chatgpt-codex-onboarding
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions require the agent to execute Python logic via bash blocks to perform OAuth tasks like checking status or starting a login flow. This is the intended delivery mechanism for script-mode skills.
- [DATA_EXFILTRATION]: The skill performs network operations to a loopback address (
localhost) to notify the agent runtime to flush its cache after credential changes. It also interacts with OpenAI APIs for the OAuth handshake, which is appropriate for the skill's primary purpose. - [DYNAMIC_EXECUTION]: The
exports.pyscript dynamically modifies the Python search path (sys.path) to load core libraries from the environment. It also uses a temporary event loop to run asynchronous tasks within a synchronous script context. - [INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface where it ingests model definitions and usage data from OpenAI's APIs.
- Ingestion points: Data enters the system via
fetch_models_from_apiandfetch_account_usageinexports.py. - Boundary markers: None; data is processed as structured JSON.
- Capability inventory: The skill has the ability to write files to the workspace and make local network calls.
- Sanitization: Standard JSON parsing is used to handle external data.
Audit Metadata