The Agent Skills Directory

[SAFE]: A comprehensive audit of the skill's instructions, advisor dossiers, and templates revealed no malicious behavior. The skill contains no executable code, avoids obfuscation, and does not perform any suspicious network operations. It utilizes local project directories (e.g., '.agents/') for context and custom data storage, which is a standard and safe practice for AI agents.
[PROMPT_INJECTION]: The skill was evaluated for Indirect Prompt Injection vulnerability surface (Category 8) due to its data ingestion capabilities.
Ingestion points: The skill ingests data from '.agents/product-marketing.md' and custom advisor dossiers created by the user in '.agents/advisors/'.
Boundary markers: The skill establishes 'Grounding Rules' and requires a mandatory simulation disclaimer at the beginning of the output to prevent the AI from adopting unverified instructions as absolute truth.
Capability inventory: The skill is restricted to reading and writing text files within the local project directory and performing web searches via standard research tools to verify citations.
Sanitization: The instructions explicitly forbid the AI from inventing views for private individuals and mandate the use of primary sources (books, blogs, newsletters) for public figures, ensuring the output remains grounded in verifiable data.

marketing-council