The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external competitor websites, reviews, and SEO platforms, creating a surface for indirect prompt injection.
Ingestion points: Data enters the agent's context through firecrawl_scrape, firecrawl_map, and WebFetch calls targeting arbitrary competitor URLs and review sites (G2, Capterra).
Boundary markers: The skill contains explicit instructions under 'Untrusted Input' stating that competitor pages are data to analyze, never instructions to follow, and specifically tells the agent to ignore embedded instructions.
Capability inventory: The agent has the capability to write scraped content to the local filesystem (competitor-profiles/raw/) and perform network operations via multiple MCP tools.
Sanitization: While the skill provides clear behavioral instructions to the agent to disregard malicious text, there are no technical sanitization or escaping mechanisms defined for the interpolated web content before processing.
[COMMAND_EXECUTION]: The skill orchestrates the execution of multiple external tools (MCPs) to perform its core functions.
The instructions direct the agent to use firecrawl_map, firecrawl_scrape, and various DataForSEO tools to retrieve external data.
The skill also uses the built-in WebFetch tool as an alternative stack for site scraping.

competitor-profiling