fix(cli): reap codex subprocess on failure in codex templates

Address Greptile review: wrap the streaming call in try/finally across the
default, sync and temporal codex templates so the codex subprocess is killed
and reaped even when auto_send_turn / yield_turn raises or the async generator
is abandoned. Previously a failed turn left codex blocked on a full stdout pipe
buffer, leaking an OS process per failure until the server/worker restarted.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: declan-scale

src/agentex/lib/cli/templates/default-codex/project/acp.py.j2

@@ -192,9 +192,15 @@ async def handle_task_event_send(params: SendEventParams):
             parent_span_id=turn_span.id if turn_span else None,
         )
 
-        result = await emitter.auto_send_turn(turn)
-
-        await process.wait()
+        # Guarantee the subprocess is reaped even if auto_send_turn raises
+        # (e.g. a Redis error); otherwise codex stays blocked writing to a full
+        # stdout pipe buffer and the OS process leaks until the server restarts.
+        try:
+            result = await emitter.auto_send_turn(turn)
+        finally:
+            if process.returncode is None:
+                process.kill()
+            await process.wait()
 
         # Record the real wall-clock duration AFTER streaming completes; setting
         # it before the stream ran would capture only subprocess spawn overhead.

src/agentex/lib/cli/templates/sync-codex/project/acp.py.j2

@@ -156,10 +156,16 @@ async def handle_message_send(
             parent_span_id=turn_span.id if turn_span else None,
         )
 
-        async for event in emitter.yield_turn(turn):
-            yield event
-
-        await process.wait()
+        # Guarantee the subprocess is reaped even if the generator is abandoned
+        # (client disconnect / GC) or yield_turn raises; otherwise codex stays
+        # blocked writing to a full stdout pipe buffer and the process leaks.
+        try:
+            async for event in emitter.yield_turn(turn):
+                yield event
+        finally:
+            if process.returncode is None:
+                process.kill()
+            await process.wait()
 
         # Record the real wall-clock duration AFTER streaming completes; setting
         # it before the stream ran would capture only subprocess spawn overhead.

src/agentex/lib/cli/templates/temporal-codex/project/activities.py.j2

@@ -134,9 +134,15 @@ async def run_codex_turn(params: RunCodexTurnParams) -> dict[str, Any]:
         trace_id=params.trace_id,
         parent_span_id=params.parent_span_id,
     )
-    result = await emitter.auto_send_turn(turn, created_at=params.created_at)
-
-    await process.wait()
+    # Guarantee the subprocess is reaped even if auto_send_turn raises;
+    # otherwise codex stays blocked writing to a full stdout pipe buffer and the
+    # OS process leaks until the worker restarts.
+    try:
+        result = await emitter.auto_send_turn(turn, created_at=params.created_at)
+    finally:
+        if process.returncode is None:
+            process.kill()
+        await process.wait()
 
     return RunCodexTurnResult(
         final_text=result.final_text,