fix: refactor: replace escalation JSONL with blocked label + diagnostic comment (#352)

Replace the unreliable escalation JSONL system (supervisor/escalations-*.jsonl consumed by gardener) with direct blocked label + diagnostic comment on the original issue. When a dev-agent or action-agent session fails (PHASE:failed, idle timeout, crash, CI exhausted): - Capture last 50 lines from tmux pane via tmux capture-pane - Post a structured diagnostic comment on the issue (exit reason, timestamp, PR number, tmux output) - Label the issue "blocked" (instead of restoring "backlog") - Remove in-progress label Removed: - Escalation JSONL write paths in dev-agent.sh, phase-handler.sh, dev-poll.sh, action-agent.sh - is_escalated() helper in dev-poll.sh - Escalation triage (P2f section) in supervisor-poll.sh - Escalation processing + recipe engine in gardener-poll.sh - ci-escalation-recipes step from run-gardener.toml formula - escalations*.jsonl from .gitignore Added: - post_blocked_diagnostic() shared helper in phase-handler.sh - ensure_blocked_label_id() helper (creates label via API if not exists) - is_blocked() helper in dev-poll.sh (replaces is_escalated) - Blocked issues listing in supervisor/preflight.sh Kept: - Matrix notifications on failure (unchanged) - CI fix counter logic (still tracks attempts) - needs_human injection in supervisor/gardener (not escalation-related) - Gardener grooming (gardener-agent.sh still invoked) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-03-21 04:18:43 +00:00 · 2026-03-21 04:18:43 +00:00 · 61c44d31b1
commit 61c44d31b1
parent 0109f0b0c3
10 changed files with 181 additions and 990 deletions
--- a/.gitignore
+++ b/.gitignore
@ -11,7 +11,6 @@ RESOURCES.md
 state.json
 *.lock
 *.pid
-escalations*.jsonl
 metrics/supervisor-metrics.jsonl

 # OS
--- a/AGENTS.md
+++ b/AGENTS.md
@ -130,9 +130,9 @@ issue is filed against).

 **Key files**:
 - `gardener/gardener-run.sh` — Cron wrapper: lock, memory guard, dedup check, files action issue
- `gardener/gardener-poll.sh` — Recipe engine: escalation-reply injection for dev sessions, processes dev-agent CI escalations via recipe engine (invoked by formula step ci-escalation-recipes)
+- `gardener/gardener-poll.sh` — Escalation-reply injection for dev sessions, invokes gardener-agent.sh for grooming
 - `gardener/gardener-agent.sh` — Orchestrator: bash pre-analysis, creates tmux session (`gardener-{project}`) with interactive `claude`, monitors phase file, parses result file (ACTION:/DUST:/ESCALATE), handles dust bundling
- `formulas/run-gardener.toml` — Execution spec: preflight, grooming, blocked-review, CI escalation recipes, agents-update, commit-and-pr
+- `formulas/run-gardener.toml` — Execution spec: preflight, grooming, blocked-review, agents-update, commit-and-pr

 **Environment variables consumed**:
 - `CODEBERG_TOKEN`, `CODEBERG_REPO`, `CODEBERG_API`, `PROJECT_NAME`, `PROJECT_REPO_ROOT`
@ -159,8 +159,8 @@ runs directly from cron like the planner and predictor.
  `run_formula_and_monitor`
 - `supervisor/preflight.sh` — Data collection: system resources (RAM, disk, swap,
  load), Docker status, active tmux sessions + phase files, lock files, agent log
-  tails, CI pipeline status, open PRs, issue counts, stale worktrees, pending
-  escalations, Matrix escalation replies
+  tails, CI pipeline status, open PRs, issue counts, stale worktrees, blocked
+  issues, Matrix escalation replies
 - `formulas/run-supervisor.toml` — Execution spec: five steps (preflight review,
  health-assessment, decide-actions, report, journal) with `needs` dependencies.
  Claude evaluates all metrics and takes actions in a single interactive session
@ -373,7 +373,7 @@ Issues flow through these states:
 |---|---|---|
 | `backlog` | Issue is queued for implementation. Dev-poll picks the first ready one. | Planner, gardener, humans |
 | `in-progress` | Dev-agent is actively working on this issue. Only one issue per project is in-progress at a time. | dev-agent.sh (claims issue) |
-| `blocked` | Issue has unmet dependencies (other open issues). | gardener, supervisor (detected) |
+| `blocked` | Issue is stuck — agent session failed, crashed, timed out, or CI exhausted. Diagnostic comment on the issue has details. Also used for unmet dependencies. | dev-agent.sh, action-agent.sh, dev-poll.sh (on failure) |
 | `tech-debt` | Pre-existing issue flagged by AI reviewer, not introduced by a PR. | review-pr.sh (auto-created follow-ups) |
 | `underspecified` | Dev-agent refused the issue as too large or vague. | dev-poll.sh (on preflight `too_large`), dev-agent.sh (on mid-run `too_large` refusal) |
 | `vision` | Goal anchors — high-level objectives from VISION.md. | Planner, humans |
--- a/action/action-agent.sh
+++ b/action/action-agent.sh
@ -304,13 +304,12 @@ case "${_MONITOR_LOOP_EXIT:-}" in
    notify_ctx \
      "session idle for $((IDLE_TIMEOUT / 3600))h — killed" \
      "session idle for $((IDLE_TIMEOUT / 3600))h — killed"
-    # Escalate to supervisor (idle_prompt already escalated via _on_phase_change callback)
-    echo "{\"issue\":${ISSUE},\"pr\":${PR_NUMBER:-0},\"reason\":\"idle_timeout\",\"ts\":\"$(date -u +%Y-%m-%dT%H:%M:%SZ)\"}" \
-      >> "${FACTORY_ROOT}/supervisor/escalations-${PROJECT_NAME}.jsonl"
+    # Post diagnostic comment + label blocked (replaces escalation JSONL)
+    post_blocked_diagnostic "idle_timeout"
    rm -f "$PHASE_FILE" "${PHASE_FILE%.phase}.context" "$IMPL_SUMMARY_FILE" "$THREAD_FILE" "$SCRATCH_FILE"
    ;;
  idle_prompt)
-    # Notification + escalation already handled by _on_phase_change(PHASE:failed) callback
+    # Notification + blocked label already handled by _on_phase_change(PHASE:failed) callback
    rm -f "$PHASE_FILE" "${PHASE_FILE%.phase}.context" "$IMPL_SUMMARY_FILE" "$THREAD_FILE" "$SCRATCH_FILE"
    ;;
  PHASE:failed)
@ -319,8 +318,7 @@ case "${_MONITOR_LOOP_EXIT:-}" in
      notify_ctx \
        "session killed — wall-clock cap ($((MAX_LIFETIME / 3600))h) reached" \
        "session killed — wall-clock cap ($((MAX_LIFETIME / 3600))h) reached"
-      echo "{\"issue\":${ISSUE},\"pr\":${PR_NUMBER:-0},\"reason\":\"max_lifetime\",\"ts\":\"$(date -u +%Y-%m-%dT%H:%M:%SZ)\"}" \
-        >> "${FACTORY_ROOT}/supervisor/escalations-${PROJECT_NAME}.jsonl"
+      post_blocked_diagnostic "max_lifetime"
    fi
    rm -f "$PHASE_FILE" "${PHASE_FILE%.phase}.context" "$IMPL_SUMMARY_FILE" "$THREAD_FILE" "$SCRATCH_FILE"
    ;;
--- a/dev/dev-agent.sh
+++ b/dev/dev-agent.sh
@ -749,18 +749,15 @@ case "${_MONITOR_LOOP_EXIT:-}" in
  idle_timeout|idle_prompt)
    if [ "${_MONITOR_LOOP_EXIT:-}" = "idle_prompt" ]; then
      notify_ctx \
-        "session finished without phase signal — killed. Escalating to gardener." \
-        "session finished without phase signal — killed. Escalating to gardener.${PR_NUMBER:+ PR <a href='${CODEBERG_WEB}/pulls/${PR_NUMBER}'>#${PR_NUMBER}</a>}"
+        "session finished without phase signal — killed. Marking blocked." \
+        "session finished without phase signal — killed. Marking blocked.${PR_NUMBER:+ PR <a href='${CODEBERG_WEB}/pulls/${PR_NUMBER}'>#${PR_NUMBER}</a>}"
    else
      notify_ctx \
-        "session idle for 2h — killed. Escalating to gardener." \
-        "session idle for 2h — killed. Escalating to gardener.${PR_NUMBER:+ PR <a href='${CODEBERG_WEB}/pulls/${PR_NUMBER}'>#${PR_NUMBER}</a>}"
+        "session idle for 2h — killed. Marking blocked." \
+        "session idle for 2h — killed. Marking blocked.${PR_NUMBER:+ PR <a href='${CODEBERG_WEB}/pulls/${PR_NUMBER}'>#${PR_NUMBER}</a>}"
    fi
-    # Escalate: write to project-suffixed escalation file so gardener picks it up
-    echo "{\"issue\":${ISSUE},\"pr\":${PR_NUMBER:-0},\"reason\":\"${_MONITOR_LOOP_EXIT:-idle_timeout}\",\"ts\":\"$(date -u +%Y-%m-%dT%H:%M:%SZ)\"}" \
-      >> "${FACTORY_ROOT}/supervisor/escalations-${PROJECT_NAME}.jsonl"
-    # Restore labels: remove in-progress, add backlog
-    restore_to_backlog
+    # Post diagnostic comment + label issue blocked (replaces escalation JSONL)
+    post_blocked_diagnostic "${_MONITOR_LOOP_EXIT:-idle_timeout}"
    if [ -n "${PR_NUMBER:-}" ]; then
      log "keeping worktree (PR #${PR_NUMBER} still open)"
    else
@ -773,8 +770,8 @@ case "${_MONITOR_LOOP_EXIT:-}" in
    ;;
  crashed)
    # Belt-and-suspenders: _on_phase_change(PHASE:crashed) handles primary
-    # cleanup (escalation, notification, labels, worktree, files).
-    restore_to_backlog
+    # cleanup (diagnostic comment, blocked label, worktree, files).
+    post_blocked_diagnostic "crashed"
    ;;
  done)
    # Belt-and-suspenders: callback in phase-handler.sh handles primary cleanup,
--- a/dev/dev-poll.sh
+++ b/dev/dev-poll.sh
@ -77,40 +77,59 @@ else:
 " 2>/dev/null || echo "exhausted:99"
 }

-# Check whether an issue/PR has been escalated (unprocessed or processed)
-is_escalated() {
-  local issue="$1" pr="$2"
-  python3 -c "
-import json, sys
-try:
-  issue, pr = int('${issue}'), int('${pr}')
-except (ValueError, TypeError):
-  sys.exit(1)
-for path in ['${FACTORY_ROOT}/supervisor/escalations-${PROJECT_NAME}.jsonl',
-             '${FACTORY_ROOT}/supervisor/escalations-${PROJECT_NAME}.done.jsonl']:
-  try:
-    with open(path) as fh:
-      for line in fh:
-        line = line.strip()
-        if not line:
-          continue
-        d = json.loads(line)
-        if d.get('issue') == issue and d.get('pr') == pr:
-          sys.exit(0)
-  except OSError:
-    pass
-sys.exit(1)
-" 2>/dev/null && return 0 || return 1
+# Check whether an issue already has the "blocked" label
+is_blocked() {
+  local issue="$1"
+  codeberg_api GET "/issues/${issue}/labels" 2>/dev/null \
+    | jq -e '.[] | select(.name == "blocked")' >/dev/null 2>&1
+}
+
+# Post a CI-exhaustion diagnostic comment and label issue as blocked.
+# Args: issue_num pr_num attempts
+_post_ci_blocked_comment() {
+  local issue_num="$1" pr_num="$2" attempts="$3"
+  local blocked_id
+  blocked_id=$(codeberg_api GET "/labels" 2>/dev/null \
+    | jq -r '.[] | select(.name == "blocked") | .id' 2>/dev/null || true)
+  if [ -z "$blocked_id" ]; then
+    blocked_id=$(curl -sf -X POST \
+      -H "Authorization: token ${CODEBERG_TOKEN}" \
+      -H "Content-Type: application/json" \
+      "${CODEBERG_API}/labels" \
+      -d '{"name":"blocked","color":"#e11d48"}' 2>/dev/null \
+      | jq -r '.id // empty' 2>/dev/null || true)
+  fi
+  [ -z "$blocked_id" ] && return 0
+
+  local comment
+  comment="### Session failure diagnostic
+
+| Field | Value |
+|---|---|
+| Exit reason | \`ci_exhausted_poll (${attempts} attempts)\` |
+| Timestamp | \`$(date -u +%Y-%m-%dT%H:%M:%SZ)\` |
+| PR | #${pr_num} |"
+
+  curl -sf -X POST \
+    -H "Authorization: token ${CODEBERG_TOKEN}" \
+    -H "Content-Type: application/json" \
+    "${CODEBERG_API}/issues/${issue_num}/comments" \
+    -d "$(jq -nc --arg b "$comment" '{body:$b}')" >/dev/null 2>&1 || true
+  curl -sf -X POST \
+    -H "Authorization: token ${CODEBERG_TOKEN}" \
+    -H "Content-Type: application/json" \
+    "${CODEBERG_API}/issues/${issue_num}/labels" \
+    -d "{\"labels\":[${blocked_id}]}" >/dev/null 2>&1 || true
 }

 # =============================================================================
-# HELPER: handle CI-exhaustion check/escalate (DRY for 3 call sites)
+# HELPER: handle CI-exhaustion check/block (DRY for 3 call sites)
 # Sets CI_FIX_ATTEMPTS for caller use. Returns 0 if exhausted, 1 if not.
 #
 # Pass "check_only" as third arg for the backlog scan path: ok-counts are
 # returned without incrementing (deferred to launch time so a WAITING_PRS
 # exit cannot waste a fix attempt). The 3→4 sentinel bump is always atomic
-# regardless of mode, preventing duplicate escalation writes from concurrent
+# regardless of mode, preventing duplicate blocked labels from concurrent
 # pollers.
 # =============================================================================
 handle_ci_exhaustion() {
@ -118,18 +137,18 @@ handle_ci_exhaustion() {
  local check_only="${3:-}"
  local result

-  # Fast path: already in the escalation file — skip without touching counter.
-  if is_escalated "$issue_num" "$pr_num"; then
+  # Fast path: already blocked — skip without touching counter.
+  if is_blocked "$issue_num"; then
    CI_FIX_ATTEMPTS=$(ci_fix_count "$pr_num")
-    log "PR #${pr_num} (issue #${issue_num}) already escalated (${CI_FIX_ATTEMPTS} attempts) — skipping"
+    log "PR #${pr_num} (issue #${issue_num}) already blocked (${CI_FIX_ATTEMPTS} attempts) — skipping"
    return 0
  fi

  # Single flock-protected call: read + threshold-check + conditional bump.
  # In check_only mode, ok-counts are returned without incrementing (deferred
  # to launch time). In both modes, the 3→4 sentinel bump is atomic, so only
-  # one concurrent poller can ever receive exhausted_first_time:3 and write
-  # the escalation entry.
+  # one concurrent poller can ever receive exhausted_first_time:3 and label
+  # the issue blocked.
  result=$(ci_fix_check_and_increment "$pr_num" "$check_only")
  case "$result" in
    ok:*)
@ -138,18 +157,17 @@ handle_ci_exhaustion() {
      ;;
    exhausted_first_time:*)
      CI_FIX_ATTEMPTS="${result#exhausted_first_time:}"
-      log "PR #${pr_num} (issue #${issue_num}) CI exhausted (${CI_FIX_ATTEMPTS} attempts) — escalated to gardener, skipping"
-      echo "{\"issue\":${issue_num},\"pr\":${pr_num},\"project\":\"${PROJECT_NAME}\",\"reason\":\"ci_exhausted_poll\",\"attempts\":${CI_FIX_ATTEMPTS},\"ts\":\"$(date -u +%Y-%m-%dT%H:%M:%SZ)\"}" \
-        >> "${FACTORY_ROOT}/supervisor/escalations-${PROJECT_NAME}.jsonl"
-      matrix_send "dev" "🚨 PR #${pr_num} (issue #${issue_num}) CI failed after ${CI_FIX_ATTEMPTS} attempts — escalated" 2>/dev/null || true
+      log "PR #${pr_num} (issue #${issue_num}) CI exhausted (${CI_FIX_ATTEMPTS} attempts) — marking blocked"
+      _post_ci_blocked_comment "$issue_num" "$pr_num" "$CI_FIX_ATTEMPTS"
+      matrix_send "dev" "🚨 PR #${pr_num} (issue #${issue_num}) CI failed after ${CI_FIX_ATTEMPTS} attempts — marked blocked" 2>/dev/null || true
      ;;
    exhausted:*)
      CI_FIX_ATTEMPTS="${result#exhausted:}"
-      log "PR #${pr_num} (issue #${issue_num}) CI exhausted (${CI_FIX_ATTEMPTS} attempts) — escalated to gardener, skipping"
+      log "PR #${pr_num} (issue #${issue_num}) CI exhausted (${CI_FIX_ATTEMPTS} attempts) — already blocked, skipping"
      ;;
    *)
      CI_FIX_ATTEMPTS=99
-      log "PR #${pr_num} (issue #${issue_num}) CI exhausted (${CI_FIX_ATTEMPTS} attempts) — escalated to gardener, skipping"
+      log "PR #${pr_num} (issue #${issue_num}) CI exhausted (${CI_FIX_ATTEMPTS} attempts) — already blocked, skipping"
      ;;
  esac
  return 0
--- a/dev/phase-handler.sh
+++ b/dev/phase-handler.sh
@ -34,6 +34,86 @@
 : "${CLAIMED:=false}"
 : "${PHASE_POLL_INTERVAL:=30}"

+# --- Look up (or create) the "blocked" label ID ---
+ensure_blocked_label_id() {
+  if [ -n "${_BLOCKED_LABEL_ID:-}" ]; then
+    printf '%s' "$_BLOCKED_LABEL_ID"
+    return 0
+  fi
+  _BLOCKED_LABEL_ID=$(codeberg_api GET "/labels" 2>/dev/null \
+    | jq -r '.[] | select(.name == "blocked") | .id' 2>/dev/null || true)
+  if [ -z "$_BLOCKED_LABEL_ID" ]; then
+    _BLOCKED_LABEL_ID=$(curl -sf -X POST \
+      -H "Authorization: token ${CODEBERG_TOKEN}" \
+      -H "Content-Type: application/json" \
+      "${CODEBERG_API}/labels" \
+      -d '{"name":"blocked","color":"#e11d48"}' 2>/dev/null \
+      | jq -r '.id // empty' 2>/dev/null || true)
+  fi
+  printf '%s' "$_BLOCKED_LABEL_ID"
+}
+
+# --- Post diagnostic comment + label issue as blocked ---
+# Replaces the old escalation JSONL write path.
+# Captures tmux pane output, posts a structured comment on the issue, removes
+# in-progress label, and adds the "blocked" label.
+#
+# Args: reason [session_name]
+# Uses globals: ISSUE, SESSION_NAME, PR_NUMBER, CODEBERG_TOKEN, API
+post_blocked_diagnostic() {
+  local reason="$1"
+  local session="${2:-${SESSION_NAME:-}}"
+
+  # Capture last 50 lines from tmux pane (before kill)
+  local tmux_output=""
+  if [ -n "$session" ] && tmux has-session -t "$session" 2>/dev/null; then
+    tmux_output=$(tmux capture-pane -p -t "$session" -S -50 2>/dev/null || true)
+  fi
+
+  # Build diagnostic comment body
+  local comment
+  comment="### Session failure diagnostic
+
+| Field | Value |
+|---|---|
+| Exit reason | \`${reason}\` |
+| Timestamp | \`$(date -u +%Y-%m-%dT%H:%M:%SZ)\` |"
+  [ -n "${PR_NUMBER:-}" ] && [ "${PR_NUMBER:-0}" != "0" ] && \
+    comment="${comment}
+| PR | #${PR_NUMBER} |"
+
+  if [ -n "$tmux_output" ]; then
+    comment="${comment}
+
+<details><summary>Last 50 lines from tmux pane</summary>
+
+\`\`\`
+${tmux_output}
+\`\`\`
+</details>"
+  fi
+
+  # Post comment to issue
+  curl -sf -X POST \
+    -H "Authorization: token ${CODEBERG_TOKEN}" \
+    -H "Content-Type: application/json" \
+    "${API}/issues/${ISSUE}/comments" \
+    -d "$(jq -nc --arg b "$comment" '{body:$b}')" >/dev/null 2>&1 || true
+
+  # Remove in-progress, add blocked
+  cleanup_labels
+  local blocked_id
+  blocked_id=$(ensure_blocked_label_id)
+  if [ -n "$blocked_id" ]; then
+    curl -sf -X POST \
+      -H "Authorization: token ${CODEBERG_TOKEN}" \
+      -H "Content-Type: application/json" \
+      "${API}/issues/${ISSUE}/labels" \
+      -d "{\"labels\":[${blocked_id}]}" >/dev/null 2>&1 || true
+  fi
+  CLAIMED=false
+}
+
 # --- Build phase protocol prompt (shared across agents) ---
 # Generates the phase-signaling instructions for Claude prompts.
 # Args: phase_file summary_file branch
@ -319,12 +399,11 @@ Write PHASE:awaiting_review to the phase file, then stop and wait for review fee
      CI_FIX_COUNT=$(( CI_FIX_COUNT + 1 ))
      _ci_pipeline_url="${WOODPECKER_SERVER}/repos/${WOODPECKER_REPO_ID}/pipeline/${PIPELINE_NUM:-0}"
      if [ "$CI_FIX_COUNT" -gt "$MAX_CI_FIXES" ]; then
-        log "CI failure not recoverable after ${CI_FIX_COUNT} fix attempts — escalating"
-        echo "{\"issue\":${ISSUE},\"pr\":${PR_NUMBER},\"reason\":\"ci_exhausted\",\"step\":\"${FAILED_STEP:-unknown}\",\"attempts\":${CI_FIX_COUNT},\"ts\":\"$(date -u +%Y-%m-%dT%H:%M:%SZ)\"}" \
-          >> "${FACTORY_ROOT}/supervisor/escalations-${PROJECT_NAME}.jsonl"
+        log "CI failure not recoverable after ${CI_FIX_COUNT} fix attempts — marking blocked"
+        post_blocked_diagnostic "ci_exhausted after ${CI_FIX_COUNT} attempts (step: ${FAILED_STEP:-unknown})"
        notify_ctx \
-          "CI exhausted after ${CI_FIX_COUNT} attempts — escalated to supervisor" \
-          "CI exhausted after ${CI_FIX_COUNT} attempts on PR <a href='${PR_URL:-${CODEBERG_WEB}/pulls/${PR_NUMBER}}'>#${PR_NUMBER}</a> | <a href='${_ci_pipeline_url}'>Pipeline</a><br>Step: <code>${FAILED_STEP:-unknown}</code> — escalated to supervisor"
+          "CI exhausted after ${CI_FIX_COUNT} attempts — issue marked blocked" \
+          "CI exhausted after ${CI_FIX_COUNT} attempts on PR <a href='${PR_URL:-${CODEBERG_WEB}/pulls/${PR_NUMBER}}'>#${PR_NUMBER}</a> | <a href='${_ci_pipeline_url}'>Pipeline</a><br>Step: <code>${FAILED_STEP:-unknown}</code> — issue marked blocked"
        printf 'PHASE:failed\nReason: ci_exhausted after %d attempts\n' "$CI_FIX_COUNT" > "$PHASE_FILE"
        # Do NOT update LAST_PHASE_MTIME here — let the main loop detect PHASE:failed
        return 0
@ -685,23 +764,13 @@ $(printf '%s' "$REFUSAL_JSON" | head -c 2000)
      return 1

    else
-      # Genuine unrecoverable failure — escalate to supervisor
+      # Genuine unrecoverable failure — label blocked with diagnostic
      log "session failed: ${FAILURE_REASON}"
      notify_ctx \
        "❌ Issue #${ISSUE} session failed: ${FAILURE_REASON}" \
        "❌ <a href='${CODEBERG_WEB}/issues/${ISSUE}'>Issue #${ISSUE}</a> session failed: ${FAILURE_REASON}${PR_NUMBER:+ | PR <a href='${CODEBERG_WEB}/pulls/${PR_NUMBER}'>#${PR_NUMBER}</a>}"
-      echo "{\"issue\":${ISSUE},\"pr\":${PR_NUMBER:-0},\"reason\":\"${FAILURE_REASON}\",\"ts\":\"$(date -u +%Y-%m-%dT%H:%M:%SZ)\"}" \
-        >> "${FACTORY_ROOT}/supervisor/escalations-${PROJECT_NAME}.jsonl"
+      post_blocked_diagnostic "$FAILURE_REASON"

-      # Restore backlog label
-      cleanup_labels
-      curl -sf -X POST \
-        -H "Authorization: token ${CODEBERG_TOKEN}" \
-        -H "Content-Type: application/json" \
-        "${API}/issues/${ISSUE}/labels" \
-        -d "{\"labels\":[${BACKLOG_LABEL_ID}]}" >/dev/null 2>&1 || true
-
-      CLAIMED=false  # Don't unclaim again in cleanup()
      agent_kill_session "$SESSION_NAME"
      if [ -n "${PR_NUMBER:-}" ]; then
        log "keeping worktree (PR #${PR_NUMBER} still open)"
@ -715,18 +784,14 @@ $(printf '%s' "$REFUSAL_JSON" | head -c 2000)
    fi

  # ── PHASE: crashed ──────────────────────────────────────────────────────────
-  # Session died unexpectedly (OOM kill, tmux crash, etc.). Escalate to
-  # supervisor and restore issue to backlog so it can be retried.
+  # Session died unexpectedly (OOM kill, tmux crash, etc.). Label blocked with
+  # diagnostic comment so humans can triage directly on the issue.
  elif [ "$phase" = "PHASE:crashed" ]; then
    log "session crashed for issue #${ISSUE}"
    notify_ctx \
-      "session crashed unexpectedly — escalating" \
-      "session crashed unexpectedly — escalating${PR_NUMBER:+ | PR <a href='${CODEBERG_WEB}/pulls/${PR_NUMBER}'>#${PR_NUMBER}</a>}"
-    echo "{\"issue\":${ISSUE},\"pr\":${PR_NUMBER:-0},\"reason\":\"crashed\",\"ts\":\"$(date -u +%Y-%m-%dT%H:%M:%SZ)\"}" \
-      >> "${FACTORY_ROOT}/supervisor/escalations-${PROJECT_NAME}.jsonl"
-
-    # Restore backlog label, clean up worktree + temp files
-    restore_to_backlog
+      "session crashed unexpectedly — marking blocked" \
+      "session crashed unexpectedly — marking blocked${PR_NUMBER:+ | PR <a href='${CODEBERG_WEB}/pulls/${PR_NUMBER}'>#${PR_NUMBER}</a>}"
+    post_blocked_diagnostic "crashed"
    [ -z "${PR_NUMBER:-}" ] && cleanup_worktree
    [ -n "${PR_NUMBER:-}" ] && log "keeping worktree (PR #${PR_NUMBER} still open)"
    rm -f "$PHASE_FILE" "$IMPL_SUMMARY_FILE" "$THREAD_FILE" "${SCRATCH_FILE:-}" \
--- a/formulas/run-gardener.toml
+++ b/formulas/run-gardener.toml
@ -1,17 +1,16 @@
 # formulas/run-gardener.toml — Gardener housekeeping formula
 #
 # Defines the gardener's complete run: grooming (Claude session via
-# gardener-agent.sh) + CI escalation recipes (bash, gardener-poll.sh)
-# + AGENTS.md maintenance + final commit-and-pr.
+# gardener-agent.sh) + blocked-review + AGENTS.md maintenance + final
+# commit-and-pr.
 #
 # No memory, no journal. The gardener does mechanical housekeeping
 # based on current state — it doesn't need to remember past runs.
 #
-# Steps: preflight → grooming → blocked-review → ci-escalation-recipes
-#        → agents-update → commit-and-pr
+# Steps: preflight → grooming → blocked-review → agents-update → commit-and-pr

 name        = "run-gardener"
-description = "Mechanical housekeeping: grooming, blocked review, CI escalation recipes, docs update"
+description = "Mechanical housekeeping: grooming, blocked review, docs update"
 version     = 1

 [context]
@ -184,40 +183,7 @@ CRITICAL: If this step fails, log the failure and move on.
 needs = ["grooming"]

 # ─────────────────────────────────────────────────────────────────────
-# Step 4: ci-escalation-recipes — recipe-driven CI failure handling
-# ─────────────────────────────────────────────────────────────────────
-
-[[steps]]
-id    = "ci-escalation-recipes"
-title = "CI escalation recipes (bash — gardener-poll.sh)"
-executor = "bash"
-script   = "gardener/gardener-poll.sh --recipes-only"
-description = """
-NOT a Claude step — executed by gardener-poll.sh before/after the Claude session.
-Documented here so the formula covers the full gardener run.
-
-gardener-poll.sh processes CI escalation entries from
-supervisor/escalations-{project}.jsonl. Each entry is a dev-agent session
-that exhausted its CI fix attempts and was escalated to the gardener.
-
-The recipe engine (match_recipe function in gardener-poll.sh) matches each
-escalation against gardener/recipes/*.toml by priority order, then executes
-the matched recipe's playbook actions via bash functions.
-
-Recipes (see gardener/recipes/*.toml for definitions):
- chicken-egg-ci (priority 10): non-blocking bypass + per-file fix issues
- cascade-rebase (priority 20): rebase via Gitea API, re-approve, retry merge
- flaky-test (priority 30): retrigger CI or quarantine
- shellcheck-violations (priority 40): per-file ShellCheck fix issues
- Generic fallback: one combined CI failure issue
-
-Special cases:
- idle_timeout / idle_prompt: investigation issues (no recipe matching)
-"""
-needs = ["grooming"]
-
-# ─────────────────────────────────────────────────────────────────────
-# Step 5: agents-update — AGENTS.md watermark staleness check
+# Step 4: agents-update — AGENTS.md watermark staleness check
 # ─────────────────────────────────────────────────────────────────────

 [[steps]]
@ -254,7 +220,7 @@ This keeps documentation fresh — runs 2x/day so drift stays small.
 CRITICAL: If this step fails for any reason, log the failure and move on.
 Do NOT let an AGENTS.md failure prevent the commit-and-pr step.
 """
-needs = ["ci-escalation-recipes"]
+needs = ["blocked-review"]

 # ─────────────────────────────────────────────────────────────────────
 # Step 6: commit-and-pr — single commit with all file changes
--- a/gardener/gardener-poll.sh
+++ b/gardener/gardener-poll.sh
@ -25,16 +25,8 @@ set -euo pipefail
 SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
 FACTORY_ROOT="$(dirname "$SCRIPT_DIR")"

-# --recipes-only: skip grooming (used by formula ci-escalation-recipes step
-# to avoid double-running grooming which the formula handles as its own step)
-RECIPES_ONLY=0
-if [ "${1:-}" = "--recipes-only" ]; then
-  RECIPES_ONLY=1
-  shift
-fi
-
 # Load shared environment (with optional project TOML override)
-# Usage: gardener-poll.sh [--recipes-only] [projects/harb.toml]
+# Usage: gardener-poll.sh [projects/harb.toml]
 export PROJECT_TOML="${1:-}"
 # shellcheck source=../lib/env.sh
 source "$FACTORY_ROOT/lib/env.sh"
@ -116,684 +108,8 @@ Instructions:
 done

 # ── Backlog grooming (delegated to gardener-agent.sh) ────────────────────
-# Skipped with --recipes-only (formula's grooming step handles this)
-if [ "$RECIPES_ONLY" -eq 0 ]; then
-  log "Invoking gardener-agent.sh for backlog grooming"
-  bash "$SCRIPT_DIR/gardener-agent.sh" "${1:-}" || log "WARNING: gardener-agent.sh exited with error"
-else
-  log "Skipping grooming (--recipes-only mode)"
-fi
+log "Invoking gardener-agent.sh for backlog grooming"
+bash "$SCRIPT_DIR/gardener-agent.sh" "${1:-}" || log "WARNING: gardener-agent.sh exited with error"


-# ── Recipe matching engine ────────────────────────────────────────────────
-RECIPE_DIR="$SCRIPT_DIR/recipes"
-
-# match_recipe — Find first matching recipe for escalation context
-# Args: $1=step_names_json  $2=output_file_path  $3=pr_info_json
-# Stdout: JSON {name, playbook} — "generic" fallback if no match
-match_recipe() {
-  _mr_stderr=$(mktemp /tmp/recipe-match-err-XXXXXX)
-  _mr_result=$(RECIPE_DIR="$RECIPE_DIR" python3 - "$1" "$2" "$3" 2>"$_mr_stderr" <<'PYEOF'
-import sys, os, re, json, glob
-try:
-    import tomllib
-except ModuleNotFoundError:
-    import tomli as tomllib  # Python < 3.11 fallback (pip install tomli)
-
-recipe_dir = os.environ["RECIPE_DIR"]
-recipes = []
-for path in sorted(glob.glob(os.path.join(recipe_dir, "*.toml"))):
-    with open(path, "rb") as f:
-        recipes.append(tomllib.load(f))
-
-recipes.sort(key=lambda r: r.get("priority", 50))
-
-step_names = json.loads(sys.argv[1])
-output_path = sys.argv[2]
-pr_info = json.loads(sys.argv[3])
-
-step_output = ""
-if os.path.isfile(output_path):
-    with open(output_path) as f:
-        step_output = f.read()
-
-for recipe in recipes:
-    trigger = recipe.get("trigger", {})
-    matched = True
-
-    if matched and "step_name" in trigger:
-        if not any(re.search(trigger["step_name"], n) for n in step_names):
-            matched = False
-
-    if matched and "output" in trigger:
-        if not re.search(trigger["output"], step_output):
-            matched = False
-
-    if matched and "pr_mergeable" in trigger:
-        if pr_info.get("mergeable") != trigger["pr_mergeable"]:
-            matched = False
-
-    if matched and "pr_files" in trigger:
-        changed = pr_info.get("changed_files", [])
-        if not any(re.search(trigger["pr_files"], f) for f in changed):
-            matched = False
-
-    if matched and "min_attempts" in trigger:
-        if pr_info.get("attempts", 1) < trigger["min_attempts"]:
-            matched = False
-
-    if matched and trigger.get("failures_on_unchanged"):
-        # Check if errors reference files NOT changed in the PR
-        # Patterns: ShellCheck "In file.sh line 5:", generic "file.sh:5:10: error",
-        #           ESLint/pylint "file.py:10:5: E123", Go "file.go:5:3:"
-        error_files = set()
-        error_files.update(re.findall(r"(?<=In )\S+(?= line \d+:)", step_output))
-        error_files.update(re.findall(r"^(\S+\.\w+):\d+", step_output, re.MULTILINE))
-        changed = set(pr_info.get("changed_files", []))
-        if not error_files or error_files <= changed:
-            matched = False
-
-    if matched:
-        print(json.dumps({"name": recipe["name"], "playbook": recipe.get("playbook", [])}))
-        sys.exit(0)
-
-print(json.dumps({"name": "generic", "playbook": [{"action": "create-generic-issue"}]}))
-PYEOF
-) || true
-  if [ -s "$_mr_stderr" ]; then
-    log "WARNING: match_recipe error: $(head -3 "$_mr_stderr" | tr '\n' ' ')"
-  fi
-  rm -f "$_mr_stderr"
-  if [ -z "$_mr_result" ] || ! echo "$_mr_result" | jq -e '.name' >/dev/null 2>&1; then
-    echo '{"name":"generic","playbook":[{"action":"create-generic-issue"}]}'
-  else
-    echo "$_mr_result"
-  fi
-}
-
-# ── Playbook action functions ────────────────────────────────────────────
-# Globals used by playbook functions (set by escalation loop):
-#   ESC_ISSUE, ESC_PR, ESC_ATTEMPTS, ESC_PIPELINE — escalation context
-#   _PB_FAILED_STEPS — "pid\tname" per line of failed CI steps
-#   _PB_LOG_DIR — temp dir with step-{pid}.log files
-#   _PB_SUB_CREATED — sub-issue counter for current escalation
-#   _esc_total_created — running total across all escalations
-
-# Create per-file ShellCheck sub-issues from CI output
-playbook_shellcheck_per_file() {
-  local step_pid step_name step_log_file step_logs
-  while IFS=$'\t' read -r step_pid step_name; do
-    [ -z "$step_pid" ] && continue
-    echo "$step_name" | grep -qi "shellcheck" || continue
-    step_log_file="${_PB_LOG_DIR}/step-${step_pid}.log"
-    [ -f "$step_log_file" ] || continue
-    step_logs=$(cat "$step_log_file")
-
-    local sc_files
-    sc_files=$(echo "$step_logs" | grep -oP '(?<=In )\S+(?= line \d+:)' | sort -u || true)
-
-    local sc_file file_errors sc_codes sub_title sub_body new_issue
-    while IFS= read -r sc_file; do
-      [ -z "$sc_file" ] && continue
-      # grep -F for literal filename match (dots in filenames are regex wildcards)
-      file_errors=$(echo "$step_logs" | grep -F -A3 "In ${sc_file} line" | head -30)
-      # SC codes only from this file's errors, not the whole step log
-      sc_codes=$(echo "$file_errors" | grep -oP 'SC\d+' | sort -u | tr '\n' ' ' | sed 's/ $//' || true)
-
-      sub_title="fix: ShellCheck errors in ${sc_file} (from PR #${ESC_PR})"
-      sub_body="## ShellCheck CI failure — \`${sc_file}\`
-
-Spawned by gardener from escalated issue #${ESC_ISSUE} (PR #${ESC_PR} failed CI after ${ESC_ATTEMPTS} attempt(s)).
-
-### Errors
-\`\`\`
-${file_errors}
-\`\`\`
-
-Fix all ShellCheck errors${sc_codes:+ (${sc_codes})} in \`${sc_file}\` so PR #${ESC_PR} CI passes.
-
-### Context
- Parent issue: #${ESC_ISSUE}
- PR: #${ESC_PR}
- Pipeline: #${ESC_PIPELINE} (step: ${step_name})"
-
-      new_issue=$(curl -sf -X POST \
-        -H "Authorization: token ${CODEBERG_TOKEN}" \
-        -H "Content-Type: application/json" \
-        "${CODEBERG_API}/issues" \
-        -d "$(jq -nc --arg t "$sub_title" --arg b "$sub_body" \
-          --argjson lid "$BACKLOG_LABEL_ID" '{"title":$t,"body":$b,"labels":[$lid]}')" 2>/dev/null | jq -r '.number // ""') || true
-
-      if [ -n "$new_issue" ]; then
-        log "Created sub-issue #${new_issue}: ShellCheck in ${sc_file} (from #${ESC_ISSUE})"
-        _PB_SUB_CREATED=$((_PB_SUB_CREATED + 1))
-        _esc_total_created=$((_esc_total_created + 1))
-        matrix_send "gardener" "📋 Created sub-issue #${new_issue}: ShellCheck in ${sc_file} (from escalated #${ESC_ISSUE})" 2>/dev/null || true
-      fi
-    done <<< "$sc_files"
-  done <<< "$_PB_FAILED_STEPS"
-}
-
-# Create per-file issues from any lint/check CI output (generic — no step name filter)
-playbook_lint_per_file() {
-  local step_pid step_name step_log_file step_logs
-  while IFS=$'\t' read -r step_pid step_name; do
-    [ -z "$step_pid" ] && continue
-    step_log_file="${_PB_LOG_DIR}/step-${step_pid}.log"
-    [ -f "$step_log_file" ] || continue
-    step_logs=$(cat "$step_log_file")
-
-    # Extract unique file paths from lint output (multiple formats):
-    #   ShellCheck: "In file.sh line 5:"
-    #   Generic:    "file.sh:5:10: error"
-    local lint_files
-    lint_files=$( {
-      echo "$step_logs" | grep -oP '(?<=In )\S+(?= line \d+:)' || true
-      echo "$step_logs" | grep -oP '^\S+\.\w+(?=:\d+)' || true
-    } | sort -u)
-
-    local lint_file file_errors sc_codes sub_title sub_body new_issue
-    while IFS= read -r lint_file; do
-      [ -z "$lint_file" ] && continue
-      # Extract errors for this file (try both formats)
-      file_errors=$(echo "$step_logs" | grep -F -A3 "In ${lint_file} line" 2>/dev/null | head -30 || true)
-      if [ -z "$file_errors" ]; then
-        file_errors=$(echo "$step_logs" | grep -F "${lint_file}:" | head -30 || true)
-      fi
-      [ -z "$file_errors" ] && continue
-      # Extract SC codes if present (harmless for non-ShellCheck output)
-      sc_codes=$(echo "$file_errors" | grep -oP 'SC\d+' | sort -u | tr '\n' ' ' | sed 's/ $//' || true)
-
-      sub_title="fix: lint errors in ${lint_file} (from PR #${ESC_PR})"
-      sub_body="## Lint CI failure — \`${lint_file}\`
-
-Spawned by gardener from escalated issue #${ESC_ISSUE} (PR #${ESC_PR} failed CI after ${ESC_ATTEMPTS} attempt(s)).
-
-### Errors
-\`\`\`
-${file_errors}
-\`\`\`
-
-Fix all errors${sc_codes:+ (${sc_codes})} in \`${lint_file}\` so PR #${ESC_PR} CI passes.
-
-### Context
- Parent issue: #${ESC_ISSUE}
- PR: #${ESC_PR}
- Pipeline: #${ESC_PIPELINE} (step: ${step_name})"
-
-      new_issue=$(curl -sf -X POST \
-        -H "Authorization: token ${CODEBERG_TOKEN}" \
-        -H "Content-Type: application/json" \
-        "${CODEBERG_API}/issues" \
-        -d "$(jq -nc --arg t "$sub_title" --arg b "$sub_body" \
-          --argjson lid "$BACKLOG_LABEL_ID" '{"title":$t,"body":$b,"labels":[$lid]}')" 2>/dev/null | jq -r '.number // ""') || true
-
-      if [ -n "$new_issue" ]; then
-        log "Created sub-issue #${new_issue}: lint in ${lint_file} (from #${ESC_ISSUE})"
-        _PB_SUB_CREATED=$((_PB_SUB_CREATED + 1))
-        _esc_total_created=$((_esc_total_created + 1))
-        matrix_send "gardener" "📋 Created sub-issue #${new_issue}: lint in ${lint_file} (from escalated #${ESC_ISSUE})" 2>/dev/null || true
-      fi
-    done <<< "$lint_files"
-  done <<< "$_PB_FAILED_STEPS"
-}
-
-# Create one combined issue for non-ShellCheck CI failures
-playbook_create_generic_issue() {
-  local generic_fail="" step_pid step_name step_log_file step_logs esc_section
-  while IFS=$'\t' read -r step_pid step_name; do
-    [ -z "$step_pid" ] && continue
-    # Skip shellcheck steps (handled by shellcheck-per-file action)
-    echo "$step_name" | grep -qi "shellcheck" && continue
-    step_log_file="${_PB_LOG_DIR}/step-${step_pid}.log"
-    [ -f "$step_log_file" ] || continue
-    step_logs=$(cat "$step_log_file")
-
-    esc_section="=== ${step_name} ===
-$(echo "$step_logs" | tail -50)"
-    if [ -z "$generic_fail" ]; then
-      generic_fail="$esc_section"
-    else
-      generic_fail="${generic_fail}
-${esc_section}"
-    fi
-  done <<< "$_PB_FAILED_STEPS"
-
-  [ -z "$generic_fail" ] && return 0
-
-  local sub_title sub_body new_issue
-  sub_title="fix: CI failures in PR #${ESC_PR} (from issue #${ESC_ISSUE})"
-  sub_body="## CI failure — fix required
-
-Spawned by gardener from escalated issue #${ESC_ISSUE} (PR #${ESC_PR} failed CI after ${ESC_ATTEMPTS} attempt(s)).
-
-### Failed step output
-\`\`\`
-${generic_fail}
-\`\`\`
-
-### Context
- Parent issue: #${ESC_ISSUE}
- PR: #${ESC_PR}${ESC_PIPELINE:+
- Pipeline: #${ESC_PIPELINE}}"
-
-  new_issue=$(curl -sf -X POST \
-    -H "Authorization: token ${CODEBERG_TOKEN}" \
-    -H "Content-Type: application/json" \
-    "${CODEBERG_API}/issues" \
-    -d "$(jq -nc --arg t "$sub_title" --arg b "$sub_body" \
-      --argjson lid "$BACKLOG_LABEL_ID" '{"title":$t,"body":$b,"labels":[$lid]}')" 2>/dev/null | jq -r '.number // ""') || true
-
-  if [ -n "$new_issue" ]; then
-    log "Created sub-issue #${new_issue}: CI failures for PR #${ESC_PR} (from #${ESC_ISSUE})"
-    _PB_SUB_CREATED=$((_PB_SUB_CREATED + 1))
-    _esc_total_created=$((_esc_total_created + 1))
-    matrix_send "gardener" "📋 Created sub-issue #${new_issue}: CI failures for PR #${ESC_PR} (from escalated #${ESC_ISSUE})" 2>/dev/null || true
-  fi
-}
-
-# Create issue to make failing CI step non-blocking (chicken-egg-ci)
-playbook_make_step_non_blocking() {
-  local failing_steps sub_title sub_body new_issue
-  failing_steps=$(echo "$_PB_FAILED_STEPS" | cut -f2 | tr '\n' ', ' | sed 's/,$//' || true)
-
-  sub_title="fix: make CI step non-blocking for pre-existing failures (PR #${ESC_PR})"
-  sub_body="## Chicken-egg CI failure
-
-PR #${ESC_PR} (issue #${ESC_ISSUE}) introduces a CI step that fails on pre-existing code.
-
-Failing step(s): ${failing_steps}
-
-### Playbook
-1. Add \`|| true\` to the failing step(s) in the Woodpecker config
-2. This makes the step advisory (non-blocking) until pre-existing violations are fixed
-
-### Context
- Parent issue: #${ESC_ISSUE}
- PR: #${ESC_PR}${ESC_PIPELINE:+
- Pipeline: #${ESC_PIPELINE}}"
-
-  new_issue=$(curl -sf -X POST \
-    -H "Authorization: token ${CODEBERG_TOKEN}" \
-    -H "Content-Type: application/json" \
-    "${CODEBERG_API}/issues" \
-    -d "$(jq -nc --arg t "$sub_title" --arg b "$sub_body" \
-      --argjson lid "$BACKLOG_LABEL_ID" '{"title":$t,"body":$b,"labels":[$lid]}')" 2>/dev/null | jq -r '.number // ""') || true
-
-  if [ -n "$new_issue" ]; then
-    log "Created #${new_issue}: make step non-blocking (chicken-egg from #${ESC_ISSUE})"
-    _PB_SUB_CREATED=$((_PB_SUB_CREATED + 1))
-    _esc_total_created=$((_esc_total_created + 1))
-    matrix_send "gardener" "📋 Created #${new_issue}: make CI step non-blocking (chicken-egg, from #${ESC_ISSUE})" 2>/dev/null || true
-  fi
-}
-
-# Create follow-up issue to remove || true bypass (chicken-egg-ci)
-playbook_create_followup_remove_bypass() {
-  local sub_title sub_body new_issue
-  sub_title="fix: remove || true bypass once pre-existing violations are fixed (PR #${ESC_PR})"
-  sub_body="## Follow-up: remove CI bypass
-
-After all pre-existing violation issues from PR #${ESC_PR} are resolved, remove the \`|| true\` bypass from the CI step to make it blocking again.
-
-### Depends on
-All per-file fix issues created from escalated issue #${ESC_ISSUE}.
-
-### Context
- Parent issue: #${ESC_ISSUE}
- PR: #${ESC_PR}"
-
-  new_issue=$(curl -sf -X POST \
-    -H "Authorization: token ${CODEBERG_TOKEN}" \
-    -H "Content-Type: application/json" \
-    "${CODEBERG_API}/issues" \
-    -d "$(jq -nc --arg t "$sub_title" --arg b "$sub_body" \
-      --argjson lid "$BACKLOG_LABEL_ID" '{"title":$t,"body":$b,"labels":[$lid]}')" 2>/dev/null | jq -r '.number // ""') || true
-
-  if [ -n "$new_issue" ]; then
-    log "Created follow-up #${new_issue}: remove bypass (from #${ESC_ISSUE})"
-    _PB_SUB_CREATED=$((_PB_SUB_CREATED + 1))
-    _esc_total_created=$((_esc_total_created + 1))
-  fi
-}
-
-# Rebase PR onto main branch (cascade-rebase)
-playbook_rebase_pr() {
-  log "Rebasing PR #${ESC_PR} onto ${PRIMARY_BRANCH}"
-  local result
-  local http_code
-  http_code=$(curl -s -o /dev/null -w '%{http_code}' -X POST \
-    -H "Authorization: token ${CODEBERG_TOKEN}" \
-    -H "Content-Type: application/json" \
-    "${CODEBERG_API}/pulls/${ESC_PR}/update" \
-    -d '{"style":"rebase"}' 2>/dev/null) || true
-
-  if [ "${http_code:-0}" -ge 200 ] && [ "${http_code:-0}" -lt 300 ]; then
-    log "Rebase initiated for PR #${ESC_PR} (HTTP ${http_code})"
-    _PB_SUB_CREATED=$((_PB_SUB_CREATED + 1))
-    matrix_send "gardener" "🔄 Rebased PR #${ESC_PR} onto ${PRIMARY_BRANCH} (cascade-rebase, from #${ESC_ISSUE})" 2>/dev/null || true
-  else
-    log "WARNING: rebase API call failed for PR #${ESC_PR} (HTTP ${http_code:-error})"
-  fi
-}
-
-# Re-approve PR if review was dismissed by force-push (cascade-rebase)
-playbook_re_approve_if_dismissed() {
-  local reviews dismissed
-  reviews=$(curl -sf -H "Authorization: token ${CODEBERG_TOKEN}" \
-    "${CODEBERG_API}/pulls/${ESC_PR}/reviews" 2>/dev/null || true)
-  [ -z "$reviews" ] || [ "$reviews" = "null" ] && return 0
-
-  dismissed=$(echo "$reviews" | jq -r '[.[] | select(.state == "APPROVED" and .dismissed == true)] | length' 2>/dev/null || true)
-  if [ "${dismissed:-0}" -gt 0 ]; then
-    curl -sf -X POST \
-      -H "Authorization: token ${CODEBERG_TOKEN}" \
-      -H "Content-Type: application/json" \
-      "${CODEBERG_API}/pulls/${ESC_PR}/reviews" \
-      -d '{"event":"APPROVED","body":"Re-approved after rebase (cascade-rebase recipe)"}' 2>/dev/null || true
-    log "Re-approved PR #${ESC_PR} after rebase"
-    _PB_SUB_CREATED=$((_PB_SUB_CREATED + 1))
-  fi
-}
-
-# Retry merging the PR (cascade-rebase)
-playbook_retry_merge() {
-  local result
-  result=$(curl -sf -X POST \
-    -H "Authorization: token ${CODEBERG_TOKEN}" \
-    -H "Content-Type: application/json" \
-    "${CODEBERG_API}/pulls/${ESC_PR}/merge" \
-    -d '{"Do":"rebase","delete_branch_after_merge":true}' 2>/dev/null) || true
-
-  if [ -n "$result" ]; then
-    log "Merge retry initiated for PR #${ESC_PR}"
-    _PB_SUB_CREATED=$((_PB_SUB_CREATED + 1))
-    matrix_send "gardener" "✅ Merge retry for PR #${ESC_PR} (cascade-rebase, from #${ESC_ISSUE})" 2>/dev/null || true
-  else
-    log "WARNING: merge retry failed for PR #${ESC_PR}"
-  fi
-}
-
-# Retrigger CI pipeline (flaky-test)
-playbook_retrigger_ci() {
-  [ -z "$ESC_PIPELINE" ] && return 0
-  # Max 2 retriggers per issue spec
-  if [ "${ESC_ATTEMPTS:-1}" -ge 3 ]; then
-    log "Max retriggers reached for pipeline #${ESC_PIPELINE} (${ESC_ATTEMPTS} attempts)"
-    return 0
-  fi
-  log "Retriggering CI pipeline #${ESC_PIPELINE} (attempt ${ESC_ATTEMPTS})"
-  local http_code
-  http_code=$(curl -s -o /dev/null -w '%{http_code}' -X POST \
-    -H "Authorization: Bearer ${WOODPECKER_TOKEN}" \
-    "${WOODPECKER_SERVER}/api/repos/${WOODPECKER_REPO_ID}/pipelines/${ESC_PIPELINE}" 2>/dev/null) || true
-
-  if [ "${http_code:-0}" -ge 200 ] && [ "${http_code:-0}" -lt 300 ]; then
-    log "Pipeline #${ESC_PIPELINE} retriggered (HTTP ${http_code})"
-    _PB_SUB_CREATED=$((_PB_SUB_CREATED + 1))
-    matrix_send "gardener" "🔄 Retriggered CI for PR #${ESC_PR} (flaky-test, attempt ${ESC_ATTEMPTS})" 2>/dev/null || true
-  else
-    log "WARNING: retrigger failed for pipeline #${ESC_PIPELINE} (HTTP ${http_code:-error})"
-  fi
-}
-
-# Quarantine flaky test and create fix issue (flaky-test)
-playbook_quarantine_test() {
-  # Only quarantine if retriggers exhausted
-  if [ "${ESC_ATTEMPTS:-1}" -lt 3 ]; then
-    return 0
-  fi
-
-  local failing_steps sub_title sub_body new_issue
-  failing_steps=$(echo "$_PB_FAILED_STEPS" | cut -f2 | tr '\n' ', ' | sed 's/,$//' || true)
-
-  sub_title="fix: quarantine flaky test (PR #${ESC_PR}, from #${ESC_ISSUE})"
-  sub_body="## Flaky test detected
-
-CI for PR #${ESC_PR} (issue #${ESC_ISSUE}) failed intermittently across ${ESC_ATTEMPTS} attempts.
-
-Failing step(s): ${failing_steps:-unknown}
-
-### Playbook
-1. Identify the flaky test(s) from CI output
-2. Quarantine (skip/mark pending) the flaky test(s)
-3. Create targeted fix for the root cause
-
-### Context
- Parent issue: #${ESC_ISSUE}
- PR: #${ESC_PR}${ESC_PIPELINE:+
- Pipeline: #${ESC_PIPELINE}}"
-
-  new_issue=$(curl -sf -X POST \
-    -H "Authorization: token ${CODEBERG_TOKEN}" \
-    -H "Content-Type: application/json" \
-    "${CODEBERG_API}/issues" \
-    -d "$(jq -nc --arg t "$sub_title" --arg b "$sub_body" \
-      --argjson lid "$BACKLOG_LABEL_ID" '{"title":$t,"body":$b,"labels":[$lid]}')" 2>/dev/null | jq -r '.number // ""') || true
-
-  if [ -n "$new_issue" ]; then
-    log "Created quarantine issue #${new_issue} for flaky test (from #${ESC_ISSUE})"
-    _PB_SUB_CREATED=$((_PB_SUB_CREATED + 1))
-    _esc_total_created=$((_esc_total_created + 1))
-    matrix_send "gardener" "📋 Created #${new_issue}: quarantine flaky test (from #${ESC_ISSUE})" 2>/dev/null || true
-  fi
-}
-
-# run_playbook — Execute matched recipe's playbook actions
-# Args: $1=recipe_json from match_recipe
-run_playbook() {
-  local recipe_json="$1"
-  local recipe_name actions action
-  recipe_name=$(echo "$recipe_json" | jq -r '.name')
-  actions=$(echo "$recipe_json" | jq -r '.playbook[].action' 2>/dev/null || true)
-
-  while IFS= read -r action; do
-    [ -z "$action" ] && continue
-    case "$action" in
-      shellcheck-per-file)           playbook_shellcheck_per_file ;;
-      lint-per-file)                 playbook_lint_per_file ;;
-      create-generic-issue)          playbook_create_generic_issue ;;
-      make-step-non-blocking)        playbook_make_step_non_blocking ;;
-      create-followup-remove-bypass) playbook_create_followup_remove_bypass ;;
-      rebase-pr)                     playbook_rebase_pr ;;
-      re-approve-if-dismissed)       playbook_re_approve_if_dismissed ;;
-      retry-merge)                   playbook_retry_merge ;;
-      retrigger-ci)                  playbook_retrigger_ci ;;
-      quarantine-test)               playbook_quarantine_test ;;
-      label-backlog)                 ;; # default label, no-op (issues created with backlog)
-      *)                             log "WARNING: unknown playbook action '${action}' in recipe '${recipe_name}'" ;;
-    esac
-  done <<< "$actions"
-}
-
-# ── Process dev-agent escalations (per-project, recipe-driven) ───────────
-ESCALATION_FILE="${FACTORY_ROOT}/supervisor/escalations-${PROJECT_NAME}.jsonl"
-ESCALATION_DONE="${FACTORY_ROOT}/supervisor/escalations-${PROJECT_NAME}.done.jsonl"
-
-if [ -s "$ESCALATION_FILE" ]; then
-  # Atomically snapshot the file before processing to prevent race with
-  # concurrent dev-poll appends: new entries go to a fresh ESCALATION_FILE
-  # while we process the snapshot, so nothing is ever silently dropped.
-  ESCALATION_SNAP="${ESCALATION_FILE}.processing.$$"
-  mv "$ESCALATION_FILE" "$ESCALATION_SNAP"
-
-  ESCALATION_COUNT=$(wc -l < "$ESCALATION_SNAP")
-  log "Processing ${ESCALATION_COUNT} escalation(s) for ${PROJECT_NAME}"
-  _esc_total_created=0
-
-  while IFS= read -r esc_entry; do
-    [ -z "$esc_entry" ] && continue
-
-    ESC_ISSUE=$(echo "$esc_entry" | jq -r '.issue // empty')
-    ESC_PR=$(echo "$esc_entry" | jq -r '.pr // empty')
-    ESC_ATTEMPTS=$(echo "$esc_entry" | jq -r '.attempts // 3')
-    ESC_REASON=$(echo "$esc_entry" | jq -r '.reason // empty')
-
-    if [ -z "$ESC_ISSUE" ] || [ -z "$ESC_PR" ]; then
-      echo "$esc_entry" >> "$ESCALATION_DONE"
-      continue
-    fi
-
-    log "Escalation: issue #${ESC_ISSUE} PR #${ESC_PR} reason=${ESC_REASON} (${ESC_ATTEMPTS} CI attempt(s))"
-
-    # Handle idle_timeout / idle_prompt escalations — no CI steps to inspect, just notify
-    if [[ "$ESC_REASON" == idle_timeout* || "$ESC_REASON" == idle_prompt* ]]; then
-      _issue_url="https://codeberg.org/${CODEBERG_REPO}/issues/${ESC_ISSUE}"
-      if [[ "$ESC_REASON" == idle_prompt* ]]; then
-        sub_title="chore: investigate idle prompt for issue #${ESC_ISSUE}"
-        sub_body="## Dev-agent idle prompt
-
-The dev-agent session for issue #${ESC_ISSUE} returned to the prompt without writing a phase signal.$([ "${ESC_PR:-0}" != "0" ] && printf '\n\nPR #%s may still be open.' "$ESC_PR")
-
-### What to check
-1. Did Claude finish without signalling a phase? Check for missing phase-file writes.
-2. Was the issue spec ambiguous or missing acceptance criteria?
-3. Re-run the issue by restoring the \`backlog\` label if the spec is clear.
-
-### Context
- Issue: [#${ESC_ISSUE}](${_issue_url})$([ "${ESC_PR:-0}" != "0" ] && printf '\n- PR: #%s' "$ESC_PR")"
-      else
-        sub_title="chore: investigate idle timeout for issue #${ESC_ISSUE}"
-        sub_body="## Dev-agent idle timeout
-
-The dev-agent session for issue #${ESC_ISSUE} was idle for 2h without a phase update and was killed.$([ "${ESC_PR:-0}" != "0" ] && printf '\n\nPR #%s may still be open.' "$ESC_PR")
-
-### What to check
-1. Was the agent stuck waiting for input? Check the issue spec for ambiguity.
-2. Was there an infrastructure issue (tmux crash, disk full, etc.)?
-3. Re-run the issue by restoring the \`backlog\` label if the spec is clear.
-
-### Context
- Issue: [#${ESC_ISSUE}](${_issue_url})$([ "${ESC_PR:-0}" != "0" ] && printf '\n- PR: #%s' "$ESC_PR")"
-      fi
-
-      new_issue=$(curl -sf -X POST \
-        -H "Authorization: token ${CODEBERG_TOKEN}" \
-        -H "Content-Type: application/json" \
-        "${CODEBERG_API}/issues" \
-        -d "$(jq -nc --arg t "$sub_title" --arg b "$sub_body" \
-          --argjson lid "$BACKLOG_LABEL_ID" '{"title":$t,"body":$b,"labels":[$lid]}')" 2>/dev/null | jq -r '.number // ""') || true
-
-      if [ -n "$new_issue" ]; then
-        log "Created idle sub-issue #${new_issue} for #${ESC_ISSUE} (${ESC_REASON})"
-        _esc_total_created=$((_esc_total_created + 1))
-        matrix_send "gardener" "⏱ Created #${new_issue}: ${ESC_REASON} on #${ESC_ISSUE}" 2>/dev/null || true
-      fi
-
-      echo "$esc_entry" >> "$ESCALATION_DONE"
-      continue
-    fi
-
-    # Fetch PR metadata (SHA, mergeable status)
-    ESC_PR_DATA=$(curl -sf -H "Authorization: token ${CODEBERG_TOKEN}" \
-      "${CODEBERG_API}/pulls/${ESC_PR}" 2>/dev/null || true)
-    ESC_PR_SHA=$(echo "$ESC_PR_DATA" | jq -r '.head.sha // ""' 2>/dev/null || true)
-    _PB_PR_MERGEABLE=$(echo "$ESC_PR_DATA" | jq '.mergeable // null' 2>/dev/null || true)
-
-    ESC_PIPELINE=""
-    if [ -n "$ESC_PR_SHA" ]; then
-      # Validate SHA is a 40-char hex string before interpolating into SQL
-      if [[ "$ESC_PR_SHA" =~ ^[0-9a-fA-F]{40}$ ]]; then
-        ESC_PIPELINE=$(wpdb -c "SELECT number FROM pipelines WHERE repo_id=${WOODPECKER_REPO_ID} AND commit='${ESC_PR_SHA}' ORDER BY created DESC LIMIT 1;" 2>/dev/null | xargs || true)
-      else
-        log "WARNING: ESC_PR_SHA '${ESC_PR_SHA}' is not a valid hex SHA — skipping pipeline lookup"
-      fi
-    fi
-
-    # Fetch failed CI steps and their logs into temp dir
-    _PB_FAILED_STEPS=""
-    _PB_LOG_DIR=$(mktemp -d /tmp/recipe-logs-XXXXXX)
-    _PB_SUB_CREATED=0
-    _PB_LOGS_AVAILABLE=0
-
-    if [ -n "$ESC_PIPELINE" ]; then
-      _PB_FAILED_STEPS=$(curl -sf \
-        -H "Authorization: Bearer ${WOODPECKER_TOKEN}" \
-        "${WOODPECKER_SERVER}/api/repos/${WOODPECKER_REPO_ID}/pipelines/${ESC_PIPELINE}" 2>/dev/null | \
-        jq -r '.workflows[]?.children[]? | select(.state=="failure") | "\(.pid)\t\(.name)"' 2>/dev/null || true)
-
-      while IFS=$'\t' read -r step_pid step_name; do
-        [ -z "$step_pid" ] && continue
-        [[ "$step_pid" =~ ^[0-9]+$ ]] || { log "WARNING: invalid step_pid '${step_pid}' — skipping"; continue; }
-        step_logs=$(woodpecker-cli pipeline log show "${CODEBERG_REPO}" "${ESC_PIPELINE}" "${step_pid}" 2>/dev/null | tail -150 || true)
-        if [ -n "$step_logs" ]; then
-          echo "$step_logs" > "${_PB_LOG_DIR}/step-${step_pid}.log"
-          _PB_LOGS_AVAILABLE=1
-        fi
-      done <<< "$_PB_FAILED_STEPS"
-    fi
-
-    # Fetch PR changed files for recipe matching
-    _PB_PR_FILES_JSON="[]"
-    _PB_PR_FILES=$(curl -sf -H "Authorization: token ${CODEBERG_TOKEN}" \
-      "${CODEBERG_API}/pulls/${ESC_PR}/files" 2>/dev/null | jq -r '.[].filename // empty' 2>/dev/null || true)
-    if [ -n "$_PB_PR_FILES" ]; then
-      _PB_PR_FILES_JSON=$(echo "$_PB_PR_FILES" | jq -Rsc 'split("\n") | map(select(length > 0))')
-    fi
-
-    # Build recipe matching context
-    _RECIPE_STEP_NAMES=$(echo "$_PB_FAILED_STEPS" | cut -f2 | jq -Rsc 'split("\n") | map(select(length > 0))')
-    _RECIPE_OUTPUT_FILE="${_PB_LOG_DIR}/all-output.txt"
-    cat "${_PB_LOG_DIR}"/step-*.log > "$_RECIPE_OUTPUT_FILE" 2>/dev/null || touch "$_RECIPE_OUTPUT_FILE"
-    _RECIPE_PR_INFO=$(jq -nc \
-      --argjson m "${_PB_PR_MERGEABLE:-null}" \
-      --argjson a "${ESC_ATTEMPTS}" \
-      --argjson files "${_PB_PR_FILES_JSON}" \
-      '{mergeable:$m, attempts:$a, changed_files:$files}')
-
-    # Match escalation against recipes and execute playbook
-    MATCHED_RECIPE=$(match_recipe "$_RECIPE_STEP_NAMES" "$_RECIPE_OUTPUT_FILE" "$_RECIPE_PR_INFO")
-    RECIPE_NAME=$(echo "$MATCHED_RECIPE" | jq -r '.name')
-    log "Recipe matched: ${RECIPE_NAME} for #${ESC_ISSUE} PR #${ESC_PR}"
-
-    run_playbook "$MATCHED_RECIPE"
-
-    # Fallback: no sub-issues created — create investigation issue
-    if [ "$_PB_SUB_CREATED" -eq 0 ]; then
-      sub_title="fix: investigate CI failure for PR #${ESC_PR} (from issue #${ESC_ISSUE})"
-      if [ "$_PB_LOGS_AVAILABLE" -eq 1 ]; then
-        sub_body="## CI failure — investigation required
-
-Spawned by gardener from escalated issue #${ESC_ISSUE} (PR #${ESC_PR} failed CI after ${ESC_ATTEMPTS} attempt(s)). Recipe '${RECIPE_NAME}' matched but produced no sub-issues.
-
-Check PR #${ESC_PR} CI output, identify the failing checks, and fix them so the PR can merge."
-      else
-        sub_body="## CI failure — investigation required
-
-Spawned by gardener from escalated issue #${ESC_ISSUE} (PR #${ESC_PR} failed CI after ${ESC_ATTEMPTS} attempt(s)). CI logs were unavailable at escalation time.
-
-Check PR #${ESC_PR} CI output, identify the failing checks, and fix them so the PR can merge."
-      fi
-
-      new_issue=$(curl -sf -X POST \
-        -H "Authorization: token ${CODEBERG_TOKEN}" \
-        -H "Content-Type: application/json" \
-        "${CODEBERG_API}/issues" \
-        -d "$(jq -nc --arg t "$sub_title" --arg b "$sub_body" \
-          --argjson lid "$BACKLOG_LABEL_ID" '{"title":$t,"body":$b,"labels":[$lid]}')" 2>/dev/null | jq -r '.number // ""') || true
-
-      if [ -n "$new_issue" ]; then
-        log "Created fallback sub-issue #${new_issue} for escalated #${ESC_ISSUE}"
-        _esc_total_created=$((_esc_total_created + 1))
-        matrix_send "gardener" "📋 Created sub-issue #${new_issue}: investigate CI for PR #${ESC_PR} (from escalated #${ESC_ISSUE})" 2>/dev/null || true
-      fi
-    fi
-
-    # Cleanup temp files
-    rm -rf "$_PB_LOG_DIR"
-
-    # Mark as processed
-    echo "$esc_entry" >> "$ESCALATION_DONE"
-  done < "$ESCALATION_SNAP"
-
-  rm -f "$ESCALATION_SNAP"
-  log "Escalations processed — moved to $(basename "$ESCALATION_DONE")"
-
-  # Report resolution count to supervisor for its fixed() summary
-  if [ "${_esc_total_created:-0}" -gt 0 ]; then
-    printf '%d %s\n' "$_esc_total_created" "$PROJECT_NAME" \
-      >> "${FACTORY_ROOT}/supervisor/gardener-esc-resolved.log"
-  fi
-fi
-
 log "--- Gardener poll done ---"
--- a/supervisor/preflight.sh
+++ b/supervisor/preflight.sh
@ -158,21 +158,16 @@ done
 [ "$_found_wt" = false ] && echo "  None"
 echo ""

-# ── Pending Escalations ──────────────────────────────────────────────────
+# ── Blocked Issues ────────────────────────────────────────────────────────

-echo "## Pending Escalations"
-_found_esc=false
-for _esc_file in "${FACTORY_ROOT}/supervisor/escalations-"*.jsonl; do
-  [ -f "$_esc_file" ] || continue
-  [[ "$_esc_file" == *.done.jsonl ]] && continue
-  _esc_count=$(wc -l < "$_esc_file" 2>/dev/null || echo 0)
-  [ "${_esc_count:-0}" -gt 0 ] || continue
-  _found_esc=true
-  echo "### $(basename "$_esc_file") (${_esc_count} entries)"
-  cat "$_esc_file"
-  echo ""
-done
-[ "$_found_esc" = false ] && echo "  None"
+echo "## Blocked Issues"
+_blocked_issues=$(codeberg_api GET "/issues?state=open&labels=blocked&type=issues&limit=50" 2>/dev/null || echo "[]")
+_blocked_n=$(echo "$_blocked_issues" | jq 'length' 2>/dev/null || echo 0)
+if [ "${_blocked_n:-0}" -gt 0 ]; then
+  echo "$_blocked_issues" | jq -r '.[] | "  #\(.number): \(.title)"' 2>/dev/null || echo "  (query failed)"
+else
+  echo "  None"
+fi
 echo ""

 # ── Escalation Replies from Matrix ────────────────────────────────────────
--- a/supervisor/supervisor-poll.sh
+++ b/supervisor/supervisor-poll.sh
@ -231,26 +231,6 @@ for logfile in "${FACTORY_ROOT}"/{dev,review,supervisor}/*.log; do
  fi
 done

-# Report pending escalations (processing has moved to gardener-poll.sh per-project)
-for _esc_file in "${FACTORY_ROOT}/supervisor/escalations-"*.jsonl; do
-  [ -f "$_esc_file" ] || continue
-  [[ "$_esc_file" == *.done.jsonl ]] && continue
-  _esc_count=$(wc -l < "$_esc_file" 2>/dev/null || true)
-  [ "${_esc_count:-0}" -gt 0 ] || continue
-  _esc_proj=$(basename "$_esc_file" .jsonl)
-  _esc_proj="${_esc_proj#escalations-}"
-  flog "${_esc_proj}: ${_esc_count} escalation(s) pending (gardener will process)"
-done
-
-# Pick up escalation resolutions handled by gardener
-_gesc_log="${FACTORY_ROOT}/supervisor/gardener-esc-resolved.log"
-if [ -f "$_gesc_log" ]; then
-  while IFS=' ' read -r _gn _gp; do
-    [ -n "${_gn:-}" ] && fixed "${_gp:-unknown}: gardener created ${_gn} sub-issue(s) from escalations"
-  done < "$_gesc_log"
-  rm -f "$_gesc_log"
-fi
-
 # #############################################################################
 #                      LAYER 2: PER-PROJECT CHECKS
 #               (iterated over projects/*.toml, config-driven)
@ -342,149 +322,6 @@ check_project() {
    find "$_RETRY_DIR" -type f -mmin +1440 -delete 2>/dev/null || true
  fi

-  # ===========================================================================
-  # P2f: ESCALATION TRIAGE — auto-retrigger ci_exhausted if infra-only
-  # ===========================================================================
-  if [ "${CHECK_INFRA_RETRY:-true}" = "true" ]; then
-    status "P2f: ${proj_name}: triaging ci_exhausted escalations"
-
-    _esc_file="${FACTORY_ROOT}/supervisor/escalations-${proj_name}.jsonl"
-    if [ -f "$_esc_file" ] && [ -s "$_esc_file" ]; then
-      _esc_tmp="${_esc_file}.sup.$$"
-      : > "$_esc_tmp"
-
-      while IFS= read -r _esc_line; do
-        [ -z "$_esc_line" ] && continue
-
-        _esc_reason=$(printf '%s' "$_esc_line" | jq -r '.reason // ""' 2>/dev/null)
-
-        # Only triage ci_exhausted entries (from dev-agent or dev-poll)
-        case "$_esc_reason" in
-          ci_exhausted)      ;;
-          ci_exhausted_poll) ;;
-          *) printf '%s\n' "$_esc_line" >> "$_esc_tmp"; continue ;;
-        esac
-
-        _esc_pr=$(printf '%s' "$_esc_line" | jq -r '.pr // 0' 2>/dev/null)
-        _esc_issue=$(printf '%s' "$_esc_line" | jq -r '.issue // 0' 2>/dev/null)
-        _esc_ts=$(printf '%s' "$_esc_line" | jq -r '.ts // ""' 2>/dev/null)
-
-        # Validate pr/issue are numeric
-        [[ "$_esc_pr" =~ ^[0-9]+$ ]] || { printf '%s\n' "$_esc_line" >> "$_esc_tmp"; continue; }
-        [[ "$_esc_issue" =~ ^[0-9]+$ ]] || { printf '%s\n' "$_esc_line" >> "$_esc_tmp"; continue; }
-
-        # Cooldown: 30 min from last escalation timestamp
-        _esc_epoch=0
-        [ -n "$_esc_ts" ] && _esc_epoch=$(date -d "$_esc_ts" +%s 2>/dev/null || echo 0)
-        _esc_age_min=$(( ($(date +%s) - _esc_epoch) / 60 ))
-
-        if [ "$_esc_age_min" -lt 30 ]; then
-          flog "${proj_name}: PR #${_esc_pr} ci_exhausted cooldown (${_esc_age_min}/30min)"
-          printf '%s\n' "$_esc_line" >> "$_esc_tmp"
-          continue
-        fi
-
-        # Get the PR's branch and state from Codeberg
-        _esc_pr_json=$(codeberg_api GET "/pulls/${_esc_pr}" 2>/dev/null) || {
-          flog "${proj_name}: PR #${_esc_pr}: failed to fetch PR info, keeping escalation"
-          printf '%s\n' "$_esc_line" >> "$_esc_tmp"; continue
-        }
-        _esc_pr_state=$(printf '%s' "$_esc_pr_json" | jq -r '.state // ""' 2>/dev/null)
-        if [ "$_esc_pr_state" != "open" ]; then
-          flog "${proj_name}: PR #${_esc_pr} is ${_esc_pr_state:-unknown} — discarding stale escalation"
-          continue  # PR merged/closed externally; escalation no longer actionable
-        fi
-        _esc_branch=$(printf '%s' "$_esc_pr_json" | jq -r '.head.ref // ""' 2>/dev/null)
-        if [ -z "$_esc_branch" ]; then
-          printf '%s\n' "$_esc_line" >> "$_esc_tmp"; continue
-        fi
-
-        # Validate branch name to prevent SQL injection
-        if ! [[ "$_esc_branch" =~ ^[a-zA-Z0-9/_.-]+$ ]]; then
-          flog "${proj_name}: PR #${_esc_pr}: unsafe branch name, keeping escalation"
-          printf '%s\n' "$_esc_line" >> "$_esc_tmp"; continue
-        fi
-
-        # Find the latest failed pipeline for this PR's branch via Woodpecker DB
-        _esc_pip=$(wpdb -A -c "
-          SELECT number FROM pipelines
-          WHERE repo_id = ${WOODPECKER_REPO_ID}
-            AND branch = '${_esc_branch}'
-            AND status IN ('failure', 'error')
-            AND finished > 0
-          ORDER BY number DESC LIMIT 1;" 2>/dev/null \
-          | tr -d ' ' | grep -E '^[0-9]+$' | head -1 || true)
-
-        if [ -z "$_esc_pip" ]; then
-          flog "${proj_name}: PR #${_esc_pr}: no failed pipeline for branch ${_esc_branch}, keeping escalation"
-          printf '%s\n' "$_esc_line" >> "$_esc_tmp"; continue
-        fi
-
-        # Classify failure type via ci-helpers
-        _esc_failure=$(classify_pipeline_failure "${WOODPECKER_REPO_ID}" "$_esc_pip" 2>/dev/null || echo "code")
-
-        if [[ "$_esc_failure" != infra* ]]; then
-          flog "${proj_name}: PR #${_esc_pr} pipeline #${_esc_pip}: code failure — leaving escalation for human"
-          printf '%s\n' "$_esc_line" >> "$_esc_tmp"; continue
-        fi
-
-        # Infra-only — push empty commit to retrigger CI via temporary worktree
-        _esc_wt="/tmp/${proj_name}-sup-retry-${_esc_pr}"
-        _esc_retrigger_ok=false
-        if [ -d "${PROJECT_REPO_ROOT:-}" ]; then
-          # Clean up any leftover temp worktree from a previous failed run
-          git -C "${PROJECT_REPO_ROOT}" worktree remove --force "${_esc_wt}" 2>/dev/null || true
-
-          if git -C "${PROJECT_REPO_ROOT}" fetch origin "${_esc_branch}" --quiet 2>/dev/null && \
-             git -C "${PROJECT_REPO_ROOT}" worktree add --quiet --detach \
-               "${_esc_wt}" "origin/${_esc_branch}" 2>/dev/null; then
-            if git -C "${_esc_wt}" \
-                 -c user.email="supervisor@factory" \
-                 -c user.name="Supervisor" \
-                 commit --allow-empty --no-verify \
-                 -m "chore: retrigger CI after infra-only exhaustion" \
-                 --quiet 2>/dev/null && \
-               git -C "${_esc_wt}" push origin \
-                 "HEAD:refs/heads/${_esc_branch}" --quiet 2>/dev/null; then
-              _esc_retrigger_ok=true
-            fi
-            git -C "${PROJECT_REPO_ROOT}" worktree remove --force "${_esc_wt}" 2>/dev/null || true
-          fi
-        fi
-
-        if [ "$_esc_retrigger_ok" = true ]; then
-          # Reset CI fix counter so dev-poll can spawn the agent again if needed
-          _ci_fix_file="${FACTORY_ROOT}/dev/ci-fixes-${proj_name}.json"
-          _ci_fix_lock="${_ci_fix_file}.lock"
-          flock "$_ci_fix_lock" python3 -c "
-import json, os
-f='${_ci_fix_file}'
-if not os.path.exists(f):
-    exit()
-d = json.load(open(f))
-d.pop(str(${_esc_pr}), None)
-json.dump(d, open(f, 'w'))
-" 2>/dev/null || true
-
-          fixed "${proj_name}: auto-retriggered CI for PR #${_esc_pr} after infra-only exhaustion"
-          flog "${proj_name}: auto-retriggered CI for PR #${_esc_pr} (issue #${_esc_issue}) after infra-only exhaustion"
-          matrix_send "supervisor" "♻️ auto-retriggered CI for PR #${_esc_pr} (issue #${_esc_issue}) after infra-only exhaustion" 2>/dev/null || true
-          # Escalation removed — do NOT write to _esc_tmp
-        else
-          p2 "${proj_name}: PR #${_esc_pr}: infra-only CI exhaustion but retrigger push failed"
-          # Bump timestamp to now so the 30-min cooldown resets; prevents alert flood
-          # on persistent push failures (SSH key issue, Codeberg outage, etc.)
-          _esc_now=$(date -u +%Y-%m-%dT%H:%M:%SZ)
-          _esc_bumped=$(printf '%s' "$_esc_line" | jq -c --arg ts "$_esc_now" '.ts = $ts' 2>/dev/null \
-            || printf '%s' "$_esc_line")
-          printf '%s\n' "$_esc_bumped" >> "$_esc_tmp"
-        fi
-      done < "$_esc_file"
-
-      mv "$_esc_tmp" "$_esc_file"
-    fi
-  fi
-
  # Dev-agent health (only if monitoring enabled)
  if [ "${CHECK_DEV_AGENT:-true}" = "true" ]; then
    DEV_LOCK="/tmp/dev-agent-${PROJECT_NAME}.lock"