@rallycry/conveyor-agent 10.8.1 → 10.8.2

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@rallycry/conveyor-agent",
-  "version": "10.8.1",
+  "version": "10.8.2",
   "description": "Conveyor Agent Runner v10 - PTY harness for the task chat (SDK harness for audit/project-chat). Agent-as-User architecture with BaseService patterns. Works locally too.",
   "keywords": [
     "agent",

package/runtime/entrypoint.sh

@@ -835,6 +835,22 @@ trap 'echo "[pool] SIGTERM received, forwarding to agent..."; pkill -TERM -f con
 # sleep so the pod idles instead of spinning.
 GIT_PREP_MAX_RETRIES=3
 _git_prep_attempts=0
+# Bounded agent-crash supervision. An agent that keeps dying is a pod that keeps
+# burning; cap the restarts, report every attempt to the API (the report posts to
+# the task's activity log, which is also what keeps the pod alive through
+# recovery), and exit for good once the cap is hit.
+AGENT_CRASH_ATTEMPTS=0
+AGENT_CRASH_MAX=3
+
+report_agent_crash() {
+  # $1 = exit code, $2 = attempt, $3 = final (true/false). Best-effort.
+  curl -s -m 10 -X POST \
+    -H "Authorization: Bearer ${POD_BOOTSTRAP_TOKEN}" \
+    -H "Content-Type: application/json" \
+    -d "{\"exitCode\":${1},\"attempt\":${2},\"final\":${3}}" \
+    "${CONVEYOR_API_URL}/api/v3/pods/agent-crash" >/dev/null 2>&1 || true
+}
+
 while true; do
   if [ -f "$GIT_FAILED_MARKER" ]; then
     if [ "$_git_prep_attempts" -lt "$GIT_PREP_MAX_RETRIES" ]; then
@@ -852,11 +868,18 @@ while true; do
     echo "[pool] agent exited cleanly (code 0), shutting down pod."
     exit 0
   fi
+  AGENT_CRASH_ATTEMPTS=$((AGENT_CRASH_ATTEMPTS + 1))
+  if [ "$AGENT_CRASH_ATTEMPTS" -ge "$AGENT_CRASH_MAX" ]; then
+    echo "[pool] agent crashed (code $_exit_code) — attempt cap ${AGENT_CRASH_MAX} reached, giving up."
+    report_agent_crash "$_exit_code" "$AGENT_CRASH_ATTEMPTS" true
+    exit 1
+  fi
+  report_agent_crash "$_exit_code" "$AGENT_CRASH_ATTEMPTS" false
   if [ -f "$GIT_FAILED_MARKER" ] && [ "$_git_prep_attempts" -ge "$GIT_PREP_MAX_RETRIES" ]; then
-    echo "[pool] agent crashed (code $_exit_code) after git prep exhausted retries, backing off (60s)..."
+    echo "[pool] agent crashed (code $_exit_code) after git prep exhausted retries, backing off (60s, attempt ${AGENT_CRASH_ATTEMPTS}/${AGENT_CRASH_MAX})..."
     sleep 60
   else
-    echo "[pool] agent crashed (code $_exit_code), retrying in 10s..."
+    echo "[pool] agent crashed (code $_exit_code), retrying in 10s (attempt ${AGENT_CRASH_ATTEMPTS}/${AGENT_CRASH_MAX})..."
     sleep 10
   fi
 done