Skip to content
Back to skills

Ci Debug

ASecurity

Diagnose a failing GitHub Actions check by reading the real logs and finding the first true failure.

  • 113 stars
  • 0 votes
  • 0 copies
  • 3 views
  • Added September 2, 2026
ai-agentsgit

Security analysis

A100/100

Scanned September 2, 2026

npx -y skills add SDSLeon/lightcode --skill ci-debug --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Ci Debug?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Ci Debug
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/sdsleon-ci-debug/badge)](https://www.skillsdirectory.com/skills/sdsleon-ci-debug)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: ci-debug
description: "Diagnose a failing GitHub Actions check by reading the real logs and finding the first true failure."
---

# CI Debug

Find out why a check is failing, from evidence rather than from the check's name.

## Find the real failure

Start from the check run, get its job, and read the **log**. A job summary tells you something failed; the log tells
you what.

Scan for the _first_ failure, not the loudest one. A long red log is usually one root cause followed by cascading
noise — later errors are often just the same failure surfacing again downstream.

Note the step, the command, and the exact error text. Copy the error verbatim; do not paraphrase it into something
that sounds cleaner than it was.

## Separate the failure from the change

Before blaming the diff, check whether the same job fails on the base branch or on unrelated recent runs. Flaky and
pre-existing failures look identical to real ones in a single run.

Say explicitly which of these you concluded:

- the change caused it,
- it was already broken,
- it is flaky (and what evidence supports that),
- you could not tell from the available logs.

"I could not tell" is a legitimate answer. An invented root cause is not.

## Environment differences

When something passes locally and fails in CI, compare the things that actually differ: OS and runner image, tool and
dependency versions, environment variables and secrets, working directory, and whether the job runs against a merge
commit rather than the branch head.

## Fix and verify

Propose the narrowest fix that addresses the root cause. Re-run only the failed jobs when you can. If a re-run is
needed to confirm a flake, say that is what you are doing and why.

Do not re-run, cancel, or dispatch a workflow merely to gather more evidence unless the user authorized that GitHub
action. When a fix is local, run the closest equivalent check before asking CI to confirm it.

## Report

Lead with the root cause and the evidence line from the log. Then the fix. If the fix is unverified because CI has not
re-run yet, say so.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.