analysis
Pass
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data extracted from binaries (such as strings and function names) which could contain malicious instructions designed to influence the agent's analysis or behavior.
- Ingestion points: Binary strings, import names, and function metadata are retrieved via SQL queries and processed within the agent's context (SKILL.md, references/network-detection.md).
- Boundary markers: Absent; data retrieved from the binary is presented as raw content without specific delimiters or instructions to ignore embedded commands.
- Capability inventory: The skill has access to local file system tools (Bash, Read, Glob, Grep) and can modify the IDA database structure (dirtree).
- Sanitization: No sanitization or escaping of binary-derived strings is documented before they are provided to the model for interpretation.
Audit Metadata