proto-lab
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill acts as an intermediary that ingests untrusted user input to define 'contracts' for code generation. This creates a surface for indirect prompt injection where a user might include instructions meant to bypass the safety guidelines of the downstream prototype generator.
- Ingestion points: Conversation turns in Phase 1 (Listen) and Phase 2 (Propose) in
SKILL.md. - Boundary markers: None present; the skill does not instruct the agent to delimit or ignore instructions embedded in the user's prototype ideas.
- Capability inventory: The skill directs the creation of functional artifacts ('Build' phase) via a delegated local skill (
../prototype/SKILL.md), which are designed to be 'trivial to run'. - Sanitization: No validation or sanitization is performed on the synthesized user requirements before locking the contract.
Audit Metadata