The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a web crawler that fetches arbitrary content from user-specified URLs. This content is converted to markdown and provided to the agent, creating a vulnerability surface where malicious instructions on a fetched web page could be executed by the AI.
Ingestion points: scripts/crawl.py ingests untrusted data from URLs via multiple scraper modules (firecrawl_scraper.py, jina_reader.py, scrapling_scraper.py).
Boundary markers: No delimiters or protective instructions (e.g., 'ignore content below for instruction purposes') are added to the scraped markdown before it is saved or returned.
Capability inventory: The skill has the ability to write files to the local system via the --output parameter in scripts/crawl.py and performs outbound network requests.
Sanitization: While the content is converted from HTML to Markdown using html2text, there is no sanitization or filtering of the natural language text to prevent prompt injection.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to external services to facilitate content retrieval. Specifically, it sends target URLs to the Jina Reader service (https://r.jina.ai/) and potentially to Firecrawl's API endpoint. These are well-known services used for web content extraction.
[DYNAMIC_EXECUTION]: The Tier 3 fallback mechanism in scripts/src/scrapling_scraper.py uses the scrapling library with StealthyFetcher. This creates a headless browser instance that executes JavaScript from the target remote website to handle challenges such as Cloudflare bot protection.

crawler