The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/render_pdf.mjs executes local Chromium or Google Chrome binaries via child_process.spawn to generate PDF sales collateral from HTML. The execution is limited to locally generated HTML files and utilizes specific, predefined binary paths based on common installation locations. The input HTML is generated from escaped fields, minimizing risk.\n- [PROMPT_INJECTION]: The skill processes data from public web sources (company websites, directories, news) during the research phase to identify potential leads, creating a surface for indirect prompt injection.\n * Ingestion points: The research subcommand and scripts/import_leads.mjs ingest data discovered by the agent.\n * Boundary markers: The skill relies on mandatory human approval for all outreach messages rather than explicit prompt delimiters.\n * Capability inventory: The skill can send emails (network) and execute local browser binaries (subprocess).\n * Sanitization: Content is escaped for HTML display, and the required manual review by the user provides the primary security control.\n- [SAFE]: Secret management for SMTP credentials is implemented using the platform's Vault system via scripts/configure_smtp.mjs. The skill avoids hardcoding credentials, uses environment variables for sensitive inputs to prevent command history logging, and stores only reference keys in the database to prevent plain-text exposure.

kelly-sales-outreach