[omarchybot]

A follow-up to #14040, from reviewing it at 8d957f3.

open_json in shell/plugins/agents/collectors.py reads a successful reply's body through _read_by, which stops at three times the socket timeout, because the socket timeout bounds only each wait and a server dripping a byte at a time would otherwise hold the collector forever. An HTTP error status skipped that: its body was read with error.read(limit), bounded only per wait. A provider answering 429 or 500 and dripping its body held that collector indefinitely, and since omarchy-agent-usage-update waits on every collector and the panel queues refreshes behind a running update, every later refresh stalled with it.

The error body now goes through _read_by as well. Past the deadline it is left out, and the HttpError still carries the status and headers, which is all any caller reads (Claude's Retry-After handling uses the headers; nothing reads the body). The new check in agent-usage-shared-test.sh serves a 500 that drips its body and fails without the change.

Codex Medium raised the defect in the review of #14040; Codex Medium and muse-spark-1.3 (pi, high) reviewed this change and found nothing to change.

[greptile-apps]

Confidence Score: 5/5

[Medium impact] The PR appears safe to merge.

Summary

open_json now reads HTTP error bodies through the same deadline-aware reader as successful replies.

• HTTP error replies stop waiting at the shared body-read deadline.

_{Reviews (1) · Last reviewed commit: "Hold an error reply's body to the same d..." · Reviewed by Greptile}