The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The framework uses dynamic imports to load controller files from the filesystem.
Evidence: src/loader.ts lines 180-184 use import(path) with paths resolved from a glob search in the Loader.from method.
Risk: This allows the execution of arbitrary JavaScript/TypeScript code if an attacker can write files to the scanned controller directory.
[PROMPT_INJECTION]: The skill provides a framework for handling external HTTP requests, which creates an indirect prompt injection surface.
Ingestion points: src/find-my-way.ts and src/controller.ts (via ctx.params, ctx.headers, and ctx.request.body) expose request data to application logic.
Boundary markers: No specific boundary markers or 'ignore' instructions are implemented in the framework code to separate data from instructions.
Capability inventory: The framework supports dynamic code loading and network service creation.
Sanitization: No built-in sanitization or validation of external content is performed; security is deferred to the user implementation.

hile-http