The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted content from the internet, creating a vulnerability surface for indirect prompt injection attacks where malicious instructions on a website could influence the agent's behavior.
Ingestion points: The skill utilizes the scrapling MCP server tools (fetch_page, fetch_dynamic, fetch_stealthy) and provided Python scripts to ingest HTML and JSON data from arbitrary URLs.
Boundary markers: The guidance does not specify the use of delimiters or 'ignore' instructions to separate external content from the agent's primary instructions.
Capability inventory: The agent is granted capabilities to write to the local filesystem (to_jsonl, save_image), execute shell commands via mcporter, and perform further network operations.
Sanitization: No sanitization or escaping of the fetched external content is mentioned or implemented before the data is processed by the agent.
[COMMAND_EXECUTION]: The skill encourages the execution of shell commands for installing dependencies, running MCP servers, and using the mcporter CLI to interact with scraping tools.
[EXTERNAL_DOWNLOADS]: Recommends the installation of the scrapling library via pip and the download of browser binaries using playwright install chromium.
[DATA_EXFILTRATION]: While designed for scraping, the tools provide a mechanism for reading data from external network sources into the agent's environment, which could be misused to exfiltrate information if the agent is directed to sensitive internal URLs.
[REMOTE_CODE_EXECUTION]: A recipe in references/spider-recipes.md uses xml.etree.ElementTree.fromstring to parse remote sitemaps. The Python documentation warns that this module is not secure against maliciously constructed XML data, which could lead to vulnerabilities when processing untrusted input.

scrapling-mcp