The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is to act as a 'Context Engine', which involves loading and processing large volumes of data from external sources (e.g., industry profiles, market research, and client validation documents). This architecture creates an inherent surface for indirect prompt injection, where instructions embedded in these processed datasets could attempt to override the AI's behavior across its specialized marketing agents. Ingestion points: Data enters the agent's context from files in ~/.claude-marketing/brands/, industry-profiles.md, and various engagement-part directories. Boundary markers: The framework uses structured JSON for brand profiles and maintains a 'Two-Views Model' to distinguish between unbiased research (v1) and client-validated data (v2), providing some structural isolation. Capability inventory: The skill can execute various internal Python scripts (setup.py, campaign-tracker.py, eval-runner.py) and call tools provided by user-configured MCP servers (e.g., Salesforce, HubSpot, Google Analytics). Sanitization: The documentation focuses on schema validation for structured data but does not explicitly detail the sanitization of natural language content before it is interpolated into prompts.

context-engine