evals-init
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data by scanning project manifests (e.g., package.json, requirements.txt, Cargo.toml, go.mod) to detect the tech stack and recommend evaluation frameworks.
- Ingestion points: Phase 1 (Tech Stack Detection) in SKILL.md.
- Boundary markers: The instructions do not specify boundary markers or delimiters for the content read from these external manifest files.
- Capability inventory: The skill has the capability to create directory structures and write multiple files, including executable Python scripts in the evals/graders/ directory.
- Sanitization: No explicit sanitization or validation of the content found within project manifests is documented before it influences the skill's logic or output.
- [DYNAMIC_EXECUTION]: The skill generates multiple configuration and executable files at runtime, including Python-based security graders.
- Evidence: Phase 2 in SKILL.md describes the automated creation of graders/check_pii_leakage.py, graders/check_prompt_injection.py, and framework-specific configuration files (config.js or config.py).
- Context: While these appear to be standard templates for a security baseline, the generation and subsequent execution of these files represents a dynamic execution surface.
Audit Metadata