ai-observability-promptfoo
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill describes how to implement custom evaluation logic using JavaScript and Python scripts. These are used for advanced validation scenarios, such as the
transformproperty for pre-processing outputs and custom assertion files loaded via thefile://protocol. - [EXTERNAL_DOWNLOADS]: The documentation facilitates the use of the
npxutility to fetch and run thepromptfooCLI tool directly from the official NPM package registry. - [INDIRECT_PROMPT_INJECTION]: The skill addresses the evaluation of untrusted LLM outputs, which presents a surface for indirect instructions. It demonstrates safe patterns for handling this data, including the use of regular expressions to extract specific content and structured JSON parsing to validate response schemas.
- Ingestion points:
examples/core.md(transform),examples/model-graded.md(custom assertions). - Boundary markers: Nunjucks templating is used for variable interpolation in prompts.
- Capability inventory: JavaScript and Python execution is available for complex validation tasks.
- Sanitization: Examples show data isolation via JSON parsing and regex extraction.
Audit Metadata