The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes candidate data (names, platform headlines, and search contexts) and scoring criteria hypothesis provided by the user.
Ingestion points: Data is entered through the /kelly-instructor-sourcing review and criteria subcommands and the corresponding Hono-based web interface, stored in the kelly-instructor-sourcing-candidates and kelly-instructor-sourcing-criteria bases.
Boundary markers: The skill documentation explicitly states a 'sandbox-only' policy with deterministic data and lacks any capability to perform automated messaging or outbound platform connections, which significantly mitigates the impact of potential injection content.
Capability inventory: The skill can perform authenticated mutations on Busabase records using records.changeRequest and bases.createChangeRequest. It does not have access to shell execution, sensitive local files, or unauthorized network operations.
Sanitization: The application logic in app.js employs a robust escapeHtml utility to sanitize all user-supplied content before rendering it in the user interface, preventing cross-site scripting (XSS).
[REMOTE_CODE_EXECUTION]: The project includes a build script in package.json that utilizes esbuild-wasm to bundle the Busabase SDK for browser use. This is a standard development task for preparing vendor assets and does not involve the execution of untrusted remote code at runtime.

kelly-instructor-sourcing