The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill incorporates a 'live research pass' that ingest data from external web sources and video transcripts to update its advisor simulations. This functionality introduces an inherent surface for indirect prompt injection, where maliciously crafted external content could attempt to subvert the agent's instructions.
Ingestion points: External web search results, YouTube transcripts (via optional watch-video skill), and user-provided marketing questions defined in SKILL.md.
Boundary markers: The skill requires the agent to explicitly label sessions as simulations and provides 'Grounding rules' in SKILL.md to maintain persona fidelity and prevent unauthorized assertions.
Capability inventory: The skill depends on separate research and video analysis tools for external access; it contains no native logic for shell execution, file system modification, or unauthorized network operations within the provided files.
Sanitization: The instructions emphasize the use of primary sources and verification but do not describe specific automated sanitization or filtering of fetched data.
[EXTERNAL_DOWNLOADS]: The skill references and suggests the use of external helper skills and documentation hosted on public code repositories to fulfill its research requirements.
Evidence: Mentions of the web-research skill in the Quegenx/agent-skills repository and the original source repository coreyhaines31/marketingskills in SKILL.md.

marketing-council