consult-codex
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute the
codexCLI via the Bash tool. It explicitly requires the use ofdangerouslyDisableSandbox: trueto enable network access for the CLI tool to communicate with the OpenAI API. - [DYNAMIC_EXECUTION]: The instructions involve the dynamic construction of shell commands using randomly generated session tags, turn numbers, and variable file paths. This dynamic assembly is used for session management and output redirection.
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data (project files, diffs, and context) which may contain untrusted content.
- Ingestion points: Relevant project files, command outputs, and external feedback are gathered and passed to the consultation.
- Boundary markers: The skill recommends using XML tags (e.g.,
<task>,<grounding_rules>) to structure input and delineate instructions from data. - Capability inventory: The skill utilizes the Bash tool for command execution and the Write tool for creating context and prompt files.
- Sanitization: The instructions suggest writing prompts to temporary files and piping them via stdin to avoid shell interpretation of special characters, though it does not mandate specific sanitization of the file content itself.
Audit Metadata