The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands like mkdir, cp, mv, touch, find, and grep to manage the file system. These operations are restricted to the agent's internal skill directories (~/.claude/skills/).
[DYNAMIC_EXECUTION]: The skill is designed to create TypeScript tools and execute them using the bun runtime. It includes instructions for mapping natural language user intents to CLI flags, which is a standard pattern for the described system.
[DATA_EXFILTRATION]: The skill performs curl POST requests to http://localhost:8888/notify for local voice and text notifications. Since the target is localhost, this does not represent an external exfiltration risk.
[INDIRECT_PROMPT_INJECTION]: The skill includes mechanisms to load user-defined customizations from a specific local directory (~/.claude/PAI/USER/SKILLCUSTOMIZATIONS/CreateSkill/). While this provides an entry point for external data, it is a documented feature for user preferences.
Ingestion points: Reads user preferences and resources from a dedicated local path.
Boundary markers: None; the skill assumes valid configuration files.
Capability inventory: File system modification, local network notifications, and execution of local TypeScript files via bun.
Sanitization: None; relies on the agent's internal logic to process and apply the loaded configurations.

CreateSkill