The Agent Skills Directory

[SAFE]: The skill operates within a strictly defined restricted environment. All file creation and modification activities are limited to the .research/, .requests/, and .research-archives/ directories. Metadata and workflow instructions align with the stated research orchestration purpose.
[COMMAND_EXECUTION]: The skill utilizes basic filesystem commands (ls, mkdir, zip, rm) for managing research artifacts. It includes explicit 'Mandatory Safety Rules' in both investigate.md and orchestrate.md that forbid the use of sudo, su, or doas, and block destructive commands outside of the authorized research directories.
[PROMPT_INJECTION]: The skill implements a 'Clarifier' philosophy, requiring the agent to restate its understanding and obtain explicit user confirmation via ask_user before proceeding with any research launch. This design mitigates risks of unintended action execution. Sub-agent prompts in orchestrate.md use structured delimiters (e.g., <objective>, <context>) to separate system instructions from user-provided topics.
[DATA_EXFILTRATION]: No network-capable utilities or external API communication patterns were detected. All research parameters and synthesis results are handled locally within the project structure.

research