fix(loop): refresh branches before merge and watch post-merge main CI

Adopting the six v0-era PRs surfaced two conflict-shaped cases the merge
leg handled only by luck. Behind-main branches are refused outright by
the repo's block_on_outdated_branch protection, so the leg now refreshes
every branch against origin/main before merging — which is also where a
textual conflict should surface, as diff text the loop resolves only
where it authored the hunks, otherwise abandoning the PR to a human
with a comment. And the one guard no mergeability check provides is the
push run on main after the merge: three PRs touching the same file can
merge cleanly and contradict each other, so a red main now halts the
loop instead of the tick reporting merged and moving on.

Closes #238

.pi/skills/yj-loop/SKILL.md

@@ -163,7 +163,22 @@ Merge when, and only when, **all** hold:
 - the protection contexts `CI / check` and `CI / e2e` are green on the
   PR's head, read from the API, not from the PR page's badge;
 - the PR reports mergeable;
-- the critique leg ran and no open blocker stands.
+- the critique leg ran and no open blocker stands;
+- the branch is **not behind `origin/main`** — the protection's
+  `block_on_outdated_branch: true` refuses it anyway; never
+  `force_manually_merged` around it.
+
+**Refresh before every merge.** In the loop worktree: fetch, then
+`git merge origin/main` on the PR branch, push. A textual conflict
+stops the leg there — as diff text, not as a failed merge click: hunks
+the loop authored are resolved by the loop; anything else is left with
+`⟦loop⟧` comment for a human, never forced. After any refresh push,
+re-poll the PR's own required contexts on the **new head** before
+merging.
+
+**Merges happen one at a time**, each re-reading state — the previous
+merge moved `main`, and the next PR's mergeability is recomputed at
+its own turn.
 
 ```
 curl -sS -X POST -H "Authorization: token $GITEA_TOKEN" \
@@ -172,12 +187,20 @@ curl -sS -X POST -H "Authorization: token $GITEA_TOKEN" \
   -d '{"Do":"merge","merge_message_field":"default","force_manually_merged":false}'
 ```
 
-Afterwards: `scripts/issue.sh list --state open` and check the footer
+**Afterwards watch the `push` run on `main`** — the CI the merge
-took. Close stragglers with `issue.sh close`, naming the merge commit.
+started. A red main after a loop merge is a **halt**: comment what is
-`unclaim.yml` handles the label; it is not instant; reopening does not
+known on the offending PR, mark the state file, stop taking new issues.
-restore it. Merging fans out to nothing (releases are the manual
+That run is the only thing between a clean textual merge of
-`release.yml`, which the loop never runs) — the criticism stands before
+independently-written PRs and a self-contradicting main; no
-the merge because nothing stands after it.
+mergeability check sees it. Only a green main lets the tick proceed (to
+footer verification, below).
+
+Footer verification: `scripts/issue.sh list --state open` and check
+the footer took. Close stragglers with `issue.sh close`, naming the
+merge commit. `unclaim.yml` handles the label; it is not instant;
+reopening does not restore it. Merging fans out to nothing (releases
+are the manual `release.yml`, which the loop never runs) — the
+criticism stands before the merge because nothing stands after it.
 
 ## Rails — the loop's absolute rules
 

.planning/plans/active/020-autonomous-backlog-loop.md

@@ -140,6 +140,20 @@ from a concurrent session is caught before the first edit.
 
 - **Only PRs the loop opened.** A collaborator's PR is never merged, never
   commented on for pressure, never touched.
+- **Every branch is refreshed against main before its merge**, in the
+  loop worktree — the refresh is where a textual conflict surfaces, as
+  diff text: hunks the loop authored are resolved there, anything else
+  is left to a human with a `⟦loop⟧` comment. The protection's
+  `block_on_outdated_branch` makes the refresh mandatory for adopted
+  (pre-loop) branches: behind `main`, a PR cannot merge at all.
+  Required contexts are re-polled on the refreshed head.
+- **Merges are one at a time**, each re-reading state — the previous
+  merge moved `main`, and the next PR's mergeability is recomputed at
+  its own turn.
+- **Post-merge, the `push` run on `main` is watched.** A red main after
+  a loop merge halts the loop. That run is the only guard against the
+  class no mergeability check sees: two PRs touching the same file,
+  merging cleanly, contradicting each other.
 - The gate is the protection rule itself, read from the API: contexts
   `CI / check*` and `CI / e2e*` green, PR mergeable. (Required approvals
   is 0 today; if a second person changes protection rules, the merge