macOS Desktop 0.149 regression: named permission profile overrides on-request/auto_review with never/user
What version of the Codex App are you using?
ChatGPT/Codex Desktop 26.818.41509 (build 6962)
Bundled Codex CLI: 0.149.0-alpha.4.1
What platform are you using?
macOS ARM64 (Apple Silicon)
What issue are you seeing?
Codex Desktop overrides the separately configured approval policy and approval reviewer when starting a fresh task with a named permission profile.
The relevant configuration is:
default_permissions = "Auto"
approval_policy = "on-request"
approvals_reviewer = "auto_review"
[permissions.Auto]
description = "Workspace access with automatic review for escalations."
extends = ":workspace"
The bundled runtime loads this configuration successfully:
config.load.status = ok
sandbox.helpers.approval policy = OnRequest
filesystem sandbox = restricted
network sandbox = enabled
However, the first turn of a fresh Desktop task receives:
approval_policy = never
approvals_reviewer = user
permission_profile.type = managed
active_permission_profile.id = Auto
sandbox_policy.type = workspace-write
The named permission profile itself is active, but the two independent approval settings are replaced:
on-request -> never
auto_review -> user
Because the effective approval policy is never, actions outside the permission profile cannot request escalation, and the automatic reviewer is never invoked.
No permission mode or profile transition occurred before the first turn.
Regression evidence
The same named profile and approval configuration previously worked in a Desktop task using bundled Codex CLI 0.148.0-alpha.9.
Recorded effective values in 0.148.0-alpha.9:
approval_policy = on-request
approvals_reviewer = auto_review
active_permission_profile.id = Auto
sandbox_policy.type = workspace-write
Recorded effective values in 0.149.0-alpha.4.1:
approval_policy = never
approvals_reviewer = user
active_permission_profile.id = Auto
sandbox_policy.type = workspace-write
There was no relevant permission configuration change between these observations.
This establishes a local regression boundary between 0.148.0-alpha.9 and 0.149.0-alpha.4.1.
Steps to reproduce
- Configure a named workspace permission profile together with:
```toml
default_permissions = "Auto"
approval_policy = "on-request"
approvals_reviewer = "auto_review"
[permissions.Auto]
extends = ":workspace"
```
- Fully quit Codex Desktop.
- Reopen the app.
- Create a fresh local task.
- Do not switch permission modes or profiles.
- Inspect the first turn's effective approval policy, reviewer, active permission profile, and sandbox policy.
Observed:
never / user / Auto / workspace-write
Expected:
on-request / auto_review / Auto / workspace-write
Expected behavior
Selecting or resolving a named permission profile should not replace the independently configured approval policy and reviewer.
The effective task configuration should be:
approval_policy = on-request
approvals_reviewer = auto_review
active_permission_profile.id = Auto
sandbox_policy.type = workspace-write
Related reports and changes
- #33552 reports the same
never/useroutcome on earlier Desktop builds. This report adds a known-good0.148.0-alpha.9comparison and a specific0.148 -> 0.149regression boundary. - #37437 reports a related Desktop config-to-runtime permission handoff failure involving legacy Full Access settings, but with a different input configuration and different effective approval state.
- #39597 changed how effective environment permissions and thread settings are separated and used to construct turn context. It falls within this regression window and may be relevant, but I have not bisected the regression to that change.
1 Comment
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action