pizza1
Pass
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill employs behavioral override instructions that command the agent to exclude specific information from its output. These patterns direct the model to disregard standard reasoning categories (convention, ergonomics, etc.) in subsequent turns.
- Evidence: The "Banned justifications" section explicitly instructs: "Delete these from the next response."
- Evidence: The "Contract" section mandates: "Drop every convention, ergonomics, churn, transition-size, and scope justification from the next response."
- [INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface by instructing the agent to process and evaluate external designs or codebase justifications based on a specific, non-conventional logic. If an adversary provides code comments or documentation designed to manipulate the agent's "correctness" derivation, the agent's judgment could be redirected.
- Ingestion points: Processes user-provided code, designs, and developer justifications (SKILL.md).
- Boundary markers: None present; there are no delimiters to separate instructions from the data being analyzed.
- Capability inventory: No subprocess calls, network operations, or file system write operations are defined in the skill scripts.
- Sanitization: No input validation or sanitization mechanisms are implemented to filter instructions embedded in the analyzed codebases.
Audit Metadata