fix: bug: dev-poll stale detection races with issue_claim — blocks freshly claimed issues (#471)

Add 60-second grace period to stale in-progress detection in dev-poll.sh.
When a poller sees an in-progress issue with no assignee/PR/lock, it now
checks the timeline API for when the label was added. If <60s ago, it
skips stale detection to allow the claiming agent time to finish its
assign+label sequence.

Also documents the intentional assign-before-label ordering in
issue_claim() that minimizes the race window.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

dev/dev-poll.sh

@@ -103,6 +103,37 @@ is_blocked() {
 # STALENESS DETECTION FOR IN-PROGRESS ISSUES
 # =============================================================================
 
+# Check if in-progress label was added recently (within grace period).
+# Prevents race where a poller marks an issue as stale before the claiming
+# agent's assign + label sequence has fully propagated. See issue #471.
+# Args: issue_number [grace_seconds]
+# Returns: 0 if recently added (within grace period), 1 if not
+in_progress_recently_added() {
+  local issue="$1" grace="${2:-60}"
+  local now label_ts delta
+
+  now=$(date +%s)
+
+  # Query issue timeline for the most recent in-progress label event
+  label_ts=$(curl -sf -H "Authorization: token ${FORGE_TOKEN}" \
+    "${API}/issues/${issue}/timeline" | \
+    jq -r '[.[] | select(.type == "label") | select(.label.name == "in-progress")] | last | .created_at // empty') || true
+
+  if [ -z "$label_ts" ]; then
+    return 1  # no label event found — not recently added
+  fi
+
+  # Convert ISO timestamp to epoch and compare
+  local label_epoch
+  label_epoch=$(date -d "$label_ts" +%s 2>/dev/null || echo 0)
+  delta=$(( now - label_epoch ))
+
+  if [ "$delta" -lt "$grace" ]; then
+    return 0  # within grace period
+  fi
+  return 1
+}
+
 # Check if there's an open PR for a specific issue
 # Args: issue_number
 # Returns: 0 if open PR exists, 1 if not
@@ -460,9 +491,15 @@ if [ "$ORPHAN_COUNT" -gt 0 ]; then
     fi
 
     if [ "$OPEN_PR" = false ] && [ "$BLOCKED_BY_INPROGRESS" = false ]; then
-      log "issue #${ISSUE_NUM} is stale (no assignee, no open PR, no agent lock) — relabeling to blocked"
-      relabel_stale_issue "$ISSUE_NUM" "no_assignee_no_open_pr_no_lock"
-      BLOCKED_BY_INPROGRESS=true
+      # Grace period: skip if in-progress label was added <60s ago (issue #471)
+      if in_progress_recently_added "$ISSUE_NUM" 60; then
+        log "issue #${ISSUE_NUM} in-progress label added <60s ago — skipping stale detection (grace period)"
+        BLOCKED_BY_INPROGRESS=true
+      else
+        log "issue #${ISSUE_NUM} is stale (no assignee, no open PR, no agent lock) — relabeling to blocked"
+        relabel_stale_issue "$ISSUE_NUM" "no_assignee_no_open_pr_no_lock"
+        BLOCKED_BY_INPROGRESS=true
+      fi
     fi
 
     # Formula guard: formula-labeled issues should not be worked on by dev-agent.

lib/issue-lifecycle.sh

@@ -102,7 +102,9 @@ issue_claim() {
     return 1
   fi
 
-  # Assign to self (Forgejo rejects if already assigned differently)
+  # Assign to self BEFORE adding in-progress label (issue #471).
+  # This ordering ensures the assignee is set by the time other pollers
+  # see the in-progress label, reducing the stale-detection race window.
   curl -sf -X PATCH \
     -H "Authorization: token ${FORGE_TOKEN}" \
     -H "Content-Type: application/json" \