The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data from external sources such as social media platforms (Reddit, Hacker News, X, YouTube) and individual prospect websites. This requirement introduces a vulnerability surface for indirect prompt injection, where malicious instructions embedded in external content could attempt to manipulate the agent's logic during qualification or outreach drafting.
Ingestion points: The agent mines public discourse from community forums, social media threads, and business websites via research tools like Firecrawl and Browserbase (identified in SKILL.md and references/demand-signals.md).
Boundary markers: Instructions in references/demand-signals.md direct the agent to "escape every value" when generating reports based on collected data.
Capability inventory: The skill leverages web research, data enrichment APIs, and generates structured lead sheets in markdown and CSV formats.
Sanitization: The skill recommends paraphrasing external evidence and escaping values in output reports to minimize risks from third-party content.
[SAFE]: The skill includes extensive compliance documentation in references/compliance.md, detailing adherence to GDPR, CAN-SPAM, and CASL. It explicitly mandates the use of public business contact channels and prohibits bulk scraping or the use of illicitly sourced data.
[SAFE]: The skill integrates with well-known business tools (Apollo, ZoomInfo, Clay, Clearbit, etc.) for their intended and legitimate uses within a prospecting workflow. All referenced services are reputable industry providers.
[SAFE]: The skill utilizes a local CLI tool, github-prospects.js, for specialized data enrichment from GitHub. This is a standard developer-focused workflow and does not involve remote code execution from untrusted sources.

prospecting