The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses npx tsx to execute several local TypeScript scripts (phase-scrape.ts, phase-prospeo.ts, phase-enrich.ts, phase-upload.ts) to perform web scraping and API interactions.
[EXTERNAL_DOWNLOADS]: The scripts/phase-enrich.ts and scripts/phase-scrape.ts files perform network requests using fetch to scrape external company websites and interact with Prospeo and MillionVerifier APIs.
[INDIRECT_PROMPT_INJECTION]: The skill has an inherent attack surface for indirect prompt injection because it scrapes external website content (Phase 1/4) and uses it as input for Claude to generate company analyses and personalized email copy. A malicious website could theoretically include hidden instructions in HTML tags or metadata to influence the generated outreach content.
Ingestion points: External domains provided in --domain flags and leads' company websites (scanned in scripts/phase-enrich.ts and scripts/phase-scrape.ts).
Boundary markers: Not explicitly defined in the provided script logic for the scraping phases.
Capability inventory: File system writes (local JSON state), network operations (API calls, web scraping), and Smartlead campaign creation.
Sanitization: scripts/phase-scrape.ts includes a stripHtml function that removes script and style tags, which mitigates some direct code execution risks but does not prevent natural language instruction injection.

auto-research-public