The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection because it is designed to ingest and process data from external sources and user-supplied content.
Ingestion points: The 'Live Research Pass' feature in SKILL.md allows the agent to fetch and process data from web searches, video analysis tools, and recent news feeds. Additionally, the 'Adding a Custom Advisor' workflow allows users to provide their own frameworks and heuristics which are then stored in local files.
Boundary markers: The skill mitigates risks by requiring a mandatory simulation disclaimer at the top of every output and establishing 'Grounding Rules' that explicitly forbid the fabrication of quotes and endorsements.
Capability inventory: The skill is capable of writing files to specific project directories (.agents/advisors/) and invoking external capabilities such as web search, deep-research, and watch-video.
Sanitization: Instructions emphasize the use of primary sources, explicit citations, and paraphrasing techniques to ensure that external content is anchored in reality and identified as simulation.

marketing-council