Codex Auto-Editing Files Without Permission in VS Code

[hani3]

Issue: Codex Auto-Editing Files Without Permission in VS Code

I am experiencing an issue with OpenAI Codex CLI (or its extension) in Visual Studio Code where it automatically edits project files without asking for my approval.

Previously, the tool used to request confirmation before applying any changes. However, it now directly modifies files, which is unsafe and not the expected behavior.

Details:
Codex applies changes automatically without user confirmation
No prompt or approval step before editing files
Behavior seems related to agent mode or auto-apply settings
This happens even when working in restricted or controlled environments
Expected Behavior:

Codex should always ask for user approval before making any file changes.

Codex edits files directly without any confirmation on macos

[hani3]

any help please want to solve

[OpenAI_Support]

Hey @hani3, really appreciate you flagging this, that’s definitely concerning behavior.

To help get this properly investigated, we’ll need a bit more detail from your side. The easiest way is through the CLI so it captures the full session context.

Can you try reloading your session in the interactive CLI, then run /feedback to upload the session details, and include the thread ID from this post in that report. That combo usually gives enough to see exactly what Codex was doing and why it might’ve edited files without an explicit action.

Once you’ve done that, feel free to link it back here too so others can follow along.

-Mark G.

[hani3]

I’m actually using the Codex extension inside Visual Studio Code, not the CLI, so I don’t have access to the /feedback command.

Could you let me know how I can capture or share session logs from the VS Code extension instead?

In the meantime, I can provide screenshots or reproduction steps if that helps.

[hani3]

[hani3]

i need help please to solve

[windysoliloquy]

have a chat with normal ChatGPT about it…
ask it for prompts to deal with it, and it will give you prompts that bar this sort of behavior

have you tried that yet?

prompt

We are continuing Prompt Forge semantic lane work.

Work type: 1. Shared slider wording update (pair semantics via collapse system)

Lane:

• Fantasy Illustration

Register:

• magitech

Target file:

• PromptForge.Core/Services/SliderLanguageCatalog.FantasyIllustrationPairs.cs

Implement only:

• Magitech Levity × Tension

Rules:

• target actual emitted prompt fragments, not assumed raw source-band strings
• keep the work lane-local inside SliderLanguageCatalog.FantasyIllustrationPairs.cs
• preserve existing pair paths
• preserve behavior for other Fantasy registers
• preserve behavior when Character Sketch is active
• no probe work
• build/diagnostics only
• live prompt-preview verification will be done locally after install

LEAVE UNCHANGED:

• PromptBuilderService.cs
• SliderLanguageCatalog.FantasyIllustration.cs
• SliderLanguageCatalog.cs shared fallback bands
• lane metadata
• defaults
• selector default nudges
• modifiers
• suppression behavior
• guide text
• UI
• compression profile
• unrelated lanes

Raw pair:

• Levity × Tension

Axes:

• X: austere → quickened
• Y: strain → containment-failure

Approved 5x5 fused map
(rows = Tension bands 0..4, columns = Levity bands 0..4)

Tension 0:

• austere operational strain
• warm operational strain
• lively operational strain
• spirited operational strain
• quickened operational strain

Tension 1:

• austere warded instability
• warm warded instability
• charged warded instability
• spirited warded instability
• quickened warded instability

Tension 2:

• stern routing tension
• warm routing tension
• lively routing tension
• spirited routing tension
• quickened routing tension

Tension 3:

• grim breach-and-overload strain
• warm-blooded overload strain
• lively overload strain
• spirited overload strain
• quickened overload strain

Tension 4:

• containment-failure gravity
• warm-blooded containment resolve
• steady containment resolve
• spirited containment defiance
• quickened containment defiance

Authoring requirements:

• pair output must replace both base slider fragments
• no leftover Levity fragment
• no leftover Tension fragment
• no internal commas in fused phrases
• Magitech should read as engineered arcana / workshop / guild-built / infrastructure / conduit power / constructed wonder
• keep the map image-model-friendly; prefer system-pressure language over personality language
• Levity must still govern tonal lift / liveliness / charge
• Tension must still govern strain / instability / routing pressure / overload / containment failure
• do not drift into high-magic buoyancy, low-magic gravity, or sword-and-sorcery swagger
• output must read naturally inside a comma-separated prompt stack

Risk watch:

• watch charged warded instability and steady containment resolve for stack naturalness
• watch the right side so quickened cells read as energized system pressure rather than generic speed
• watch the full stack for repetition around strain / instability / routing / overload / containment

Verification:

• confirm the collapse keys target the actual currently emitted magitech Levity and Tension fragments
• build/diagnostics only
• I will verify live prompt-preview behavior locally after install

Deliverable:

• files changed
• methods/sections touched
• exact pair entries added
• emitted fragments targeted by the collapse keys
• fallback behavior
• whether any helper was introduced or expanded
• mismatches/risks found
• build/diagnostics run
• unresolved decisions still blocking implementation

5.4 in my access point understands the repo that Codex is working on, very intimately.

It knows which files to forbid being written to depending on the work being done, as you’ll see in that example.

[hani3]

[hani3]

only default bermissions

i want chat option like windows

[thevaletutto]

Same here, but I don’t wanna be your beta test site for free, I want to get compensated or a big discount when it renews.