The Agent Skills Directory

[SAFE]: No security threats or malicious behaviors were detected during the analysis.
[INDIRECT_PROMPT_INJECTION]: The skill processes disclosure metadata from Busabase Bases.
Ingestion points: Disclosure item content is read in busabase-provider.js and tracker-model.js to populate the review UI.
Boundary markers: The skill lacks explicit prompt delimiters for external data, but uses strict domain logic to compute item statuses, reducing the risk of accidental obedience.
Capability inventory: The skill can write review decisions and execution markers back to Busabase using the busabase-sdk.
Sanitization: UI components utilize escapeHtml in app.js to prevent cross-site scripting when rendering item summaries.
[COMMAND_EXECUTION]: Utility scripts provided in the scripts/ directory (setup.mjs, generate_batch.mjs, execute_decisions.mjs, publish_airapp.mjs) are standard tools for managing the Busabase workspace and seeding mock data. They operate within expected parameters and require explicit user execution.
[EXTERNAL_DOWNLOADS]: The skill uses established packages like busabase-sdk, hono, and esbuild-wasm, which are well-known within the platform ecosystem and are sourced from official registries.

kelly-disclosure-tracker