groq-migration-deep-dive
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEDYNAMIC_CONTEXT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file uses the dynamic context injection pattern (
!npm list...) to execute a shell command that detects installed LLM SDKs upon loading. This is a legitimate use of project-specific tooling for environment discovery and does not perform any sensitive operations.\n- [INDIRECT_PROMPT_INJECTION]: The skill provides an automated migration scanner that processes local source code and configuration files, creating a potential surface for indirect prompt injection.\n - Ingestion points: The script in
references/implementation.mdreads content from thesrc/directory and.env*files usinggrep.\n - Boundary markers: The scanner does not define specific delimiters to separate the content of analyzed files from the agent's processing instructions.\n
- Capability inventory: The skill utilizes
Read,Write,Edit, andBashtools to analyze and modify the codebase.\n - Sanitization: The assessment script processes file content via text-searching utilities without implementing specific validation or sanitization of the extracted data.
Audit Metadata