The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest external data such as client decks, audit reports, and live API data. While this presents an attack surface for indirect prompt injection, the skill implements a strict interactive review phase. Users must approve or adjust each of the 13 strategy sections before finalization, which serves as a robust security control against the agent autonomously following malicious instructions embedded in third-party data.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill accesses business metrics and file systems to perform its core function. It reads from local project directories (~/marketing-plans/) and uses authorized MCP integrations to fetch data from well-known services like Stripe and Google Analytics. All network operations and data gathering are part of the stated fractional CMO research methodology and are performed within established tooling frameworks.
[COMMAND_EXECUTION]: The documentation references the installation of the agent-browser package via npm. This is a common utility for automated research in agentic workflows and is used here in a transparent, documented manner.
[REMOTE_CODE_EXECUTION]: The skill mentions external repositories and documentation from trusted organizations, such as Anthropic and Vercel Labs, for reference and idea sourcing. These references follow safe practices and do not involve unauthorized code execution.

marketing-plan