feat: watchdog + natural task creation

docs/prompt/evolve.md

@@ -350,7 +350,7 @@ You are not a task runner. You are the engineer who owns this system. Before end
 | Security / robustness | Edge cases that crash? Input validation gaps? Auto-merge exploitable? Secrets exposed? |
 
 **Constraints:**
-- **Queue-aware cap.** Count pending tasks in `docs/tasks/`. If 40+ pending, create 0 new tasks. If 30-39 pending, max 1 task. If under 30, max 3 tasks. Creating tasks is NOT mandatory — create 0 if nothing important was discovered. The queue must not grow faster than the system can close tasks.
+- **Create only what matters.** If you found a real bug, edge case, or missing test — create a task. If you didn't discover anything new — create 0. Never create tasks just to fill a quota. Check for duplicates first (scan pending tasks). The OVERSEE role handles queue cleanup.
 - **Check for duplicates first.** Scan all pending tasks in `docs/tasks/`. If a task already covers your idea, skip it or update the existing task instead.
 - **Span multiple dimensions.** If you create 3 tasks, they should not all be "code quality." Spread across at least 2 different dimensions.
 - **Vision alignment check.** Before creating tasks, read the last 5 task files (by number). Check their `vision_section` field. If 3+ target the same section, your new tasks MUST prioritize a different section. Check `docs/vision-tracker/TRACKER.md` — lower-percentage sections need more attention. Set `vision_section` in every new task's frontmatter (`loop1`, `loop2`, `self-maintaining`, `meta-prompt`, or `none`). Exception: if a section has urgent bugs or blockers, alignment can be overridden — explain why in the task description.

docs/tasks/0032.md

@@ -1,5 +1,5 @@
 ---
-status: blocked
+status: pending
 priority: normal
 environment: integration
 blocked_reason: environment

docs/tasks/0103.md

@@ -1,5 +1,5 @@
 ---
-status: blocked
+status: pending
 priority: urgent
 target: v0.0.9
 blocked_reason: design

docs/tasks/0074.md → docs/tasks/archive/0074.md

@@ -1,5 +1,5 @@
 ---
-status: pending
+status: wontfix
 priority: normal
 target: v0.0.8
 vision_section: loop2

scripts/watchdog.sh

@@ -0,0 +1,78 @@
+#!/bin/bash
+# ----------------------------------------------
+# Nightshift Watchdog
+#
+# Keeps the daemon running forever. If it crashes,
+# cleans up and restarts. Designed to run via:
+#
+#   caffeinate -s bash scripts/watchdog.sh codex 60
+#
+# Or in tmux:
+#   tmux new-session -d -s nightshift "caffeinate -s bash scripts/watchdog.sh codex 60"
+#
+# The watchdog never exits unless you kill it.
+# The daemon inside it handles its own circuit breaker,
+# budget limits, and session caps.
+# ----------------------------------------------
+
+set -uo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+REPO_DIR="$(cd "$SCRIPT_DIR/.." && pwd)"
+LOCKFILE="$REPO_DIR/.nightshift-daemon.lock"
+
+AGENT="${1:-codex}"
+PAUSE="${2:-60}"
+RESTART_DELAY=30
+MAX_RESTARTS_PER_HOUR=5
+RESTART_COUNT=0
+LAST_RESET_TIME=$(date +%s)
+
+echo "=================================================="
+echo "  NIGHTSHIFT WATCHDOG"
+echo "  Agent:         $AGENT"
+echo "  Pause:         ${PAUSE}s"
+echo "  Restart delay: ${RESTART_DELAY}s"
+echo "  Max restarts:  $MAX_RESTARTS_PER_HOUR per hour"
+echo "  Stop:          Ctrl+C"
+echo "=================================================="
+echo ""
+
+while true; do
+    # --- Rate limit restarts ---
+    NOW=$(date +%s)
+    ELAPSED=$(( NOW - LAST_RESET_TIME ))
+    if [ "$ELAPSED" -ge 3600 ]; then
+        RESTART_COUNT=0
+        LAST_RESET_TIME=$NOW
+    fi
+
+    if [ "$RESTART_COUNT" -ge "$MAX_RESTARTS_PER_HOUR" ]; then
+        echo "WATCHDOG: $MAX_RESTARTS_PER_HOUR restarts in the last hour. Something is fundamentally broken."
+        echo "WATCHDOG: Sleeping 1 hour before retrying."
+        sleep 3600
+        RESTART_COUNT=0
+        LAST_RESET_TIME=$(date +%s)
+    fi
+
+    # --- Clean stale lock ---
+    if [ -d "$LOCKFILE" ]; then
+        echo "WATCHDOG: Cleaning stale lock from previous crash."
+        rmdir "$LOCKFILE" 2>/dev/null || true
+    fi
+
+    # --- Run the daemon ---
+    echo "WATCHDOG: Starting daemon (attempt $((RESTART_COUNT + 1))) --- $(date '+%Y-%m-%d %H:%M:%S') ---"
+    bash "$SCRIPT_DIR/daemon.sh" "$AGENT" "$PAUSE"
+    EXIT_CODE=$?
+
+    RESTART_COUNT=$((RESTART_COUNT + 1))
+
+    if [ "$EXIT_CODE" -eq 0 ]; then
+        echo "WATCHDOG: Daemon exited cleanly (exit 0). Restarting in ${RESTART_DELAY}s."
+    else
+        echo "WATCHDOG: Daemon crashed (exit $EXIT_CODE). Restarting in ${RESTART_DELAY}s."
+    fi
+
+    sleep "$RESTART_DELAY"
+done