Fix some issues with local dev

scripts/dev-agent-worker

@@ -27,6 +27,8 @@ export SPOON_AGENT_WORKER_URL="${SPOON_AGENT_WORKER_URL:-http://localhost:${SPOO
 export SPOON_AGENT_WORKER_INTERNAL_TOKEN="${SPOON_AGENT_WORKER_INTERNAL_TOKEN:-${SPOON_WORKER_TOKEN:-}}"
 export SPOON_AGENT_WORKDIR="${SPOON_AGENT_LOCAL_WORKDIR:-.local/agent-work/${WITH_ENV_ENVIRONMENT:-dev}}"
 export SPOON_AGENT_JOB_IMAGE="${SPOON_AGENT_LOCAL_JOB_IMAGE:-spoon-agent-job:latest}"
+# Self-terminate if the dev runner dies, so the worker never orphans on port 3921.
+export SPOON_AGENT_DEV_WATCHDOG="${SPOON_AGENT_DEV_WATCHDOG:-1}"
 
 if [[ "$SPOON_AGENT_CONTAINER_ACCESS" == "host_port" && -z "${SPOON_AGENT_KEEP_NETWORK:-}" ]]; then
   unset SPOON_AGENT_NETWORK

scripts/export-env

@@ -24,10 +24,25 @@ fi
 command -v infisical >/dev/null 2>&1 || { echo "export-env: Infisical CLI is required." >&2; exit 1; }
 "$ROOT_DIR/scripts/infisical-account" ensure
 
-(cd "$ROOT_DIR" && infisical export --env="$INFISICAL_ENV" --format=dotenv --silent) || {
-  echo "export-env: failed to export '$INFISICAL_ENV'; check login and project access." >&2
-  exit 1
-}
+# Retry transient Infisical failures (e.g. 500s when several dev tasks fetch
+# concurrently at startup) so one flaky response doesn't kill the dev server.
+attempt=0
+while :; do
+  attempt=$((attempt + 1))
+  if EXPORT_OUT=$(cd "$ROOT_DIR" && infisical export --env="$INFISICAL_ENV" --format=dotenv --silent 2>"/tmp/export-env.$$.err"); then
+    printf '%s\n' "$EXPORT_OUT"
+    break
+  fi
+  if [ "$attempt" -ge 5 ]; then
+    cat "/tmp/export-env.$$.err" >&2 2>/dev/null || true
+    rm -f "/tmp/export-env.$$.err"
+    echo "export-env: failed to export '$INFISICAL_ENV' after $attempt attempts; check login and project access." >&2
+    exit 1
+  fi
+  echo "export-env: Infisical export failed (attempt $attempt/5), retrying in 2s..." >&2
+  sleep 2
+done
+rm -f "/tmp/export-env.$$.err"
 
 if [ -f "$STATE_FILE" ]; then
   printf '\n'