The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content from job postings, LinkedIn profiles, and professional blogs to generate research and outreach drafts. This creates a potential surface for indirect prompt injection where malicious instructions could be embedded in these public sources. * Ingestion points: SKILL.md (Step 1 and 2) identifies external sources including job postings, engineering blogs, and LinkedIn search result pages. * Boundary markers: The skill requires explicit user confirmation of the contact identity and the exact text of every message ('yes or no') before proceeding, ensuring no autonomous outreach. * Capability inventory: The skill can read/write local JSON files (pipeline.json), access the user's resume, and utilizes platform connectors for Gmail and Calendar. * Sanitization: outreach.py implements a research_problems function and a PERSONAL regex filter to identify and exclude non-professional or private information from the research process.

hunter-connector