cobol-openapi-generator
Pass
Audited by Gen Agent Trust Hub on Nov 13, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill implements a dedicated prompt injection guard that explicitly instructs the agent to treat all copybooks, COBOL comments, and repository files as untrusted data. It provides clear directives to follow only user messages and the skill's own rules, mitigating the risk of instructions embedded in source code.
- [DATA_EXFILTRATION]: The skill includes non-negotiable rules against including secrets, credentials, tokens, or personal data in the generated contracts or reports. No suspicious network activity or exfiltration patterns were identified in the associated scripts.
- [DYNAMIC_EXECUTION]: The provided scripts use secure practices such as
yaml.safe_load()andyaml.safe_dump()for configuration and contract handling, preventing potential vulnerabilities associated with unsafe deserialization of YAML data. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (COBOL files) which could contain hidden instructions. It mitigates this risk through a multi-gate workflow, structural parsing using a dedicated script, and a final validation step that ensures the output follows a strict schema, rather than obeying content within the data.
- [COMMAND_EXECUTION]: All shell command executions, such as calling the validator script, are predefined and controlled within the skill's workflow. There are no pathways for user-supplied data to influence command arguments in a way that would allow for arbitrary command injection.
Audit Metadata