crewly 1.20.355 → 1.20.358

package/config/constants.ts

@@ -1155,6 +1155,17 @@ export const AGENT_SUSPEND_CONSTANTS = {
 	 * paid that 50 times a day on steamfun-ops (2026-09-16).
 	 */
 	CRON_KEEPALIVE_WINDOW_MINUTES: 30,
+	/**
+	 * An idle member that is in a conversation with the owner (an owner thread
+	 * the sentinel watches: the owner wrote to it, or it promised the owner
+	 * something, in the last two hours) is kept at least this long before an
+	 * idle stop, whatever the configured or pressure timeout says. Under the
+	 * 10-minute pressure timeout the audit of 2026-10-16 counted 50 idle stops
+	 * in 9 h, many of them agents the owner was still talking to; every stop
+	 * costs a cold start and, past the fresh-start threshold, the conversation
+	 * (crewly#1344). Memory-pressure and critical stops still apply.
+	 */
+	OWNER_THREAD_IDLE_TIMEOUT_MINUTES: 30,
 } as const;
 
 // ========================= SERVER PROCESS CONSTANTS =========================

package/config/roles/team-leader/prompt.md

@@ -174,7 +174,7 @@ When you receive an Objective from the Orchestrator:
 1. Monitor worker status — check for Idle/Working/Error states
 2. If a worker requests information, retrieve it from the Knowledge Base or Orchestrator
 3. Use idle event subscriptions (auto-setup by delegate-task) to get notified when workers finish
-4. Periodic fallback checks ensure no task goes stale
+4. No recurring progress check is set up by default: the idle event and report-status wake you. Opt in with `--progress-check <minutes>` (minimum 30) only for a long task with no other signal
 
 ### Step 5: Result Verification (#140)
 **MANDATORY: When a worker reports task as done, you MUST verify before accepting.**

package/config/roles/team-leader/tl-addon.md

@@ -156,7 +156,7 @@ When you receive an Objective from the Orchestrator:
 1. Monitor worker status — check for Idle/Working/Error states
 2. If a worker requests information, retrieve it from the Knowledge Base or Orchestrator
 3. Use idle event subscriptions (auto-setup by delegate-task) to get notified when workers finish
-4. Periodic fallback checks ensure no task goes stale
+4. No recurring progress check is set up by default: the idle event and report-status wake you. Opt in with `--progress-check <minutes>` (minimum 30) only for a long task with no other signal
 
 #### Step 5: Result Verification (#140)
 **MANDATORY: When a worker reports task as done, you MUST verify before accepting.**

package/config/skills/team-leader/delegate-task/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: Delegate Task (TL)
-description: "Assign a task to a worker within the Team Leader's subordinate scope. Validates that the target worker's parentMemberId matches the TL's memberId before delegation. Includes auto-monitoring setup."
+description: "Assign a task to a worker within the Team Leader's subordinate scope. Validates that the target worker's parentMemberId matches the TL's memberId before delegation. Subscribes the lead to the worker's idle event; a recurring progress check is opt-in (minimum 30 minutes)."
 version: 1.0.0
 category: management
 skillType: claude-skill
@@ -52,6 +52,28 @@ Assigns a task to a worker within the Team Leader's subordinate scope. Validates
 | `--thread` | `thread` | No | Owner request from Slack: the key from the `[SLACK-THREAD:<key>]` of the owner's message. The member answers the owner in that thread itself — see Owner requests |
 | `--no-member-fits` | `noMemberFits` | No | You keep the work: record what is missing (access, tool, permission, everyone busy). Needs `--task`; nothing is delegated — see No member fits |
 | `--work-item` | `workItemId` | No | With `--no-member-fits`: the WorkItem you keep (gets a `[NO-MEMBER-FITS]` note) |
+| `--progress-check` | `monitor.progressCheckMinutes` | No | Opt in to a recurring "TL progress check" every N minutes for this delegation. Off by default; values under 30 are raised to 30. Legacy JSON `monitor.fallbackCheckMinutes` is read the same way |
+| — | `monitor.idleEvent` | No | `false` skips the idle event subscription (default `true`) |
+
+## Monitoring
+
+Each delegation subscribes you (one-shot) to the worker's `agent:idle` event,
+so you are woken when the worker goes idle; the worker also reports
+completion with `report-status`. That is enough for almost every task.
+
+No recurring progress check is scheduled unless you ask for one. Every check
+wakes your session and re-reads your whole context; the old default (every 5
+minutes per delegation) woke leads 219 times in 3.5 days. Opt in only for a
+long task with no other signal:
+
+```bash
+bash execute.sh --to worker-session --task "..." --progress-check 60
+```
+
+The minimum is 30 minutes (lower values are raised, with a note on stderr).
+The backend applies the same floor to any "TL progress check" it schedules.
+A check linked to the WorkItem stops once the item is done, verified,
+cancelled or failed.
 
 ## Who to delegate to
 

package/config/skills/team-leader/delegate-task/execute.sh

@@ -21,6 +21,7 @@ PROJECT_TICKET=""
 OWNER_THREAD=""
 NO_MEMBER_FITS=""
 KEPT_WORK_ITEM=""
+PROGRESS_CHECK_ARG=""
 
 # Detect legacy JSON argument
 if [[ $# -gt 0 && ${1:0:1} == '{' ]]; then
@@ -44,11 +45,13 @@ while [[ $# -gt 0 ]]; do
     --thread)      OWNER_THREAD="$2";   shift 2 ;;
     --no-member-fits) NO_MEMBER_FITS="$2"; shift 2 ;;
     --work-item)   KEPT_WORK_ITEM="$2"; shift 2 ;;
+    --progress-check) PROGRESS_CHECK_ARG="$2"; shift 2 ;;
     --json|-j)     INPUT_JSON="$2";     shift 2 ;;
     --help|-h)
       echo "Usage: execute.sh --to worker-session --task 'implement feature' --priority high --project /path [--team teamId] [--tl-member memberId] [--request-id <ticket id from the [TICKET:TKT-123 <id>] line>] [--ticket <project ticket id, e.g. APP-12>]"
       echo "Work for a teammate on a project always runs through a project ticket: --ticket uses that backlog/ready ticket, otherwise one is created for you."
       echo "Owner request from Slack: add --thread <key> (from [SLACK-THREAD:<key>]) — the member answers the owner in that thread itself."
+      echo "Optional recurring progress check (off by default; the idle notification already wakes you): --progress-check <minutes, min 30>"
       echo "Keeping work yourself because no member fits: execute.sh --no-member-fits '<what is missing: access / tool / permission / everyone busy>' --task '<the work>' [--work-item <id>] [--ticket <ID>]"
       exit 0
       ;;
@@ -132,6 +135,45 @@ require_param "task (--task)" "$TASK"
 # The check is NOT weakened — a marker genuinely absent from BOTH fields
 # still warns, and the missing-field list is still per-marker.
 #
+# Minutes between recurring "TL progress check" wakes for this delegation,
+# or 0 for none.
+#
+# Off by default: the one-shot agent:idle subscription below already wakes
+# the lead when the worker goes idle, and completion is reported with
+# report-status. The old default (every 5 minutes, per delegation) woke a
+# Claude session 219 times in 3.5 days (Owen 109; every 5 min 03:21-05:13).
+# An explicit opt-in (--progress-check N, or monitor.progressCheckMinutes /
+# legacy monitor.fallbackCheckMinutes in JSON) is raised to the 30-minute
+# floor. The backend enforces the same floor
+# (DELEGATION_PROGRESS_CHECK_CONSTANTS in backend/src/constants.ts).
+#
+# $1 = --progress-check value (may be empty), $2 = parsed JSON input.
+# Prints the interval on stdout; notes go to stderr.
+resolve_progress_check_minutes() {
+  local min_minutes=30
+  local requested="$1"
+  if [ -z "$requested" ]; then
+    local input="${2:-}"
+    [ -z "$input" ] && input='{}'
+    requested=$(printf '%s' "$input" | jq -r '.monitor.progressCheckMinutes // .monitor.fallbackCheckMinutes // empty' 2>/dev/null || true)
+  fi
+  if [ -z "$requested" ] || [ "$requested" = "null" ] || [ "$requested" = "0" ] || [ "$requested" = "false" ]; then
+    echo 0
+    return 0
+  fi
+  if ! [[ "$requested" =~ ^[0-9]+$ ]]; then
+    echo "{\"warning\":\"Ignoring progress check interval '${requested}': expected whole minutes. No recurring check scheduled.\"}" >&2
+    echo 0
+    return 0
+  fi
+  if [ "$requested" -lt "$min_minutes" ]; then
+    echo "{\"warning\":\"Progress check interval ${requested} min is below the ${min_minutes} min minimum; using ${min_minutes} min.\"}" >&2
+    echo "$min_minutes"
+    return 0
+  fi
+  echo "$requested"
+}
+
 # $1 = task, $2 = context (either may be empty).
 warn_missing_request_contract() {
   # Join with a newline rather than concatenating: the patterns below anchor
@@ -406,7 +448,8 @@ fi
 
 # Set up idle event subscription for TL monitoring
 MONITOR_IDLE=$(printf '%s' "$INPUT" | jq -r 'if .monitor.idleEvent == null then true else .monitor.idleEvent end')
-MONITOR_FALLBACK_MINUTES=$(printf '%s' "$INPUT" | jq -r 'if .monitor.fallbackCheckMinutes == null then 5 else .monitor.fallbackCheckMinutes end')
+# Recurring progress check: none unless asked for (see resolve_progress_check_minutes).
+MONITOR_FALLBACK_MINUTES=$(resolve_progress_check_minutes "$PROGRESS_CHECK_ARG" "$INPUT")
 
 COLLECTED_SCHEDULE_IDS="[]"
 COLLECTED_SUBSCRIPTION_IDS="[]"

package/config/skills/team-leader/delegate-task/execute.test.sh

@@ -26,6 +26,15 @@ if ! grep -q 'Request Contract incomplete' "$REAL_FN_FILE"; then
   echo "FATAL: could not extract warn_missing_request_contract() from execute.sh" >&2
   exit 1
 fi
+# Same for the progress-check interval resolution (default off, 30 min floor).
+sed -n '/^resolve_progress_check_minutes() {$/,/^}$/p' "$SCRIPT_DIR/execute.sh" >> "$REAL_FN_FILE"
+if ! grep -q 'min_minutes=' "$REAL_FN_FILE"; then
+  echo "FATAL: could not extract resolve_progress_check_minutes() from execute.sh" >&2
+  exit 1
+fi
+# The mock writes each /schedule request body here so scenarios can check the interval.
+SCHED_BODY_LOG="$(mktemp)"
+export SCHED_BODY_LOG
 # shellcheck source=/dev/null
 source "$REAL_FN_FILE"
 
@@ -133,6 +142,7 @@ TEAMJSON
   fi
 
   if [ "$method" = "POST" ] && [ "$endpoint" = "/schedule" ]; then
+    printf '%s\n' "$body" >> "$SCHED_BODY_LOG"
     echo '{"success":true,"data":{"checkId":"sched-001"}}'
     return 0
   fi
@@ -225,7 +235,9 @@ fi
 
 # --- THIS IS THE CODE BEING TESTED ---
 MONITOR_IDLE=$(echo "$INPUT" | jq -r 'if .monitor.idleEvent == null then true else .monitor.idleEvent end')
-MONITOR_FALLBACK_MINUTES=$(echo "$INPUT" | jq -r 'if .monitor.fallbackCheckMinutes == null then 5 else .monitor.fallbackCheckMinutes end')
+# Production line verbatim (the mock takes JSON only, so no --progress-check flag).
+PROGRESS_CHECK_ARG="${PROGRESS_CHECK_ARG:-}"
+eval "$(grep -E '^MONITOR_FALLBACK_MINUTES=' "${REAL_SKILL_DIR}/execute.sh")"
 
 COLLECTED_SCHEDULE_IDS="[]"
 COLLECTED_SUBSCRIPTION_IDS="[]"
@@ -415,6 +427,88 @@ else
   FAIL=$((FAIL + 1))
 fi
 
+# -----------------------------------------------------------------------
+# Scenario 7b: no monitor field — no recurring progress check by default
+# (219 "TL progress check" wakes in 3.5 days with the old 5-minute default)
+# -----------------------------------------------------------------------
+echo "Scenario 7b: no recurring progress check by default"
+: > "$SCHED_BODY_LOG"
+RAW_OUTPUT=$(env CREWLY_SESSION_NAME="sam-tl" bash "$MOCK_SCRIPT" "$SKILL_DIR" \
+  '{"to":"crewly-product-leo-dev","task":"Default monitoring"}' 2>&1)
+OUTPUT=$(echo "$RAW_OUTPUT" | grep '"monitoring"' | head -1)
+SCHED_COUNT=$(echo "$OUTPUT" | jq '.monitoring.scheduleIds | length')
+SUB_COUNT=$(echo "$OUTPUT" | jq '.monitoring.subscriptionIds | length')
+if [ "$SCHED_COUNT" = "0" ] && [ "$SUB_COUNT" = "1" ] && [ ! -s "$SCHED_BODY_LOG" ]; then
+  echo "  PASS: idle subscription only, no /schedule call"
+  PASS=$((PASS + 1))
+else
+  echo "  FAIL: expected 0 schedules + 1 subscription, got schedules=$SCHED_COUNT subs=$SUB_COUNT body=$(cat "$SCHED_BODY_LOG")"
+  FAIL=$((FAIL + 1))
+fi
+
+# -----------------------------------------------------------------------
+# Scenario 7c: opt-in below the floor is raised to 30 minutes, with a note
+# -----------------------------------------------------------------------
+echo "Scenario 7c: fallbackCheckMinutes=5 is raised to the 30 min floor"
+: > "$SCHED_BODY_LOG"
+RAW_OUTPUT=$(env CREWLY_SESSION_NAME="sam-tl" bash "$MOCK_SCRIPT" "$SKILL_DIR" \
+  '{"to":"crewly-product-leo-dev","task":"Opt in low","monitor":{"fallbackCheckMinutes":5}}' 2>&1)
+INTERVAL=$(jq -r '.intervalMinutes' "$SCHED_BODY_LOG" 2>/dev/null | head -1)
+RECURRING=$(jq -r '.isRecurring' "$SCHED_BODY_LOG" 2>/dev/null | head -1)
+if [ "$INTERVAL" = "30" ] && [ "$RECURRING" = "true" ] && echo "$RAW_OUTPUT" | grep -q "below the 30 min minimum"; then
+  echo "  PASS: interval raised to 30 and the lead is told"
+  PASS=$((PASS + 1))
+else
+  echo "  FAIL: expected recurring interval 30 + warning, got interval=$INTERVAL recurring=$RECURRING output=$RAW_OUTPUT"
+  FAIL=$((FAIL + 1))
+fi
+
+# -----------------------------------------------------------------------
+# Scenario 7d: opt-in at or above the floor is kept as asked
+# -----------------------------------------------------------------------
+echo "Scenario 7d: progressCheckMinutes=45 is kept"
+: > "$SCHED_BODY_LOG"
+env CREWLY_SESSION_NAME="sam-tl" bash "$MOCK_SCRIPT" "$SKILL_DIR" \
+  '{"to":"crewly-product-leo-dev","task":"Opt in 45","monitor":{"progressCheckMinutes":45}}' > /dev/null 2>&1 || true
+INTERVAL=$(jq -r '.intervalMinutes' "$SCHED_BODY_LOG" 2>/dev/null | head -1)
+if [ "$INTERVAL" = "45" ]; then
+  echo "  PASS: 45 min interval kept"
+  PASS=$((PASS + 1))
+else
+  echo "  FAIL: expected interval 45, got '$INTERVAL'"
+  FAIL=$((FAIL + 1))
+fi
+
+# -----------------------------------------------------------------------
+# Scenario 7e: a non-numeric interval schedules nothing
+# -----------------------------------------------------------------------
+echo "Scenario 7e: non-numeric interval is ignored"
+: > "$SCHED_BODY_LOG"
+RAW_OUTPUT=$(env CREWLY_SESSION_NAME="sam-tl" bash "$MOCK_SCRIPT" "$SKILL_DIR" \
+  '{"to":"crewly-product-leo-dev","task":"Bad interval","monitor":{"progressCheckMinutes":"soon"}}' 2>&1)
+if [ ! -s "$SCHED_BODY_LOG" ] && echo "$RAW_OUTPUT" | grep -q "expected whole minutes"; then
+  echo "  PASS: nothing scheduled, warning emitted"
+  PASS=$((PASS + 1))
+else
+  echo "  FAIL: expected no schedule + warning, got body=$(cat "$SCHED_BODY_LOG") output=$RAW_OUTPUT"
+  FAIL=$((FAIL + 1))
+fi
+
+# -----------------------------------------------------------------------
+# Scenario 7f: --progress-check flag value wins over JSON; floor and default hold
+# -----------------------------------------------------------------------
+echo "Scenario 7f: --progress-check flag resolves through the real function"
+if [ "$(resolve_progress_check_minutes "10" '{}' 2>/dev/null)" = "30" ] \
+  && [ "$(resolve_progress_check_minutes "" '{}')" = "0" ] \
+  && [ "$(resolve_progress_check_minutes "60" '{"monitor":{"progressCheckMinutes":45}}')" = "60" ] \
+  && grep -q -- '--progress-check) PROGRESS_CHECK_ARG="$2"' "$SCRIPT_DIR/execute.sh"; then
+  echo "  PASS: flag wins over JSON, floor and default hold"
+  PASS=$((PASS + 1))
+else
+  echo "  FAIL: resolve_progress_check_minutes / --progress-check parsing mismatch"
+  FAIL=$((FAIL + 1))
+fi
+
 # -----------------------------------------------------------------------
 # Scenario 8: Missing required params
 # -----------------------------------------------------------------------