Skill v1.0.0

currentAutomated scan100/100

89jobrien/godmode/ci-fix

──Details

PublishedSeptember 27, 2026 at 12:42 PM

Content Hashsha256:5f9fea81c99d91c5...

Git SHA

──Files

Files (1 file, 3.7 KB)

SKILL.md3.7 KBactive

SKILL.md · 103 lines · 3.7 KB

version: "1.0.0" name: "godmode:ci-fix" description: > Self-healing CI diagnosis and fix loop. Use when CI is failing, when asked to "fix CI", or after a push that broke the pipeline. Fetches the latest failed run, classifies the root cause, applies a targeted fix, verifies locally, and re-pushes. requires: [] next: [cap]

CI Fix — Diagnose and Repair

Fetch the failure, classify it, fix it, verify locally, push.

Workflow

Step 1: Fetch failed runs

bash

gh run list --limit 5 --status failure

Pick the most recent failure for the current branch.

Step 2: Download logs

bash

gh run view <run-id> --log-failed

Read the full failure output. Identify the first error — that's usually the root cause.

Step 3: Classify

Class	Symptoms	Fix
`compile_error`	`error[E...]`, missing match arm, wrong import	Fix source, `cargo check`
`test_failure`	`FAILED`, assertion mismatch, panic	Fix impl or test, `cargo nextest run`
`clippy_warning`	`error: ...`, `-D warnings` gate	Fix all warnings, `cargo clippy`
`fmt_check`	`Diff in ...`	Run `cargo fmt --all`
`pre_commit_hook`	gitleaks, obfsck, coursers block	Add allowlist entry
`runner_environment`	missing tool, wrong Xcode, wrong target	Update workflow YAML
`false_positive_detection`	secret scanner flags test/doc content	Add `.gitleaksignore` entry
`dependency_issue`	lockfile conflict, yanked crate, version mismatch	Update `Cargo.toml`/`Cargo.lock`

If the class is ambiguous — report it to the user before fixing.

Step 4: Fix

Apply the minimum targeted fix for the classified root cause. Do not refactor unrelated code.

Verify locally before pushing:

Run via godmode verify or use the commands below:

bash

# compile_error / test_failure / clippy_warning / fmt_check
cargo check --workspace
cargo nextest run --workspace
cargo clippy --workspace -- -D warnings
cargo fmt --all --check
 
# runner_environment — read the workflow file first (use Read tool, not cat)
# Read: .github/workflows/<file>.yml
# then edit only the failing step

Step 5: Commit and push

Run: git branch --show-current

Verify output matches the expected branch. Stop immediately if not. (guardrails.nu check-branch <expected>)

bash

git add -A
git commit -m "fix(ci): <short description of root cause>"
git push

Then verify the new run passes:

bash

gh run list --limit 3   # polling fallback — interactive run watching requires a TTY

Devloop specialization

agents/ci-failure-responder.md is a devloop-specific specialization of this workflow — same triage/classify/fix/verify loop, with devloop-only notes baked in (edition 2024 let-chain style for collapsible_if, _DEVLOOP_OP_WRAPPED=1 prefix for nextest, worktree snapshot exclusions, BAML client drift). Use it directly when working in devloop.

Guardrails

Fix one root cause per pass. If there are multiple independent failures, fix them

sequentially — most CI failures cascade from a single root.

Do NOT switch self-hosted runners to GitHub-hosted without asking.
Do NOT change model names, API keys, or secrets in workflow files.
Do NOT use --no-verify.
If the same failure class recurs across 3+ sessions, add a note to CLAUDE.md to

prevent it at the source.

All versions