codex
Pass
Audited by Gen Agent Trust Hub on Aug 8, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes the
codexCLI via a wrapper script (codex.sh). This is the core functionality of the skill and is implemented with security-conscious patterns, such as using stdin for the prompt to avoid shell injection and running inside a dispatched subagent to isolate the output. - [DATA_EXFILTRATION]: While the skill communicates with the OpenAI Codex API, this is the intended purpose of the tool. The analysis found no evidence of exfiltrating sensitive local files (e.g., SSH keys, AWS credentials) beyond the project context provided by the user for the task.
- [PROMPT_INJECTION]: The skill includes instructions to relay Codex's output verbatim. While this could theoretically be used to relay a malicious prompt from Codex to the main agent, the skill documentation explicitly warns that 'Codex's output is not ground truth' and mandates manual review of all diffs before they are accepted.
Audit Metadata