lsm / neokai / 28262654913

82%

Build Type

push

github

Committed by web-flow

Commit Message

fix(space): route reviewer merge conflicts to coder, not human (#2189)

* fix(space): route reviewer post-approval merge conflicts to coder, not human

When a reviewer's post-approval squash-merge fails on a conflict with the
base branch (`dev`), it used to escalate straight to a human via
request_human_input. Conflicts are routine coder work, so this created
unnecessary human load.

Update the shared post-approval merge instructions (step 3) so the reviewer:
- Detects the conflict (DIRTY mergeStateStatus / conflict markers in stderr)
- Records each attempt as Forge evidence (merge_conflict_loop artifact)
- Messages the upstream coder with pr_url, base branch, and conflicting files,
  instructing it to rebase onto origin/dev, resolve, test, push, report back
- Does NOT close the task or escalate to a human on a conflict alone
- Re-attempts the merge after the coder reports back, capped at 2 coder
  attempts, then escalates to space-agent with the accumulated evidence

The postApproval.instructions field is structural and synced from the template
on restamp, so existing spaces pick up the new behavior automatically.

Adds 7 unit tests verifying the conflict-routing content and ordering.

* fix(space): address review on merge-conflict routing (review_url, conflict files, re-verify)

Addresses PR #2189 review findings (reviewer + Codex) on the post-approval
merge-conflict routing template:

- Include review_url in the coder handoff payload. The Review -> Coding channel
  is guarded by review-posted-gate, which requires both pr_url and review_url
  and resets each cycle; a handoff carrying only pr_url would be blocked and the
  coder would never receive the rebase request. The reviewer now looks up the
  already-posted approval review permalink and sends both fields.
- Derive actual conflicting paths from the merge failure output / a local
  git merge-tree trial instead of `gh pr view --json files`, which lists every
  PR file rather than the conflict sub... (continued)

Coverage Stats

11421 of 16451 branches covered (69.42%)

Branch coverage included in aggregate %.

133 of 133 new or added lines in 2 files covered. (100.0%)

1 existing line in 1 file now uncovered.

96697 of 115419 relevant lines covered (83.78%)

300.93 hits per line