expose-tools-on-mcp
Pass
Audited by Gen Agent Trust Hub on Nov 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from external platform APIs that could theoretically contain malicious instructions if the upstream source were compromised.
- Ingestion points: The skill reads project metadata, tool definitions, and server configurations from the Speakeasy platform via
list_project_toolsandget_mcp(detailed inSKILL.md). - Boundary markers: There are no explicit delimiters defined for the data returned from platform tools.
- Capability inventory: The skill can modify administrative configurations using
create_mcp_from_functions,add_tools_to_mcp, andremove_tools_from_mcp(documented inSKILL.mdsteps 3 and 6). - Sanitization: The workflow significantly mitigates risk by requiring explicit user confirmation before any write operation and by strictly forbidding the agent from guessing identifiers or processing secrets.
Audit Metadata