Add a safe first-party action to start Codex Security scans

What variant of Codex are you using?

App / Codex Security / Computer Use

What feature would you like to see?

Please provide a safe first-party way for Codex agents to start a prepared Codex Security scan from the Codex Security app/workspace, without requiring Computer Use to control the Codex desktop app itself.

Today, after an agent prepares a scoped or diff Codex Security scan, the last step may require clicking Start scan in the Codex Security UI. Trying to automate that click through Computer Use is blocked with:

Computer Use is not allowed to use the app 'com.openai.codex' for safety reasons.

That safety boundary is understandable. This request is not asking for unrestricted Computer Use access to com.openai.codex. The request is for one of these safe alternatives:

1. Preferred: expose a dedicated Codex Security tool/action, for example start_codex_security_scan, that starts the already prepared scan and returns a scan id or structured status.
2. Alternative: allow an allowlisted, auditable action for the Codex Security Start scan button only.

This would let an agent complete scoped security workflows when the user has already authorized the scan, instead of stalling for a manual UI click.

Suggested behavior:

• The tool/action should show or receive the scan scope before starting.
• It should fail closed if the scan scope changed, if no prepared scan/workspace exists, or if user approval is required but missing.
• It should return a scan id, status, or structured error that the agent can use for follow-up.
• It should not enable arbitrary self-inspection or arbitrary control of the Codex desktop app.
• It should leave an audit trail in the thread/tool output.

Additional information

Related issues: #23452 and #24761 discuss the broader intentional Computer Use restriction for com.openai.codex. This request is narrower: preserve that safety restriction, but add a supported Codex Security scan-start path.

Concrete case:

• A user explicitly asks the agent to run a scoped or diff Codex Security check.
• The agent prepares the scan in the Codex Security app/workspace.
• The remaining required action is pressing Start scan.
• Computer Use cannot click it because targeting Codex itself is blocked, and the tool returns the safety refusal above.

The current workaround is manual clicking or waiting for an external scan start, which interrupts otherwise autonomous security remediation and review workflows.