The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to use tools such as web_scrape and cloud_browser_open to fetch and interact with external websites. This establishes an attack surface where malicious instructions embedded in the target website's content could potentially influence the agent's logic or downstream actions.
Ingestion points: Identified in SKILL.md through the use of web_scrape, cloud_browser_open, and take_snapshot.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions for content retrieved from external sources.
Capability inventory: The agent has access to network operations (via scraping and browser tools) and can perform stateful interactions like form filling or alert management (alert_create).
Sanitization: There is no mention of sanitizing or filtering external content before it is processed by the agent.
[DATA_EXFILTRATION]: The skill references connection to a remote Scrapfly MCP endpoint for tool discovery and execution.
Evidence: Reference to https://mcp.scrapfly.io/mcp in the Setup section of SKILL.md as a valid MCP endpoint.

scrapfly-agent-rules