ctdio/dotfiles
Overview
This skill automates iterative handling of bugbot feedback and CI failures on pull requests. It fetches unresolved bugbot comments, triages each issue, makes minimal surgical fixes or documents and resolves verified false positives, and repeats until all comments are resolved and CI passes. It is designed to run inside a ralph-loop orchestration for repeated passes.
How this skill works
The skill waits for bugbot to finish analyzing, then queries for unresolved bugbot review threads on the current PR. For each thread it triages whether the issue is a real bug, a false positive (with evidence), or needs deeper investigation, applies minimal fixes or posts a documented resolution, commits changes, and submits them via Graphite. After submission it waits for bugbot and CI to re-run, fetches failing check logs if any, runs targeted local fixes for lint/format/type/test/build errors, and repeats until success.
When to use it
- Address unresolved bugbot review comments on a PR
- Fix CI failures (lint, format, type checks, tests, build)
- Automate iterative triage and fixes as part of ralph-loop
- Quickly resolve false positives with evidence and close threads
- Perform surgical fixes while preserving existing code patterns
Best practices
- Run wait-for-bugbot first to avoid racing the analyzer
- Triage carefully: prove false positives before resolving them
- Make minimal, pattern-consistent edits; avoid refactors
- Run failing checks locally (lint, format, tests) before committing
- Always git add, git commit, then gt submit --update-only — never git push/rebase/pull
Example use cases
- Fetch all unresolved bugbot threads on a feature PR and address each one
- Repair lint and formatter failures reported by CI, commit, and resubmit
- Document and resolve a false positive with code references and a concise explanation
- Fix a failing unit test discovered by CI, commit change, and re-run checks
- Run as part of ralph-loop to iterate until no bugbot comments remain and CI is green
FAQ
Evidence must show the flagged issue has no runtime impact: point to defensive checks, type guarantees, validation code, or framework behavior that makes the warning inapplicable. Quote exact lines and explain the data flow that proves safety.
What if I can’t fix a CI failure?
Document the failure in the progress report, include failing logs and why it’s unfixable in this iteration, and let a human continue. Continue fixing other issues you can resolve.
4 skills
This skill helps you address bugbot feedback and CI failures on PRs by triaging comments, applying surgical fixes, and ensuring checks pass.
This skill deepens an existing feature plan by critiquing gaps, verifying assumptions against code, and surfacing concrete improvements before implementation.
This skill lets you consult the oracle for a second opinion on problems by sending prompts to GPT-5.2 via Codex CLI.
This skill creates structured, phased feature plans in ~/.ai/plans, enabling progressive disclosure, stackable PRs, and quick context for future agents.