The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from competitor websites, which is a common vector for indirect prompt injection attacks.
Ingestion points: Untrusted data enters the agent context through website scraping via firecrawl_map and firecrawl_scrape in Phase 1 of the research process.
Boundary markers: The skill contains a 'Core Principles' section that explicitly instructs the agent to treat scraped pages and reviews as data to analyze rather than instructions to follow, specifically warning about hidden HTML directives.
Capability inventory: The skill utilizes network-read capabilities through Firecrawl and DataForSEO tools and has the ability to write structured profiles and raw data files to the competitor-profiles/ directory on the local filesystem.
Sanitization: The instructions mandate that the agent should ignore any embedded commands in external data and note the discovery of such attempts in the generated profile instead of executing them.

competitor-profiling