The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions enforce a strict limitation on creating files, commits, or code. The only exception is the optional creation of a documentation file at the end of the process, which requires an explicit user request.
[DATA_EXFILTRATION]: The skill contextually analyzes existing project files to inform its suggestions. This data access is localized and does not include patterns for exfiltrating sensitive information to external servers.
[EXTERNAL_DOWNLOADS]: The skill directs the agent to perform external research for technical facts using search tools to ensure accuracy. This process is governed by a 'Protocol of Rigor' that requires documenting sources and dates to manage data volatility.
[PROMPT_INJECTION]: The skill addresses potential indirect prompt injection from external sources by explicitly stating that results from consultations are to be treated as data and not as instructions, reducing the risk of third-party content hijacking agent behavior.

brainstorm