codex-subagents
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the user to run various CLI tools and diagnostic scripts (e.g.,
./bin/control,node subagent-report.mjs) to manage subagent settings and verify model status within the local repository. - [DYNAMIC_EXECUTION]: A troubleshooting command is provided that uses the
node -eflag to execute a script snippet. This snippet dynamically imports a local module (./src/agent-check.mjs) to perform live capability tests on models. - [INDIRECT_PROMPT_INJECTION]: The diagnostic script
subagent-report.mjsingests data from local configuration files to present the agent with the current state of available models. - Ingestion points: Reads
merged-models.jsonandmulti-agent-settings.jsonfrom the local~/.codex/codex-routerstate directory. - Boundary markers: None; the data is delimited by standard JSON structures.
- Capability inventory: The skill can execute local Node.js scripts and project-specific binaries to modify the router configuration.
- Sanitization: Data is ingested through standard Node.js JSON parsing, minimizing the risk of arbitrary code injection from these files.
Audit Metadata