The Agent Skills Directory

[SAFE]: The skill is a well-structured simulation tool. It uses static markdown dossiers to ground persona behavior and provides clear instructions for simulation boundaries.
[INDIRECT_PROMPT_INJECTION]: The skill has a potential attack surface by reading external project files (.agents/product-marketing.md) and custom advisor files (.agents/advisors/*.md). However, this risk is mitigated by explicit grounding rules, simulation disclaimers, and instructions to paraphrase content rather than execute embedded instructions. This is a standard functional requirement for context-aware agents.
Ingestion points: Project context files (.agents/product-marketing.md, etc.), custom advisor dossiers, and live research results (web/video search).
Boundary markers: Output is wrapped in a simulation-labeled structure with mandatory disclaimers.
Capability inventory: File reading and web search (via platform tools); no dangerous system-level capabilities.
Sanitization: Instructions require paraphrasing and verification of quotes, preventing direct execution of external strings.
[REMOTE_CODE_EXECUTION]: No remote code execution patterns were found. The 'Live Research Pass' uses standard platform tools for searching public information.
[OBFUSCATION]: No obfuscated content, Base64 strings, zero-width characters, or homoglyph attacks were identified in any of the 15 files.

marketing-council