codex-mcp-researcher
Pass
Audited by Gen Agent Trust Hub on Sep 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The agent configuration allows for potential indirect prompt injection by processing untrusted external data.
- Ingestion points: The agent utilizes
web_searchand has full project read access as defined inagents/openai.yaml. - Boundary markers: The
system_promptinagents/openai.yamldoes not implement delimiters or instructions to ignore commands embedded in fetched data. - Capability inventory: The agent possesses tools for reading the codebase and writing to the
./artifactsand./research-reportsdirectories (agents/openai.yaml). - Sanitization: The instructions in
agents/openai.yamldo not specify any validation, filtering, or sanitization of external content before processing.
Audit Metadata