Prime Session Watchdog incident digest (automated, Windows) #1958
Replies: 1 comment
|
Investigated this report and produced a candidate fix, validated on a fork of this repository. Root cause, fix and validation from the working notes:
Diff: awhite0030/prime-agent@main...fix-websocket-inactivity-and-rlm-lifecycle-3642992307708068967 The change passes |
Uh oh!
There was an error while loading. Please reload this page.
Automated two-hour incident batch
This message was rendered from structured, allowlisted metadata only. It contains no prompts, raw logs, session identifiers, usernames, absolute paths, or file contents.
2026-08-31T13:49:18.192000Z—2026-09-01T14:59:40.370000Z0.2.0kernel-runtimedaemon-stale-working× 1high× 1hermes-local× 1zai/glm-5.3-flash× 1provider-retry-stateretry-exhausted-needs-input× 2high× 2hermes-local× 2zai/glm-5.3-flash× 2provider-stream-lifecycledaemon-hard-stale-streaming× 1,subagent-hard-stale-streaming× 5critical× 5,high× 1openai-codex× 6gpt-5.6-sol× 1,gpt-5.6-terra× 5rlm-lifecyclerlm-metadata-stale-running× 6,rlm-passive-completed-listed-live× 10high× 6,medium× 10Publication policy
Retries from one failure episode are merged into one case. The global publisher performs at most one GitHub write per two hours. New cases remain queued until the next permitted batch.
All reactions