Codex 0.118 stops unattended runs with opaque “Missing required artifacts and external info” approval prompts and no “Always allow” option

What version of Codex CLI is running?

0.118

What subscription do you have?

Pro

Which model were you using?

gpt-5.3-codex

What platform is your computer?

Linux 6.8.0-107-generic x86_64 x86_64

What terminal emulator and version are you using (if applicable)?

Gnome terminal

What issue are you seeing?

Summary

After upgrading to Codex CLI 0.118, unattended/autonomous workflows are no longer reliable.

I now get approval prompts that only offer:

• Allow
• Cancel

with no Always allow option.

One recurring prompt looks like this:

Tool call needs your approval. Reason: Missing required artifacts and external info

This completely breaks overnight / unattended runs, because Codex stops and waits for manual confirmation.

Why this is a problem

I use Codex specifically for long-running autonomous workflows (for example, running overnight in IntelliJ + MCP-backed project workflows).

If Codex can stop at arbitrary points with a blocking approval dialog that:

1. has no Always allow option, and
2. does not clearly explain what is actually missing,

then the autonomous workflow becomes effectively impossible.

At that point Codex is no longer usable for the main reason I adopted it.

Environment

• Codex CLI: 0.118
• IntelliJ IDEA with JetBrains MCP server
• Project path example: /home/oleg/Data/projects/ta
• MCP tool involved in at least one case: get_file_text_by_path

Related behavior

I also see inconsistent approval behavior between orchestrator and subagents.

Example:

• orchestrator asks for approval for an MCP tool call
• I select Always allow
• later, a subagent asks again for the same kind of operation

Additionally, some approval prompts do not offer Always allow at all and only show Allow / Cancel.

This suggests there may be multiple approval/interaction paths, some of which are not compatible with unattended execution.

Example prompt

Example of the blocking prompt:

Tool call needs your approval. Reason: Missing required artifacts and external info

maxLinesCount: 300
pathInProject: src/test/java/trade/ta/auto/indicator/calibration/Microba...
projectPath: /home/oleg/Data/projects/ta

1. Allow   Run the tool and continue.
2. Cancel  Cancel this tool call

Expected behavior

For autonomous / unattended workflows:

• Codex should not stop on opaque prompts that require manual confirmation unless the action is genuinely unsafe or outside the configured trust boundary.
• If a prompt is shown, it should clearly explain:
• what artifact is missing,
• what “external info” means in this context,
• why Codex cannot proceed automatically.
• If the action is safe/read-only and within already approved boundaries, approval should be reusable or inheritable.
• Subagents should respect previously granted approval where applicable.
• There should be a way to run Codex in a truly unattended mode without these blocking prompts for safe operations.

Actual behavior

Codex blocks execution with vague approval prompts, sometimes without Always allow, and waits for user interaction.
This makes overnight/autonomous workflows unreliable or impossible.

Repro (high level)

1. Run Codex CLI 0.118 on a real project with IntelliJ MCP integration.
2. Let Codex orchestrate a multi-step task that may spawn subagents.
3. Approve MCP access where requested.
4. At some point Codex stops with:

• repeated approval prompts for similar operations, or
• a blocking Missing required artifacts and external info prompt with only Allow / Cancel.

Request

Please clarify whether this is:

• an intentional behavior change,
• a regression in approval propagation / subagent inheritance,
• or a bug in the new approval / interaction flow.

Also, please provide a supported way to preserve truly unattended execution for safe project-local workflows.

What steps can reproduce the bug?

What steps can reproduce the bug?

1. Install and run Codex CLI 0.118.
2. Use Codex in a real project with IntelliJ IDEA and the JetBrains MCP server enabled.
3. Start a multi-step autonomous task that requires project file inspection and may spawn subagents.
4. When the orchestrator requests approval for an MCP tool call, choose Always allow.
5. Let the task continue until a subagent performs a similar MCP-backed operation.
6. Observe that the subagent may ask for approval again instead of reusing the existing approval decision.
7. In some runs, observe an additional blocking prompt like:

Tool call needs your approval. Reason: Missing required artifacts and external info

maxLinesCount: 300
pathInProject: src/test/java/trade/ta/auto/indicator/calibration/Microba...
projectPath: /home/oleg/Data/projects/ta

1. Allow   Run the tool and continue.
2. Cancel  Cancel this tool call

### What is the expected behavior?

_No response_

### Additional information

_No response_