fix(agent-ops): stall 이후 새 세션을 사용한다

agent-ops/.version

@@ -1 +1 @@
-1.1.199
+1.1.200

agent-ops/skills/common/orchestrate-agent-task-loop/SKILL.md

@@ -113,6 +113,7 @@ Accept self-check completion only when `## Implementation Checklist` or its supp
 - Determine liveness from PID/start-token/process-marker evidence and actual stream or native-session progress. Heartbeat mtime is never agent progress. For Codex JSONL, an unmatched `item.started` `command_execution` is an active tool interval: suspend the model-response silence timer until its matching `item.completed`, then restore normal stall detection.
 - The dispatcher model-silence safety net is 70 seconds. Downstream provider runtimes should emit their bounded terminal before that deadline; do not extend the dispatcher budget per target to cover nested retries.
 - Treat a confirmed provider transport terminal as the end of the current dispatch. Do not resume or automatically resend the same native session; an operator may start a fresh dispatch after the provider/runtime state is corrected.
+- Retry `session-stall` only with a fresh native conversation. Preserve workspace changes and logical locator evidence, but do not carry the silent conversation context into the next attempt or a restarted dispatcher.
 - Never start a duplicate attempt while owned live evidence remains.
 - Keep a 10-consecutive-failure budget per task stage. Reset only that stage's budget after success.
 - Preserve failed attempt logs. Delete successful attempt logs only after verified archive completion and no live evidence.

agent-ops/skills/common/orchestrate-agent-task-loop/scripts/dispatch.py

@@ -3253,7 +3253,7 @@ def native_resume_locator(
     if (
         not isinstance(record.get("runtime"), dict)
         or not record["runtime"].get("native_session_monitor")
-        or record.get("failure_class") not in {"context-limit", "session-stall"}
+        or record.get("failure_class") != "context-limit"
         or record.get("status") != "failed"
     ):
         return None
@@ -4690,7 +4690,7 @@ async def run_escalating(
                     ],
                 )
                 previous_locator = locator
-                native_resume_locator = locator
+                native_resume_locator = locator if failure == "context-limit" else None
                 await asyncio.sleep(min(30, 2 ** min(native_recovery_retries, 5)))
                 continue
             native_recovery_retries += 1
@@ -5341,7 +5341,11 @@ async def run_worker(
     resume_locator: Path | None = None,
 ) -> None:
     retry_context = store.task_state(task).get("retry_failover_context")
-    if resume_locator is None and isinstance(retry_context, dict):
+    if (
+        resume_locator is None
+        and isinstance(retry_context, dict)
+        and retry_context.get("failure_class") not in PROVIDER_TRANSPORT_FAILURES
+    ):
         locator_value = retry_context.get("locator")
         if isinstance(locator_value, str) and locator_value:
             resume_locator = Path(locator_value)