regex-vs-llm-structured-text
Pass
Audited by Gen Agent Trust Hub on Jul 29, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill exhibits a susceptibility to indirect prompt injection within the
validate_with_llmfunction implementation. - Ingestion points: Untrusted content enters the agent's context through the
original_textparameter in thevalidate_with_llmfunction, which originates from external documents being processed. - Boundary markers: The prompt template directly interpolates the
{original_text}variable without using strong delimiters (e.g., XML tags or triple backticks) or explicit instructions to ignore any commands embedded within the text. - Capability inventory: The skill utilizes an LLM (
client.messages.create) to perform extraction and validation, returning the results to the calling application. - Sanitization: No sanitization or escaping logic is applied to the input text before it is included in the LLM prompt, potentially allowing an attacker to embed instructions that override the model's intended parsing logic.
Audit Metadata