eval-builder
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill directs the agent to read and process various external project files (DNA rules, FlowMind definitions, and Lev configurations), which constitutes a potential attack surface for indirect prompt injection if those files are poisoned.
- Ingestion points: The skill specifies multiple ingestion points in the 'Project resolution' section, including
.lev/config.yaml,dna/core/eval.dna.yaml, and.lev/pm/tasks/*/dna.yaml. - Boundary markers: The skill proactively mitigates this risk by instructing the agent that 'Deterministic code validates schemas... before quality' and requiring 'immutable execution requirements' with pinned digests.
- Capability inventory: The agent is expected to use file-system read capabilities to inspect the listed project references.
- Sanitization: The instructions mandate deterministic schema validation and the use of a 'deterministic assurance compiler' to process rules before they influence agent behavior.
Audit Metadata