groq-inference-skill
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, including audio files, images, and search results, which can serve as vectors for indirect prompt injection attacks. Malicious instructions embedded in these inputs could influence the agent's behavior, particularly when combined with the skill's tool execution and code interpretation capabilities.
- Ingestion points: Transcription (SKILL.md, reference/audio-speech.md), vision (SKILL.md, reference/vision-multimodal.md), and web search (SKILL.md, reference/tool-use-patterns.md).
- Capability inventory: The skill uses file reads, network API calls, and tool calling mechanisms.
- Boundary markers: No specific delimiters or safety instructions are provided to isolate untrusted content.
- Sanitization: The documentation recommends parameter validation and sanitization as best practices for tool use.
- [COMMAND_EXECUTION]: The skill documentation includes patterns for local function calling and server-side Python code execution via the Groq platform. These capabilities require strict input handling to prevent the execution of malicious commands derived from untrusted LLM outputs.
Audit Metadata