code-review
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [Indirect Prompt Injection Surface]: The skill ingests untrusted code data via the
@$1argument (such as PR text, file paths, or diffs) to perform analysis. If the code or PR comments under review contain adversarial text or injection prompts, the agent could potentially be influenced by those embedded instructions. - Ingestion points: The user-supplied payload is passed via the
@$1placeholder inSKILL.md. - Boundary markers: The prompt lacks explicit structural delimiters or specific isolation instructions to separate the input code from the system instructions.
- Capability inventory: The skill contains no active shell execution commands, file modification tasks, or external network calls within the analyzed file.
- Sanitization: No input filtering or content sanitization is defined for the target code payload prior to processing.
Audit Metadata