skills/dzhng/skills/codex/Gen Agent Trust Hub

codex

Fail

Audited by Gen Agent Trust Hub on Nov 11, 2026

Risk Level: HIGHCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [PRIVILEGE_ESCALATION]: The skill directs the agent to use the --dangerously-bypass-approvals-and-sandbox flag when tasks require browser verification, dev servers, or full test runs. This flag explicitly disables the security sandbox and approval requirements for the Codex CLI, granting it full OS access.
  • [COMMAND_EXECUTION]: The skill provides instructions for the agent to modify the ~/.codex/config.toml file to set trust_level = "trusted" for specific directory paths, which elevates the execution context for the Codex CLI. It also uses various shell utilities like mktemp, git, and kill to manage the lifecycle of secondary agent processes.
  • [EXTERNAL_DOWNLOADS]: The skill provides instructions to download and install the Codex CLI from https://developers.openai.com/codex/cli, which is an official resource provided by OpenAI.
  • [INDIRECT_PROMPT_INJECTION]: The skill involves passing repository contents and user tasks to a secondary agent (Codex). There is a risk that malicious content within the repository (e.g., in documentation or code comments) could influence the behavior of the secondary agent. The skill attempts to mitigate this by requiring the agent to author prompts "end-to-end" and avoid relaying third-party text.
  • [DYNAMIC_EXECUTION]: The skill revolves around the execution of code generated by the Codex model within the local workspace. When the sandbox is bypassed, this generated code runs with the user's full privileges on the local machine.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Nov 11, 2026, 11:01 AM