The Agent Skills Directory

[PROMPT_INJECTION]: Deceptive Content and Bloating. The skill folder contains eight massive reference files (such as architecture-patterns.md, security-best-practices.md, and testing-strategies.md) that are not mentioned in the main SKILL.md reference list. These files consist of approximately 1,350 sections of repetitive technical jargon shuffled to mimic advanced documentation. This technique functions as a form of obfuscation by creating extreme noise, potentially designed to overwhelm security scanners or exhaust the agent's context window to degrade performance.
[PROMPT_INJECTION]: Indirect Prompt Injection Surface. In SKILL.md, the agent is instructed to ingest untrusted user-supplied data (such as analytics platform names, user journey descriptions, and North Star metrics) into a strictly formatted response. The skill lacks explicit boundary markers or instructions to disregard embedded commands in the input context, providing a surface for attackers to influence the agent's behavior through crafted metadata.

product-analytics