The Agent Skills Directory

[PROMPT_INJECTION]: The skill describes an automated pipeline that processes external data from Aitable and lead capture forms, creating a surface for indirect prompt injection.
Ingestion points: External data enters the system through Aitable Clients row updates and the POST /api/leads/landing-page endpoint.
Boundary markers: The skill mentions phone number validation and rate limiting, but does not specify explicit boundary markers to isolate instructions within the generated strategies or landing page content.
Capability inventory: The system uses an lp-builder worker which involves Stitch MCP scaffold and Claude Code wire, suggesting the generation of code or configuration based on input data.
Sanitization: Basic phone number validation and deduplication are implemented, but broader content sanitization for the strategy doc generation is not detailed.
[COMMAND_EXECUTION]: The architecture describes a dynamic build process where an lp-builder worker utilizes Stitch MCP scaffold and Claude Code wire. This pattern indicates the runtime assembly and potential execution of code scaffolding or configurations during the landing page generation phase.
[DATA_EXFILTRATION]: The skill contains a hardcoded Cloudflare R2 bucket identifier (pub-f1692e774ca04e3b9e495f7d3c85a759.r2.dev) within the anti-patterns section. While listed as a negative example to prevent developers from hardcoding such strings, it exposes an internal infrastructure identifier.

lead-pages