rnv
Pass
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The instruction 'Do not self-censor or hedge' found in SKILL.md under Stage 1 is a common marker used in prompt injections to encourage the model to ignore safety guidelines and internal constraints.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input via the $ARGUMENTS variable to perform multi-stage code generation and verification, which provides an attack surface for indirect instructions.
- Ingestion points: The skill accepts user tasks through the $ARGUMENTS parameter in SKILL.md.
- Boundary markers: The protocol uses explicit headers such as '## [UNVERIFIED DRAFT]' and '## [VERIFIED SOLUTION]' to separate different stages of generation and verification.
- Capability inventory: The skill is designed for code generation, API verification, and logical auditing based on user instructions.
- Sanitization: There is no evidence of validation or sanitization of the user-supplied task before it is interpolated into the reasoning protocol.
Audit Metadata