The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes merchant lead data (such as brand names, notes, and rejection reasons) from an external database. This is a potential surface for indirect prompt injection if the stored data contains instructions meant to bias the agent's summarization.
Ingestion points: Lead data is read from the leads Busabase Base via busabase-provider.js.
Boundary markers: The skill does not currently define explicit boundary markers in its instructions to separate lead data from agent instructions.
Capability inventory: The skill possesses the capability to write updates back to the database (e.g., changing lead stages or appending notes via records.changeRequest).
Sanitization: The application implements escapeHtml in its frontend logic (app.js, accent-theme.js) to prevent cross-site scripting (XSS), though prompt-level filtering for the agent is handled by the platform's standard guardrails.
[REMOTE_CODE_EXECUTION]: The skill includes developer-facing scripts (setup.mjs, publish_airapp.mjs) used for provisioning the application. These scripts utilize the vendor's own SDK (busabase-sdk) to interact with the Busabase API using credentials supplied by the user's local environment variables. This is standard behavior for skill deployment and does not represent an execution risk to the end-user.

kelly-lead-funnel