Local agent counterpart to App /heyeddi fails: fetch failing Checks, write a diagnosis report, fix actionable failures, verify with evidenced commands. Stack-agnostic (log-driven).
Default behavior (no confirmation)
Run this skill automatically on any PR work — open PR, PR branch, review request, or before @heyeddi-pr-respond. Do not wait for the user to mention CI or ask you to fix failures.
- Always fetch and diagnose failing Checks
- Always apply fixes when the diagnosis identifies actionable failures
- Always commit + push fixes as part of the PR workflow
- Only merge still requires authorize merge in the current turn
Ephemeral artifacts (do not commit)
| File | Role |
|---|---|
pr-<N>-ci-fails-raw.json |
Raw gh evidence |
pr-<N>-ci-fails.md |
Diagnosis report |
Never commit these. GitHub Checks remain SSOT.
Pipeline
fetch_failing_checks --pr <N>
write_ci_fails_report --pr <N> --force
→ fill likely cause / proposed fix from logs
apply fixes when diagnosis identifies actionable failures (default — do not ask)
discover_and_verify [--run]
assert_no_merge --check
→ commit + push to PR branch when fixes applied
Also tell the user they can comment /heyeddi fails on the PR for hosted App analysis (billable when that path applies).
Hard rules
- Never invent failure causes without citing check evidence
- Never merge without authorize merge
- Never claim Spot/GHA jobs ran without Check or Spot evidence
Tools
| Tool | Purpose |
|---|---|
fetch_failing_checks |
Read-only failing Checks rollup |
write_ci_fails_report |
Write pr-<N>-ci-fails.md |
discover_and_verify |
Evidenced verify after fixes |
assert_no_merge |
Merge hard gate |
When complete
python .agents/skills/heyeddi/scripts/suggest_next_skill.py \
--current-skill heyeddi-ci-fails --project-root .