prompt-engineering-patterns
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements various prompt template systems that use string interpolation to incorporate user-provided data into instructions for LLMs.
- Ingestion points:
PromptTemplate.renderandConditionalTemplate.renderinreferences/prompt-templates.mdinterpolate keyword arguments directly into templates.PromptOptimizerinscripts/optimize-prompt.pyalso formats templates with test case inputs. - Boundary markers: The skill documentation mentions "Safety guidelines" and "Verification" steps, but the core template implementations in
references/prompt-templates.mddo not automatically include delimiters or "ignore" instructions for interpolated content. - Capability inventory: The skill is designed to interact with LLMs (e.g., via the
openailibrary inreferences/chain-of-thought.md) which can perform complex tasks based on the generated prompts. - Sanitization: There is no evidence of input sanitization or escaping in the provided template rendering logic.
Audit Metadata