The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read .claude/product-marketing-context.md for context before creating content. If an attacker can modify this file, they could inject instructions that override the agent's behavior. Additionally, the 'Reverse Engineering' section (references/reverse-engineering.md) suggests using scraping tools like Apify or manual research to collect large amounts of social media posts for analysis. If the scraped content contains malicious instructions (e.g., in HTML comments or post metadata), the agent could be vulnerable to indirect prompt injection when analyzing this data.

social-content