The Agent Skills Directory

[SAFE]: The skill implements a robust review workflow where no external actions (sending emails) occur without explicit human approval within the companion app UI.
[SAFE]: Secret management follows best practices by using environment variables (e.g., BUSABASE_API_KEY) and explicitly instructing the user never to paste credentials into chat or logs.
[INDIRECT_PROMPT_INJECTION]: The skill processes external email content and subscriber segments from Busabase. It mitigates injection risks by using a quality gate (email-quality-auditor) and sanitizing data before rendering in the companion app UI.
Ingestion points: Data is fetched from the sends, segments, and suppression Busabase bases within busabase-provider.js.
Boundary markers: Mandatory human approval steps and the EQS scoring system provide a functional gate before any processing affects the send queue.
Capability inventory: The skill uses busabase-sdk to read and write records, and a separate script execute_decisions.mjs for scheduling authorized sends.
Sanitization: Employs escapeHtml in the frontend (e.g., in app.js and campaign-views.js) to prevent XSS and related data injection issues.
[SAFE]: Dependencies are restricted to standard, versioned packages from the NPM registry such as busabase-sdk and hono.
[SAFE]: Network operations are restricted to the configured Busabase instance and the author's documentation site.

kelly-campaigns