loopflow 0.12.10

Run steps and flows with coding agents
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61

---
requires: wave PR with failing CI checks (or CI failure message context)
produces: branch with CI failures fixed
diff_files: false
action_style: procedural
---
Fix failing CI checks for the current wave PR.

## Goal

Find the wave's current PR, identify the latest failing checks on its head commit, fix them, and leave the branch ready to push.

## Workflow

1. **Resolve the PR for this wave**
   - If the prompt message already includes check metadata (PR number, branch, commit SHA, logs URL), use it directly.
   - Otherwise, resolve the PR for the current branch with the GitHub CLI:
     ```bash
     gh pr view --json number,headRefName,headRefOid,url
     ```
   - If there is no PR, stop and explain what is missing.

2. **Fetch latest check-run failures**
   - Query checks for the PR head SHA (from `headRefOid`):
     ```bash
     repo=$(gh repo view --json nameWithOwner -q .nameWithOwner)
     sha=$(gh pr view <pr> --json headRefOid -q .headRefOid)
     gh api "repos/$repo/commits/$sha/check-runs"
     ```
   - Focus on failed checks from the most recent completed runs.
   - Use the check `html_url`/logs link for repro details.

3. **Fix one failing check at a time**
   - Reproduce the failure locally from the current branch.
   - Apply the smallest correct fix.
   - Run only the relevant local checks first, then broader checks if needed.
   - Repeat until failing checks for the head SHA are addressed or a real blocker remains.

4. **Verify and report**
   - Summarize what failed, what was changed, and what commands were run.
   - If no failing checks remain, say so clearly.
   - If blocked (missing secrets, flaky upstream, infra outage), give exact
     blocker details and the next manual action. Name the failing capability
     (provider, github-observation, secrets) — the Task runner independently
     detects provider outages and GitHub observation failures and transitions
     the Task to Blocked with that capability, so report it precisely. A PR
     whose head is still Failing after you finish (you could not repair it)
     also blocks the Task actionably rather than silently returning to Waiting.

## Guardrails

- Stay scoped to CI failures on this PR.
- Prefer targeted fixes over broad refactors.
- Do not ignore failing tests to get green.
- Do not fake progress. If you cannot repair the head, say so and stop — the
  runner blocks the Task on the unchanged failing head. Pushing a real fix
  (the head advances) is the only signal that another iteration should run.

## Adaptation

After fixing the immediate failure, ask: why did this get to CI? Could gate have caught it? If the answer points to a missing check in gate, a missing convention in repo docs, or a recurring ci-fix pattern — make that update. The fix addresses the symptom; the step or doc update prevents recurrence.