The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions create a vulnerability surface for indirect prompt injection by directing the agent to read and trust product marketing context from files like .agents/product-marketing.md or .claude/product-marketing.md before performing campaign operations.\n- Ingestion points: The agent is instructed to read context from .agents/product-marketing.md, .claude/product-marketing.md, and product-marketing-context.md (SKILL.md).\n- Boundary markers: There are no explicit instructions to use delimiters or ignore potentially malicious instructions embedded in the ingested marketing context.\n- Capability inventory: The skill is intended to be used with powerful tool integrations (Google Ads, Meta Ads, LinkedIn Ads, TikTok Ads) that allow the agent to modify live advertising accounts and spend budget.\n- Sanitization: The instructions do not specify any validation or sanitization steps for the data retrieved from external files.

ads