The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to scrape and process content from external websites (e.g., Binance Square). This content enters the agent's context, creating a vulnerability where malicious instructions embedded in web pages (such as hidden text or HTML comments) could attempt to override the agent's behavior.
Ingestion points: Web content extracted via cdp("Runtime.evaluate", ...) and innerText lookups in binance-square-scraper/SKILL.md and domain-skills/binance-square/scraping.md.
Boundary markers: The skill does not explicitly define sanitization or boundary markers for the scraped content before it is processed by the AI.
Capability inventory: The skill has extensive capabilities, including executing arbitrary Python, making network requests via http_get, and controlling the browser.
Sanitization: No explicit sanitization of DOM text is mentioned before it is returned to the agent.
[DYNAMIC_EXECUTION]: The browser-harness utility operates as a dynamic execution environment, accepting Python code via a heredoc or stdin and executing it locally. Furthermore, it uses the Runtime.evaluate CDP method to execute arbitrary JavaScript within the browser's context to extract data and manipulate the DOM.
[COMMAND_EXECUTION]: The instructions provide shell commands for system discovery and process management, such as querying localhost:9222 to find the browser's WebSocket URL and using pkill to reset the Chrome environment.
[EXTERNAL_DOWNLOADS]: The skill references and provides instructions for cloning the browser-harness repository from GitHub (browser-use/browser-harness). While this is a well-known project for browser automation, it represents an external code dependency that is required for the skill to function.

browser-harness