codex-dispatch
Fail
Audited by Gen Agent Trust Hub on Jul 19, 2026
Risk Level: HIGHCREDENTIALS_UNSAFECOMMAND_EXECUTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill directs the agent to read and potentially modify
~/.codex/config.toml. As stated in the skill's own documentation, this environment is authenticated via OpenAI API keys or subscription tokens, making this file a high-value target for credential exposure. - [COMMAND_EXECUTION]: The skill employs a shell glob pattern (
node ~/.claude/plugins/cache/openai-codex/codex/*/scripts/codex-companion.mjs) to locate and execute a Node.js script. This dynamic path resolution for executable code allows for potential execution of unauthorized scripts if a malicious directory is introduced into the plugin cache. - [COMMAND_EXECUTION]: The instructions command the agent to use
grepandcatto inspect the internal logic of plugin scripts and model caches (~/.codex/models_cache.json), which can expose system metadata and operational details. - [CREDENTIALS_UNSAFE]: The skill explicitly notes that the sandbox allows reads outside of the workspace and encourages providing absolute paths to files. This capability can be leveraged to exfiltrate sensitive files (such as SSH keys or environment variables) by passing them to the Codex dispatch service.
Recommendations
- AI detected serious security threats
Audit Metadata