OpenAI Codex团队修复OpenClaw认证漏洞

Gold Financial reports that on April 23rd, the OpenAI Codex team is focusing on optimizing the experience of OpenAI models within OpenClaw. Codex engineering lead Tibo Sottiaux stated that the team is collaborating with OpenClaw creator Peter Steinberger and that more progress will be announced next week.
Codex product lead Nik Pash discovered a critical issue during debugging: when configuring OpenClaw to use the Codex harness with OpenAI models, the authentication process fails silently, causing the system to fallback to Pi harness, leading users to believe that the Codex harness is functioning normally, which is not the case. Pash submitted two PRs, one fixing the authentication bridging and the other preventing silent fallback.
The behavior of the agent before and after the fix differs significantly. When using Pi harness, the agent performs only shallow polling on each heartbeat: reading the heartbeat file, checking Discord, returning HEARTBEAT_OK, ignoring other commands, sometimes inferring actions but not issuing tool calls. After switching to the Codex harness, the agent enters a full working cycle: reading workspace context, parsing task lists, checking repositories, executing edits, attempting validation, and subsequent heartbeats can connect progress rather than repeat work.
Pash stated that the agent’s prompt and upper-layer workflow logic have not been changed; the improvement entirely comes from the underlying runtime adapter (harness) switch, meaning the way OpenClaw communicates with the model API has changed.