The Agent Skills Directory

[SAFE]: The skill implements a comprehensive 'Quarantine Classification' system specifically designed to identify and isolate adversarial content. It provides the agent with heuristics to detect indirect prompt injection patterns such as 'ignore previous instructions' or 'system:' markers within untrusted data.
[COMMAND_EXECUTION]: The instructions utilize the Bash tool to run the crwl (Crawl4AI) and docling (Docling) command-line interfaces. The skill uses quoted variables (e.g., "<url>") in its command templates, which is a standard practice to mitigate basic shell injection risks.
[EXTERNAL_DOWNLOADS]: The skill is configured to fetch web content from external sources as part of its primary research function. These operations are governed by budget constraints (max pages, max per domain) defined in a manifest, reducing the risk of unintended resource exhaustion or uncontrolled crawling.

research-collect