agent-tool-builder
Pass
Audited by Gen Agent Trust Hub on Aug 14, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documents how to create tools that process external data provided by an LLM (e.g., in the
get_weatherandcreate_calendar_eventexamples). This inherently creates an attack surface where maliciously crafted user inputs could attempt to influence tool logic. However, the skill explicitly mitigates this risk by advocating for validation gates, strict schema compliance, and input sanitization. - Ingestion points: Tool arguments in Python functions and MCP server request handlers.
- Boundary markers: The skill recommends using explicit validation logic and JSON Schema 'strict' modes.
- Capability inventory: Demonstrates network interactions (API fetches) and implies database operations.
- Sanitization: Explicitly mandates input validation before execution and parameterized queries for SQL to prevent injection attacks.
- [COMMAND_EXECUTION]: The provided implementation patterns use standard SDK methods and do not demonstrate unsafe execution of shell commands or arbitrary code. The use of MCP and Anthropic SDKs follows official developer guidelines.
- [EXTERNAL_DOWNLOADS]: The skill references standard library dependencies from established ecosystems (Node.js/npm and Python/PyPI) for tool development. These include the Anthropic SDK, Zod, and the Model Context Protocol (MCP) SDK, which are standard for the described use cases.
Audit Metadata