diagnosis
Pass
Audited by Gen Agent Trust Hub on Jul 19, 2026
Risk Level: SAFENO_CODEPROMPT_INJECTION
Full Analysis
- [NO_CODE]: The skill consists only of instructional Markdown and evaluation JSON. It does not include any Python, Node.js, or shell scripts, eliminating the risk of direct malicious code execution.
- [PROMPT_INJECTION]: The skill identifies a surface for indirect prompt injection because user-provided lyrics are processed by the agent without explicit delimiters or escaping. While an adversary could include instructions inside a lyric to manipulate the critique, the impact is limited by the agent's internal safety guardrails and the lack of sensitive system capabilities.
- Ingestion points: The $ARGUMENTS variable in SKILL.md captures user input for audit and diagnosis actions.
- Boundary markers: None present; input is treated as raw text payload within the prompt instructions.
- Capability inventory: File system access is limited to reading bundled research context and optional project-level templates.
- Sanitization: No input validation or escaping logic is implemented.
Audit Metadata