The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user-supplied data regarding business goals, budget, and competitor tactics. This creates a surface for indirect prompt injection where a user or processed data could attempt to influence the agent's behavior. However, the skill includes clear boundary instructions and a rigid scoring rubric that requires the agent to justify recommendations with evidence, effectively mitigating this risk.
[DATA_EXPOSURE]: The skill instructions direct the agent to read context from specific local paths such as .agents/product-marketing.md and references/ideas-by-category.md. These are localized reference files intended for the skill's operation and do not expose sensitive user credentials or system-level configuration files.
[COMMAND_EXECUTION]: The skill defines clear boundaries that explicitly forbid the agent from launching campaigns, spending money, or creating accounts without authorization. There are no shell commands or subprocess calls identified in the instructions or provided metadata files.

suede-marketing-ideas