no-cutting-corners
Pass
Audited by Gen Agent Trust Hub on Jun 19, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill provides instructions focused on quality control and thoroughness. It explicitly notes that its mandates are not an override for confidentiality, instruction hierarchy, privacy, or security boundaries, ensuring it does not encourage bypassing safety filters.
- [DATA_EXFILTRATION]: A detailed 'Disclosure Boundary' section is included, which instructs the agent to identify and redact secrets, PII, and customer data. This ensures that sensitive information is protected and not transmitted externally, even when the agent is asked for 'complete' reporting.
- [EXTERNAL_DOWNLOADS]: The skill contains references to official research and documentation from well-known organizations such as OpenAI, Anthropic, and ArXiv for conceptual grounding. These references are static and do not involve downloading or executing untrusted remote code.
- [COMMAND_EXECUTION]: While the skill mentions the use of tools like 'curl' and 'npm test', these are provided as examples of how an agent should verify its claims rather than as a vector for arbitrary command injection. The instructions emphasize using deterministic methods and verification receipts.
Audit Metadata