The Agent Skills Directory

Management of User-Defined Instructions: The skill establishes procedures for handling custom instructions created by users. Since these instructions are eventually loaded into the agent's execution context, they represent a potential surface for indirect prompt injection. The skill notes that the system may reject certain content during the creation or publishing phase, which serves as a validation boundary.
Ingestion points: User-provided content passed to the create and publish tools.
Boundary markers: Mentions of system rejection for invalid descriptions or instructions.
Capability inventory: Access to tools (create, publish, set_mode, remove) that alter the agent's behavior or persona.
Sanitization: Dependent on the validation logic of the underlying tools referenced in the instructions.
Automated Tool Interaction and Approvals: The instructions direct the agent to use specific tools for lifecycle management and to rely on system-generated approval cards for confirmation. This process is designed to ensure that significant changes, especially in shared environments, are reviewed through the platform's native authorization mechanism rather than through conversational confirmation.

manage