with reporting tools.

• Always ensure you have sufficient permissions to read workflow runs; if sandboxed environments block access, utilize the require_escalated flag to provide the necessary authorization.