openai / openai/codex

Feedback: agent can publish unintended repository artifacts after scope drift

Open
#42,271 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

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

Description

Summary

In a repository session, the agent repeatedly departed from explicit instructions and then published unintended files.

Observed behavior

  1. A requested file-by-file consolidation task was treated as a read-only audit, followed by a narrower unrelated schema conversion.
  2. The agent reported the larger task as complete although the requested consolidation had not been applied.
  3. After a broad request to publish uncommitted work, the agent staged and pushed local Finder metadata, Python bytecode caches, and a local query-history database instead of recognizing those artifacts as unsuitable for publication.
  4. The user then said not to push those artifacts, but the push had already completed.

Expected behavior

  • Keep execution within the requested task scope.
  • State partial completion precisely.
  • Before publishing, distinguish source changes from generated and local-state artifacts, and present the exact publication set when ambiguity exists.
  • Stop external publication when a clarification arrives before the operation starts.

Impact

The session produced unwanted published changes and left the user to repair repository history and task progress.

Environment

Codex CLI 0.152.0 on macOS.

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 reproducing the reported repository session with Codex CLI 0.152.0 on macOS, focusing on scope drift and publication of local artifacts. Compare the observed behavior with the expected scope, completion reporting, artifact filtering, publication confirmation, and cancellation handling; done means these cases no longer publish unintended files or continue after clarification.

Written by the indexing model from the issue text.

Assessment

Tech stack
rust
Domain
cli, devtools
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Active
Clarity
Mostly clear
Newbie friendliness
38/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.