feat(bin): add bounded PR delivery loop

.agents/skills/pr-delivery/SKILL.md

@@ -0,0 +1,39 @@
+---
+name: pr-delivery
+description: Agent-only procedure for handling check: pr-delivery wakes from the bounded main-home PR delivery loop. Use on every check: pr-delivery wake to enumerate merge-eligible PRs, merge under configured authority via fm-pr-merge.sh, then defer post-merge orchestration to its owning flow.
+user-invocable: false
+metadata:
+  internal: true
+---
+
+# pr-delivery
+
+`bin/fm-pr-delivery.sh` discovers open PRs for merge-capable registered projects and classifies each head with live gh evidence.
+The watcher and locked session start run its bounded `scan` adjunct; when a merge-eligible PR exists, they queue `check: pr-delivery` with the scan payload.
+
+This skill owns every handling turn for that wake.
+Do not invent a parallel PR poll or merge path.
+
+## On wake
+
+1. Read the wake payload from the drain.
+   It is `merge-eligible: ...` with `project=`, `repo=`, `pr=`, `task=`, `url=`, and `head=` fields.
+2. Run `bin/fm-pr-delivery.sh show` when you need the current reason-coded blocked queue for operator context.
+
+## merge-eligible
+
+1. Reconcile the named task with `bin/fm-crew-state.sh` when current state matters.
+2. Confirm the PR URL, task id, project, and expected head still match the payload and the blocked queue does not show a stronger hold.
+3. Decide merge authority:
+   - With standing or task `yolo=on`, merge when the scan classified the PR as eligible.
+   - Otherwise escalate to the captain for explicit merge approval before calling `bin/fm-pr-merge.sh`.
+4. Merge only through `bin/fm-pr-merge.sh <task-id> <full-pr-url> --expected-head <payload-head>`.
+   If the expected-head guard refuses, leave the PR unmerged and let the delivery scan classify the new head.
+   Never call `gh` or `gh-axi pr merge` directly around that helper.
+5. After a successful merge, stop this delivery handling turn.
+   Project and secondmate fleet sync, validation and ready reporting, and teardown remain outside this delivery loop.
+
+## Holds this scan respects
+
+The scan already blocks on checks, mergeability, review issues, active task blockers, migration evidence, unresolved authority decisions, destructive or real-contact decisions, and missing task linkage.
+Do not override a hold the scan reported without new evidence or captain authority.

AGENTS.md

@@ -402,9 +402,10 @@ Handle actionable wakes as follows:
 1. For `signal:`, read the listed event lines first, then reconcile current state only where action depends on it.
 2. For `stale:`, inspect the recorded endpoint and load `stuck-crewmate-recovery` for a stopped, looping, confused, or unresponsive worker; a deep-inspection reason also requires current-state and validation-log inspection.
 3. For `check:`, act on the named poll result, including merges, Relay events, process-to-event source results, and captain inbox notes; a handled inbox note is also acknowledged with `bin/fm-inbox.sh drain --ack <id>`, or it stays counted as still waiting for firstmate.
+   `pr-delivery` obligations are handled by loading the `pr-delivery` skill.
 4. For `heartbeat:`, review the whole fleet from the structured fleet view, reconcile suspicious tasks and PR state, update the backlog, and never report an unchanged fleet as progress.
 
-When any wake reports a merged PR for a project cloned in this home, refresh that clone through the guarded fleet-sync path.
+When a non-`pr-delivery` wake reports a merged PR for a project cloned in this home, refresh that clone through the guarded fleet-sync path.
 When Relay-linked work reaches a milestone or terminal state, load `fmx-respond`; before terminal teardown, use its promised-final reconciliation when a typed public commitment exists, otherwise post the final completion follow-up so the link clears even if earlier follow-ups were spent.
 
 A secondmate's idle endpoint is healthy, and parent supervision relies on its routed status rather than treating a quiet pane as stale.
@@ -538,6 +539,7 @@ These skills are not captain-invocable; load them only at their precise triggers
 - `captain-hold-lifecycle` - load before treating an investigation or visual review as complete, before ending a visual review that exposed a captain decision, when recording or routing the captain's answer, and on any `RECORD DIVERGENCE` line from the wake drain.
 - `process-event-sources` - load before arming a long-polling source, before registering a deterministic condition->action watch (do X as soon as Y is true), and on any `procevent <adapter> <source-id> <sequence>` check wake.
   Never run a registered source's blocking command yourself in a conversational turn.
+- `pr-delivery` - load on a `check: pr-delivery` wake to handle a merge-eligible PR from the bounded PR delivery scan.
 - `fmx-respond` - load on an `x-mention <request_id>` `check:` wake to handle the mention, on an `x-mode-error ...` `check:` wake to report the Relay configuration blocker, on a `public-followup ...` `check:` wake or a startup-surfaced public commitment, and on any milestone or terminal wake for a Relay-linked task before posting its completion follow-up; relevant only when Relay is on.
 - `firstmate-codexapp` - load before coordinating a visible Codex Desktop thread, evaluating a Codex App backend request, or reconciling Codex Desktop host-tool smoke evidence for Firstmate work.
 - `firstmate-coding-guidelines` - load before changing firstmate's shared, tracked material, as defined by section 1's list, whether editing directly or briefing a crewmate for a firstmate-repo task.

bin/fm-pr-check.sh

@@ -65,18 +65,18 @@ fi
 "$FM_ROOT/bin/fm-guard.sh" || true
 
 # pr_head is recorded only when the forge's CLI can supply it. gh exposes the
-# head commit as a selectable field; plain glab exposes it only inside its JSON
-# output, which would need a JSON processor firstmate does not require, so a
-# GitLab task records no pr_head. Both consumers already treat it as optional:
+# head commit as a selectable field and accepts the canonical PR URL without a
+# local repository. Plain glab exposes it only inside its JSON output, which
+# would need a JSON processor firstmate does not require, so a GitLab task
+# records no pr_head. Both consumers already treat it as optional:
 # bin/fm-teardown.sh reads the head from the forge at teardown rather than from
 # metadata and falls back to its provider-agnostic content check, and
 # bin/fm-review-diff.sh resolves the head from the remote when none is recorded.
 # bin/fm-pr-merge.sh reads a GitLab head live at merge time for the same reason,
 # and treats a recorded value that disagrees as stale rather than authoritative.
-WT=$(grep '^worktree=' "$META" | tail -1 | cut -d= -f2- || true)
 PR_HEAD=
-if [ "$PROVIDER" = github ] && [ -n "$WT" ] && [ -d "$WT" ] && command -v gh >/dev/null 2>&1; then
-  if REMOTE_HEAD=$(cd "$WT" && gh pr view "$URL" --json headRefOid -q .headRefOid 2>/dev/null) \
+if [ "$PROVIDER" = github ] && command -v gh >/dev/null 2>&1; then
+  if REMOTE_HEAD=$(gh pr view "$URL" --json headRefOid -q .headRefOid 2>/dev/null) \
     && fm_pr_head_valid "$REMOTE_HEAD"; then
     PR_HEAD=$REMOTE_HEAD
   fi