ci-debug

name	ci-debug
description	Diagnose and fix GitHub Actions CI failures. Given a URL, PR number, or "CI is failing", parses logs, identifies root cause, fixes locally, and verifies. Use when CI is red and you need to get it green.
allowed-tools	Bash, Read, Edit, Write, Glob, Grep, Task, Agent, Skill
skills	["dancing-bear-rules"]

CI Debug

Delegates to workflows/code/ci-debug.yaml.

When to Use

• User pastes a GitHub Actions run URL
• User says "CI is failing" or "fix CI"
• After a push when CI goes red
• As Phase 4 of /review-fix

How to Run

IMPORTANT: Use the /workflow skill — do NOT call ./bin/workflow run --execute directly. It only writes dispatch files and exits (status=pending). The /workflow skill is what actually spawns agents, waits for results, and handles human gates.

# From a GitHub Actions run URL
Skill(skill="workflow", args="--workflow workflows/code/ci-debug.yaml --params github-actions_url=<RUN_URL> test_cmd='PYTHONPATH="$PWD/src" python3 -m unittest discover -s tests/<domain> -t . -f -q' source_root='<domain>/'")

# From a PR number
Skill(skill="workflow", args="--workflow workflows/code/ci-debug.yaml --params pr_number=<PR_NUMBER> test_cmd='PYTHONPATH="$PWD/src" python3 -m unittest discover -s tests/<domain> -t . -f -q' source_root='<domain>/'")

# Both (run URL for logs + qlty PR-scoped check)
Skill(skill="workflow", args="--workflow workflows/code/ci-debug.yaml --params github-actions_url=<RUN_URL> pr_number=<PR_NUMBER> test_cmd='PYTHONPATH="$PWD/src" python3 -m unittest discover -s tests/<domain> -t . -f -q' source_root='<domain>/'")

At least one of github-actions_url or pr_number must be provided.

Workflow Params

Param	Default	Description
github-actions_url	""	GitHub Actions run URL (optional if pr_number given)
pr_number	""	PR number to check (optional if github-actions_url given)
test_cmd	""	Test invocation for local verify (e.g. PYTHONPATH="$PWD/src" python3 -m unittest discover -s tests/core -t . -f -q; -f is unittest's failfast — -x is pytest syntax and errors here. The PYTHONPATH prefix is required in a worktree or imports resolve to the main checkout.)
source_root	""	Source dir for coverage scoping (e.g. core/)
min_coverage	"80"	Minimum new coverage % required by qlty gate

Workflow Stages

1. fetch-failures — fetch GitHub Actions failure logs and qlty gate status in parallel. Uses gh run view <RUN_ID> --log-failed for the failing job output and gh pr checks <PR> --json name,state,detailsUrl to resolve which run to inspect.
2. diagnose — parse raw output, extract structured failure records, write findings.json
3. fix — apply targeted fixes for each failure; spawn ci-fixer agent for complex multi-file cases
4. verify — run the specific previously-failing tests to confirm each fix landed
5. recheck — run the full domain test suite and re-check qlty for regressions
6. human-gate — present structured report (N failures found, root cause, fix applied, verification result)
7. sq-check-qlty — qlty quality gate fix loop (from shared fragment)
8. sq-verify-qlty — final qlty confirmation (from shared fragment)