The Agent Skills Directory

Indirect Prompt Injection (HIGH): The skill parses untrusted tool definitions and Zod schemas from src/server/modules/mcp*/server.ts. Because the agent uses this data to generate and then suggest the execution of TypeScript scripts, a malicious actor could embed instructions in tool descriptions to compromise the system.
Ingestion points: src/server/modules/mcp*/server.ts (tool names, descriptions, and schemas).
Boundary markers: None specified. The skill assumes tool definitions are safe to extract and interpolate directly into code templates.
Capability inventory: The skill has file system write access to ./scripts/sdk/ and triggers command execution via npx tsx for generated scripts.
Sanitization: No sanitization or validation of the extracted metadata is performed before it is used to generate executable code.
Dynamic Execution (MEDIUM): The skill generates and promotes the execution of dynamic TypeScript code at runtime based on patterns it discovers in the project.
Evidence: The workflow involves generating client.ts and examples/*.ts files using templates and extracted metadata, which are subsequently recommended for execution to interact with backend services.

ts-agent-sdk