Chat deadline timeouts with partial responses (ChatDeadlineExpiredPartial)

[krishagel]

Summary

Agents are hitting ChatDeadlineExpiredPartial errors — the chat deadline expires with a partial response (last_state=aborted). Two occurrences in the last 4 hours, both on the same session. The agent processes a large number of internal events (72-163 agent events) before the deadline fires, suggesting the agent is running out of time while still processing.

Definition of Done

• Root cause identified: determine whether the deadline is too short, the agent is doing too much work per turn, or there is a performance bottleneck
• Event counts and timing analyzed to find the bottleneck
• Proposed fix documented (e.g., increase chat deadline, optimize agent event processing, or add progress checkpoints)
• (manual) Fix applied and verified in staging

Affected failures

ID	Occurred At (UTC)	User	Source	Error
255	2026-07-29 23:11:51	hagelk@psd401.net	harness	ChatDeadlineExpiredPartial — chat deadline expired with partial response (last_state=aborted)
251	2026-07-29 22:34:00	hagelk@psd401.net	harness	ChatDeadlineExpiredPartial — chat deadline expired with partial response (last_state=aborted)

Affected users

• hagelk@psd401.net

Error details

ID 255:

error_class: ChatDeadlineExpiredPartial
error_message: chat deadline expired with partial response (last_state=aborted)
context: {"phase": "deadline", "last_state": "aborted", "event_counts": {"res": 1, "event:chat": 4, "event:tick": 5, "event:agent": 72, "event:health": 3}}

ID 251:

error_class: ChatDeadlineExpiredPartial
error_message: chat deadline expired with partial response (last_state=aborted)
context: {"phase": "deadline", "last_state": "aborted", "event_counts": {"res": 1, "event:chat": 6, "event:tick": 10, "event:agent": 163, "event:health": 5}}

Acceptance tests / E2E flows

N/A — this is a performance/configuration issue.

Affected areas

• Chat deadline configuration
• Agent event processing loop
• Harness timeout settings

Out of scope

• Changing the agent architecture
• Removing health checks

Risk & rollback

• Risk: medium — deadline changes affect all agent turns
• Rollback: revert deadline configuration to previous value

[krishagel]

This issue was created in the wrong repo. The correct issue is in psd401/aistudio.