The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and process untrusted content from the web, which creates a potential surface for indirect prompt injection attacks where malicious instructions hidden in web pages or feeds could attempt to influence the agent's summary or subsequent actions.
Ingestion points: fetch_url, extract_page_content, and fetch_feed functions in scripts/research_utils.py ingest data from arbitrary external URLs.
Boundary markers: The skill does not explicitly define or enforce strict delimiters or instructions to the agent to disregard commands within the fetched content.
Capability inventory: The skill utilizes the requests library for network operations and writes to a local .research_cache directory for data persistence.
Sanitization: Content is processed using BeautifulSoup to remove script and style elements, but the resulting text is not sanitized for potential natural language instructions.
[EXTERNAL_DOWNLOADS]: The skill fetches data from several well-known and reputable security sources, including the National Vulnerability Database (NVD) at services.nvd.nist.gov, CISA Alerts, and the MITRE ATT&CK framework hosted on GitHub.
[SAFE]: The implementation includes a RateLimiter class and a ResearchCache mechanism in scripts/research_utils.py to ensure responsible use of external APIs and improve performance through local data storage.

research