The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The documentation provides a URL to download a setup script directly from the vendor's domain.
Evidence: https://scrapfly.io/install/byop-connector.sh?api_key=YOUR_API_KEY in SKILL.md.
[REMOTE_CODE_EXECUTION]: The setup instructions include downloading and piping a remote shell script directly to the command line.
Evidence: curl -fsSL ... | sh pattern used for BYOP connector installation.
[COMMAND_EXECUTION]: The skill documentation suggests using shell commands to install and manage vendor-provided egress tools.
Evidence: Usage of piped shell execution and scrapfly exit-peer CLI commands.
[DYNAMIC_EXECUTION]: The skill uses Playwright capabilities to execute JavaScript within the remote browser context for data extraction.
Evidence: Implementation of page.evaluate() in several code examples within SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill navigates to and retrieves content from arbitrary websites, creating a pathway for untrusted data to enter the agent context.
Ingestion points: Data retrieved through page.content(), page.title(), and inner_text() after visiting URLs with page.goto().
Boundary markers: No delimiters or explicit instructions to ignore embedded commands are present in the code examples.
Capability inventory: Browser navigation, JavaScript execution, and local file writing via page.screenshot().
Sanitization: No sanitization or filtering logic is demonstrated for web content before it is passed to the AI agent.

scrapfly-browser