The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill incorporates a specialized utility (scripts/render_pdf.mjs) for generating PDF collateral. This script utilizes child_process.spawn to invoke a local Chromium or Chrome instance. The execution is limited to rendering locally generated HTML files into PDF format and uses standard, non-shell execution patterns to prevent command injection.
[INDIRECT_PROMPT_INJECTION]: As the skill is designed to research target companies via public web URLs, it possesses an attack surface for indirect prompt injection. To address this, the skill instructions and the application logic enforce a strict human-approval workflow. Every email draft generated from external data must be manually reviewed and approved by the user in the dashboard before it can be sent, effectively neutralizing the risk of the agent following malicious instructions embedded in web content.
[COMMAND_EXECUTION]: The skill provides a suite of Node.js scripts for environment setup, data import, and outreach management. These scripts facilitate communication with the Busabase backend. The architecture ensures that sensitive actions, such as SMTP email transmission, are handled by trusted scripts in a local environment rather than the browser-based dashboard.
[CREDENTIALS_UNSAFE]: The skill manages SMTP credentials for email outreach. It correctly utilizes a secure 'Vault' mechanism via the busabase-sdk. Sensitive values like SMTP_PASS are never displayed in the UI or stored in plain text within the database; instead, they are accessed at runtime by the trusted sending script, which adheres to secret management best practices.

kelly-sales-outreach