The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes curl as a fallback method to fetch web content if the preferred /read skill is missing. This is a standard operation for a research-oriented tool and is limited to retrieving content from URLs identified during the discovery phase.
[DATA_EXFILTRATION]: While the skill performs network operations to fetch external content, these are directed at primary sources (papers, blogs, repositories) for research purposes. No sensitive local files or credentials are accessed or transmitted.
[PROMPT_INJECTION]: The skill includes specific formatting instructions (e.g., prefixing lines with an emoji) and workflow constraints, but these are functional directives and do not attempt to bypass the agent's safety filters or core instructions.
[SAFE]: The research workflow is transparent, requiring user confirmation for modes and publishing. It explicitly instructs the agent to stop after the draft is ready and not to perform automated publishing actions without explicit request.

learn