orchestration: run workers as a sprite service; record wave 4 completion

setsid nohup did not survive orchestrator teardown on 09-06 (uptime was
continuous, so this was a teardown kill, not the hazard-#5 suspend).
tasks/service-worker.sh runs a worker under the sprite service supervisor
instead, which both outlives the orchestrator and holds the box awake, making
the task-lease guard redundant. It is sentinel-guarded so a supervisor restart
does not re-run a finished wave, and it stops its own service afterwards so the
sprite can suspend.

logs/WAVE4-DONE records that F3's files were all complete but the worker was
stuck re-running verification it could not finish, so the orchestrator ran the
verification itself and accepted the work. F3's own written report — including
the design critique of the rendered screens it was asked for — was never
produced; that gap is recorded in the sentinel.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014TyzeWmdTqi7U85iYNGy7P

tasks/service-worker.sh

@@ -0,0 +1,63 @@
+#!/usr/bin/env bash
+# service-worker.sh <task-name> <prompt-file> <sentinel-name>
+#
+# Runs ONE worker under the sprite service supervisor instead of `setsid nohup`.
+#
+# WHY THIS EXISTS (wave-4 post-mortem, second failure):
+# `setsid nohup ./tasks/run-task.sh ...` does NOT reliably outlive the orchestrator.
+# On 09-06 at 19:04 the orchestrator's session was torn down and the detached worker
+# AND its wave-guard died with it -- no reboot (uptime was continuous), so this was a
+# teardown kill, not the hazard-#5 suspend. `setsid` did not save it.
+#
+# /.sprite/llm.txt: services are "long-running processes that persist across reboots"
+# and "services and sessions keep sprites alive". A service therefore fixes BOTH
+# failure modes at once -- it survives orchestrator teardown AND holds the box awake --
+# which makes the task-lease guard redundant.
+#
+# IDEMPOTENCE MATTERS HERE: the supervisor restarts services. Without the sentinel
+# guard below, a finished wave would be re-run on every restart and burn quota
+# redoing completed work.
+set -u
+NAME="$1"; PROMPT="$2"; SENT="$3"
+L="$HOME/bookshelf/logs"; DONE="$L/$SENT"
+SVC="${SVC_NAME:-bookshelf-worker}"
+
+log() { echo "[$(date -Is)] service-worker: $*" >> "$L/service-worker.log"; }
+
+# Already finished -> do not re-run. Idle briefly so a restart loop stays slow.
+if [ -f "$DONE" ]; then
+  log "$SENT already exists; nothing to do (service can be deleted)"
+  sleep 300
+  exit 0
+fi
+
+log "starting worker $NAME (prompt=$PROMPT, sentinel=$SENT)"
+"$HOME/bookshelf/tasks/run-task.sh" "$NAME" "$PROMPT"
+rc=$?
+log "worker $NAME exited rc=$rc; writing $SENT"
+
+{
+  echo "=== $SENT written $(date -Is) ==="
+  echo "Worker finished under the service supervisor. The orchestrator was NOT"
+  echo "necessarily alive for this. run-task.sh exit code: $rc"
+  echo
+  echo "--- $NAME ---"
+  grep -hE 'SUCCESS|GIVING UP|WALL CLOCK|QUOTA' "$L/$NAME.state" 2>/dev/null | tail -5
+  if [ -s "$L/$NAME.json" ]; then
+    jq -r '"cost=$" + ((.total_cost_usd//0)|tostring) + " turns=" + ((.num_turns//0)|tostring)' "$L/$NAME.json" 2>/dev/null
+    jq -r '"is_error=" + ((.is_error//false)|tostring)' "$L/$NAME.json" 2>/dev/null
+  else
+    echo "!! $NAME.json is 0 bytes -> worker was KILLED, not finished (hazard #3)"
+  fi
+  echo
+  echo "NEXT: orchestrator must independently verify before accepting:"
+  echo "  cd ~/bookshelf && ./tasks/gw assembleDebug && ./tasks/gw testDebugUnitTest"
+  echo "  git status --porcelain   # what did the worker actually touch?"
+} > "$DONE"
+sync
+
+# Sentinel is durable now, so it is safe to stop ourselves. This lets the sprite
+# suspend instead of idling hot on the owner's dime.
+log "stopping service $SVC so the sprite can suspend"
+sprite-env services stop "$SVC" >/dev/null 2>&1
+sleep 60