The Agent Skills Directory

[SAFE]: The skill implements a set of non-negotiable 'Grounding Rules' designed to prevent the AI from fabricating quotes or inventing endorsements. This is a safety best practice for persona-based simulations.
[COMMAND_EXECUTION]: The skill instructions allow for a 'Live Research Pass' that can utilize external skills like deep-research, watch-video, or standard web searches to find current information about advisors. This capability is scoped to the primary purpose of information gathering.
[DATA_EXFILTRATION]: The agent is instructed to read local project files such as .agents/product-marketing.md or .claude/product-marketing.md to gain business context. This is a legitimate use of project data for personalized assistance and does not involve accessing sensitive system credentials or exfiltrating data to untrusted external domains.
[PROMPT_INJECTION]: The 'Add a Custom Advisor' feature facilitates the creation of new persona files based on user-supplied data or project files. This functionality represents an indirect prompt injection surface where a malicious dossier could attempt to influence the agent's behavior. However, the skill provides a strict template and instructions for the agent to interview the user rather than inventing content, which acts as a safeguard.

marketing-council