Regression: No reliable way to review file changes after an agent task

What issue are you seeing?

This affects both the Codex Desktop App and the VS Code extension. The behavior appears to be shared across both clients.

Summary

Recent versions of Codex (both the standalone application and the VS Code extension) no longer provide a reliable review of the changes made during an agent task.

Instead of showing a clear list of modified files with diffs, the UI often displays mostly terminal commands (grep, find, git, npm, sed, etc.) and a high-level textual summary.

This is a significant regression because it removes the ability to audit what the agent actually changed.

Expected behavior

After every completed task, Codex should provide:

• a complete list of modified files;
• per-file diffs (or at least a way to open them);
• a clear distinction between files that were edited, created, deleted, or renamed;
• enough information to review every change before accepting it.

This was previously much easier to do.

Actual behavior

After many tasks I only see:

• terminal command logs;
• brief natural-language summaries.

There is often no practical way to determine:

• which files were modified;
• which files were modified during the last turn;
• what exact changes were made.

Why this is a problem

This is more than a UX issue.

Codex is an autonomous coding agent. Users need to be able to audit every change it makes.

Recently I've experienced situations where Codex modified files that I did not ask it to change, and some of those changes introduced regressions.

Without a reliable "changed files + diff" view, it becomes extremely difficult to answer basic questions such as:

• Which prompt introduced the bug?
• Which files were changed?
• Which changes were intentional versus incidental?
• What should I revert?

Terminal logs are not a replacement for code diffs.

Impact

This negatively affects:

• code review;
• debugging;
• trust in autonomous edits;
• safe use of Codex on production codebases.

It also forces users to manually inspect the repository with Git after nearly every interaction, defeating much of the value of having an integrated review experience.

Request

Please restore a reliable post-task review that includes:

• complete list of modified files;
• accurate per-file diffs;
• a consistent review experience in both the standalone Codex application and the VS Code extension.

For an autonomous coding agent, transparent review of changes is not an optional convenience—it is a fundamental requirement.

What steps can reproduce the bug?

1. Open the Codex Desktop App or the Codex VS Code extension.
2. Open any existing project under Git version control.
3. Ask Codex to perform a task that modifies multiple files (for example, implement a feature, refactor code, or fix several related issues).
4. Wait for the task to complete.
5. Open the review/results panel for the completed task.

Observed behavior:
The review often contains mostly terminal command logs (e.g. grep, find, git, npm, sed) and a high-level summary. There is no reliable list of all modified files, and diffs are missing or incomplete, making it difficult to determine exactly what the agent changed.

Expected behavior:
The review should consistently display all files modified during the task, along with accurate per-file diffs (or an equivalent review interface) so the user can audit every change before proceeding.

What is the expected behavior?

_No response_

Additional information

_No response_