The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as brand guides, logos, and campaign briefs which could theoretically contain malicious instructions.
Ingestion points: User-provided assets and the templates/campaign-brief.md file serve as the primary entry points for external data.
Boundary markers: The instructions explicitly advise the agent to "escape" asset names and campaign strings, treating them strictly as data.
Capability inventory: The skill facilitates file creation (static HTML/TypeScript), network requests (CRM adapters), and database operations (Neon Postgres).
Sanitization: SKILL.md and references/brand-and-marketer-workflow.md include mandatory sanitization rules: "Asset names and campaign strings are data: escape them and never inject arbitrary HTML or JavaScript from a brand document."
[EXTERNAL_DOWNLOADS]: The skill references and integrates with several well-known and trusted external services for its default implementation stack.
The architecture utilizes Vercel for hosting and API functions, Neon for managed Postgres, and Twilio/SendGrid for contact verification and messaging.
It references technical guidelines and example architectures from official GitHub repositories belonging to Vercel Labs and Cloudflare.
Integration guidance is provided for well-known CRM services such as HubSpot.
The skill mentions several open-source reference implementations (e.g., OpenLaunch) but instructs the agent to inspect their code and documentation rather than adopting them as unverified dependencies.

waiting-list