User story
As an agent running several read-only verification calls in one MCP session, I need every run to return a terminal structured result so I can determine which checks completed.
Summary
A multi-call MCP run ended after the lifecycle event that started its final read-only preview query. It emitted neither a terminal lifecycle event for that call nor the run's structured JSON result.
What the agent tried
- Started one MCP run containing four sequential read-only calls.
- Observed successful terminal lifecycle events for the first three calls.
- Waited for the fourth preview query and the run's structured result.
Expected behavior
The final call should emit either success or failure, and the run should always return structured JSON that includes terminal status and any partial call results.
Actual result
The process ended after reporting that the final preview query had started. No success or failure event and no structured run result were returned.
Recovery attempts
- Polled the active process twice for additional output.
- Inspected the lifecycle stream to confirm the earlier calls had completed before the missing terminal event.
User impact
Batch verification cannot be parsed or trusted, so each read-only check must be repeated in a separate process and automation cannot distinguish termination from success.
Technical context
The workflow used the source-tree CLI in MCP execution mode. The final operation was a read-only asset query preview; no write was in progress.
Agent environment
- Client: Codex
- Provider: OpenAI
- Model: unknown
- Reasoning effort: unknown
- Trigger: automatic-friction
- Category: hang-or-crash
Acceptance criteria
- Every MCP run emits a structured terminal result on both success and failure.
- Unexpected termination preserves completed call results and identifies the unfinished call.
- A regression test covers a multi-call run that terminates during a read-only preview operation.
User story
As an agent running several read-only verification calls in one MCP session, I need every run to return a terminal structured result so I can determine which checks completed.
Summary
A multi-call MCP run ended after the lifecycle event that started its final read-only preview query. It emitted neither a terminal lifecycle event for that call nor the run's structured JSON result.
What the agent tried
Expected behavior
The final call should emit either success or failure, and the run should always return structured JSON that includes terminal status and any partial call results.
Actual result
The process ended after reporting that the final preview query had started. No success or failure event and no structured run result were returned.
Recovery attempts
User impact
Batch verification cannot be parsed or trusted, so each read-only check must be repeated in a separate process and automation cannot distinguish termination from success.
Technical context
The workflow used the source-tree CLI in MCP execution mode. The final operation was a read-only asset query preview; no write was in progress.
Agent environment
Acceptance criteria