The Agent Skills Directory

[SAFE]: The skill serves as a meta-framework for analytical reasoning and does not contain malicious patterns or obfuscated code.
[COMMAND_EXECUTION]: The skill uses curl to send status updates to a local voice notification service at http://localhost:8888/notify. This is a documented feature for user feedback and targets a whitelisted domain (localhost).
[COMMAND_EXECUTION]: The BeCreative sub-skill utilizes the llm CLI tool (e.g., gemini-3-pro-preview) to execute specialized model tasks for technical creativity.
[DATA_EXFILTRATION]: Network operations were evaluated; all detected curl calls are restricted to localhost, and no sensitive system or user data is transmitted.
[PROMPT_INJECTION]: The skill processes user-provided inputs for analysis and debate. It possesses a potential surface for indirect prompt injection, which is managed via structural boundaries.
Ingestion points: User-provided 'problems', 'challenges', and 'strategies' processed across all 53 workflow files.
Boundary markers: The skill consistently uses Markdown headers (##), blockquotes, and XML-like tags (<instructions>) to separate system instructions from user data.
Capability inventory: Includes local curl notifications, llm CLI tool execution, and local file I/O for persistent world models.
Sanitization: No programmatic sanitization of user input is present; the skill relies on the structural separation defined in the templates.

thinking