The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external, untrusted public sources including news outlets and competitor websites as described in the SKILL.md and README.md. This ingestion of external content presents a surface for indirect prompt injection attacks.
Ingestion points: The agent is instructed to "browse current public sources", "news sources", and "competitor URLs" to collect evidence-backed signals.
Boundary markers: The skill incorporates a "review cockpit" (App UI) where a human operator must review and approve signals, actions, and drafts. Instructions specify that outbound messages and regulated claims are "approval-required".
Capability inventory: The skill uses busabase-sdk to write interpreted data to a database and includes a script (scripts/execute_decisions.mjs) to update record statuses after a human has performed an external handoff.
Sanitization: The application includes escapeHtml utility functions in app/app/app.js and app/app/js/connect-gate.js to sanitize data when rendering it in the UI.
[COMMAND_EXECUTION]: The skill includes a server component (app/server.js) and a decision-tracking script (scripts/execute_decisions.mjs). These components facilitate the storage and management of data within a Busabase workspace. The execution of the execute_decisions.mjs script is controlled and primarily serves to log actions and update statuses in the database rather than executing arbitrary code derived from untrusted input.
[EXTERNAL_DOWNLOADS]: The skill depends on several external Node.js packages, including hono, @hono/node-server, and busabase-sdk. These are well-known libraries in the ecosystem. The application also connects to busabase.com for OAuth authentication and data storage, which is the intended purpose of the skill.

kelly-ai-newsroom