prompt-engineering
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The
PromptEvaluatorandPromptOptimizerclasses inscripts/prompt_optimizer.pyexhibit an indirect prompt injection attack surface where untrusted data is processed. - Ingestion points: Data enters the agent context through the
test_case.inputvariable and theoutputreceived from the LLM, which are both used inscripts/prompt_optimizer.py. - Boundary markers: The script lacks clear boundary markers or delimiters when interpolating external content into its evaluation meta-prompts (e.g., in
_score_relevanceand_score_clarity). - Capability inventory: The skill possesses the capability to perform iterative LLM generations (
llm.generate) and modify prompt templates dynamically based on these results. - Sanitization: There is no evidence of input validation, escaping, or instruction-filtering for the content being interpolated into the scoring prompts.
Audit Metadata