building-agents
Pass
Audited by Gen Agent Trust Hub on Aug 21, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill promotes secure development practices, including the use of credential connectors instead of hardcoded secrets and enforcing conversation isolation by authenticated user. It also advises on setting operational bounds like token budgets and step caps to prevent runaway loops.
- [INDIRECT_PROMPT_INJECTION]: The skill documents an architecture where agents process external data, creating an inherent surface for indirect prompt injection.
- Ingestion points: External data enters through the buildInput function and tools defined in the tools/ directory.
- Boundary markers: Developers are instructed to use system.md to define the agent's identity and constraints, including manual confirmation for irreversible actions.
- Capability inventory: Agent capabilities are defined by the tools implemented by the developer, which may include network or file system operations.
- Sanitization: The instructions recommend structured data formats and typed schemas for tools to reduce ambiguity, though input sanitization is left to the developer's implementation.
Audit Metadata