openai / openai/codex

Repeated malformed automation responses with unrelated Chinese text and apparent tool-routing leakage

Open
#46,667 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

app automations bug model-behavior
Dominant language
Rust
Stars
125k
Forks
19.4k
PR merge metrics
PR metrics pending

Description

What version of the Codex App are you using (From “About Codex” dialog)?

26.903.71938 (build 8576)

What subscription do you have?

pro 200

What platform is your computer?

Darwin 24.2.0 arm64 arm

What issue are you seeing?

An English-language scheduled monitoring task has produced malformed
visible responses at least twice.

First occurrence: September 19, 2026, 08:16:42 PDT
UTC: 2026-09-19T15:16:42.609Z

Actual output:

<heartbeat>
  <automation_id>monitor-xmr-one-minute-reversals</automation_id>
  <decision>NOTIFY</decision>
  天天中彩票粵user

The Chinese characters are present in the saved assistant output_text
in the local session JSONL, not only in the displayed screenshot.
The expected monitoring message and closing heartbeat tag are missing.

Second occurrence:
The visible response again began with heartbeat markup, then contained
unexpected non-English characters followed by:

assistant to=functions.exec
// @exec: {"max_output_tokens": 3414}

Expected:
A complete English-language monitoring update.

Actual:
Unrelated multilingual text, incomplete heartbeat markup, and apparent
tool-routing/execution formatting displayed as response content.

Context:
This occurred during repeated scheduled updates in an existing task.
One subsequent update returned to English, but the failure later recurred.
I do not have a reliable minimal reproduction.

Task ID: 01a0b9b1-f336-75e0-aeb0-62107ab18fae
First affected turn ID: 01a0ba3c-f8fd-72a3-a011-90eec4de2c89

Please investigate the malformed output and possible internal
message/tool-format leakage. The root cause has not been established.

What steps can reproduce the bug?
  1. A scheduled monitor ran in an existing task.
  2. The run appended its monitoring results to a local journal.
  3. Its final response contained the malformed text above instead of
    the expected monitoring summary.
  4. After I asked about the Chinese text, the assistant called it a
    "formatting glitch" without establishing the cause.
  5. The following scheduled response contained an English update,
    although it still displayed heartbeat markup.
What is the expected behavior?

Expected behavior

A complete English-language monitoring update, without unrelated
Chinese text or exposed automation markup.

Image Image
Additional information

No response

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by inspecting the local session JSONL for task 01a0b9b1-f336-75e0-aeb0-62107ab18fae and turn 01a0ba3c-f8fd-72a3-a011-90eec4de2c89, comparing the malformed responses with the later English update. Done means identifying the cause of the leaked automation or tool-routing content and preventing it from appearing in scheduled monitoring responses.

Written by the indexing model from the issue text.

Assessment

Domain
devtools
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Active
Clarity
Needs clarification
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.