The Agent Skills Directory

[SAFE]: The skill implements a secure 'review-then-execute' pattern for all platform mutations (bids, budgets, keywords). No actions that incur costs or change campaign settings are performed automatically; they are queued as adjustment cards for human approval.
[SAFE]: Data ingestion is handled via trusted scripts (ingest_reports.mjs) that validate payloads and normalize metrics before writing to the database. These scripts perform no direct network calls to ad platforms, relying on the agent to gather data using the user's local credentials.
[SAFE]: sensitive information management is handled correctly. Ad platform credentials are never stored in the database or passed into the UI; they remain in the user's local environment, following a 'least privilege' model.
[SAFE]: The AirApp itself is limited to reading and writing Busabase records, maintaining a clear security boundary by not interacting with external networks or executing arbitrary commands.
[SAFE]: All dependencies, such as busabase-sdk and hono, are standard and well-known libraries for building agent-integrated applications.

kelly-ads