security-threat-model
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted repository content, including source code and configuration files, which could contain malicious instructions designed to influence the agent's analysis.
- Ingestion points: The workflow in
SKILL.mdand the prompts inreferences/prompt-template.mddirect the agent to read and summarize files from a user-provided repository path. - Boundary markers: While the
references/prompt-template.mdprovides a structured system prompt and grounding rules, it does not utilize explicit delimiters or unique boundary markers to encapsulate the untrusted code content being analyzed. - Capability inventory: The skill utilizes file read operations, the Ripgrep (
rg) search tool, and filesystem write operations to save the final Markdown threat model (SKILL.md). - Sanitization: The skill includes instructions to redact sensitive secrets like tokens and passwords but does not specify methods for sanitizing or escaping code content to prevent instructions within the code from being interpreted as prompts by the underlying model.
Audit Metadata