Emit AgentStarted timeline event for 3p harnesses#12031
Conversation
|
I'm starting a first review of this pull request. You can view the conversation on Warp. I completed the review and no human review was requested for this pull request. Comment Powered by Oz |
There was a problem hiding this comment.
Overview
This PR renames the setup timeline event enum to cover Oz run timeline events and emits agent_started after Claude Code, Codex, and Gemini harness commands are launched. I did not find security findings, and spec_context.md contains no approved spec commitments to compare against.
Concerns
- The new
agent_startedPOST is awaited after the harness process starts but before the runner is markedRunning, so a slow or stuck client-event request can delay harness monitoring and leave early session updates/saves observingPreexec. app/src/ai/agent_sdk/setup_observability_tests.rsis listed with an unavailable patch in the attached diff, so this review could only validate the visible hunks.
Verdict
Found: 0 critical, 1 important, 0 suggestions
Request changes
Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).
Powered by Oz
|
/oz-review |
|
I'm re-reviewing this pull request in response to a review request. You can view the conversation on Warp. I completed the review and no human review was requested for this pull request. Comment Powered by Oz |
There was a problem hiding this comment.
Overview
This PR emits an agent_started Oz run timeline event after Claude Code, Codex, and Gemini harness commands successfully start, and renames the existing setup timeline event enum to cover both setup and run lifecycle events.
Concerns
- No blocking correctness, security, or spec-alignment concerns found in the annotated diff.
Verdict
Found: 0 critical, 0 important, 0 suggestions
Approve
Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).
Powered by Oz
Description
For 3p harnesses, the client needs to emit an event for us to know that the agent has begun work in the handoff case. We rely on this instead of creating the conversation, because creating the conversation only happens when it's not a handoff.
Testing
Tested with https://github.com/warpdotdev/warp-server/pull/11462 using claude code - lifecycle through shutdown and follow up