The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted client materials (decks, research docs, audits) stored in the 'materials/' directory during the research phase. Malicious instructions hidden within these external documents could potentially manipulate the agent's behavior during plan synthesis. No explicit boundary markers or 'ignore' instructions are provided to mitigate this risk. Evidence: Ingestion point at ~/marketing-plans/{client-slug}/materials/; Capability inventory includes file-write, network operations via MCPs, and subprocess calls; Sanitization is absent.
[EXTERNAL_DOWNLOADS]: The skill documentation references and suggests the installation of external tools such as 'agent-browser' and the use of various MCP-based services (Ahrefs, Customer.io, Stripe). These are well-known technology services required for the stated marketing functionality.
[DATA_EXFILTRATION]: The skill includes functionality to read sensitive financial and customer data via Stripe and GA4 MCPs and offers to publish the compiled plan to external GitHub repositories. This represents a data transfer surface, although it is designed as a user-directed feature for client strategy sharing.

marketing-plan