investigate-ci-failure

name	investigate-ci-failure
description	Investigate CI job failures on a GitHub pull request (checks, Prow, or GitHub Actions). Use when the user pastes a PR URL and asks about CI failures or red checks.
disable-model-invocation	true

Investigate CI Failure

Default repo: openshift/lightspeed-agentic-sandbox. Adjust org/repo if the user gives another URL.

Workflow

1. Extract PR info

gh api repos/{org}/{repo}/pulls/{pr} --jq '{title, state, user: .user.login, head_sha: .head.sha}'
gh api repos/{org}/{repo}/pulls/{pr}/files --jq '.[].filename'

2. Get check statuses

gh pr checks {pr} --repo {org}/{repo}
gh api repos/{org}/{repo}/commits/{head_sha}/check-runs --jq '.check_runs[] | select(.conclusion == "failure") | {name, html_url}'

For OpenShift Prow, gh pr checks often surfaces Prow job links; use the target URL from failed contexts to open logs.

3. Prow / GCS artifacts (when applicable)

From a Prow target_url like:

https://prow.ci.openshift.org/view/gs/test-platform-results/pr-logs/pull/{org}_{repo}/{pr}/{job_name}/{build_id}

Derive artifact bases as in the upstream skill (GCS browser + raw storage.googleapis.com URLs) and fetch the failing step log (build, unit test, lint).

4. Konflux / Tekton

This repo uses Konflux (.tekton/). If checks reference Tekton/Konflux:

• Open the failed pipeline run from the GitHub check details.
• Read the failing task log (often build, prefetch-dependencies, or EC).

5. Diagnose

• Classify: dependency prefetch vs lint vs test vs image build.
• Tie failure to a file or command in this repo (Makefile target, pyproject.toml, Containerfile).
• Suggest the smallest fix; do not guess — quote the log line.

6. Report

• Failed job name(s) + link(s)
• Root cause (one paragraph)
• Concrete next step (file/command)