agent-management

To create agents, prompts, and skills, instruct the system to generate files with appropriate naming conventions like PascalCase filenames, placing them in the correct ~/.agents subdirectories to maintain local configuration safely.

Agents perform tasks, skills define reusable capabilities, prompts provide specific instructions, and hooks trigger actions based on events. Decision guidance helps determine which primitive to use based on whether you need autonomous execution, reusable logic, or event-driven workflows.

Agent, skill, prompt, and instruction files are stored in the ~/.agents directory and local configuration repositories. These user-local files should not be committed to version control to protect your workspace state.

Safely edit agents and instructions by starting with read operations like list_dir to verify file locations, following naming conventions, and ensuring changes remain local within the ~/.agents subdirectories without committing user-local configuration files.

Yes, you can review past conversations to derive improvements for prompts and skills. The workflow outlines roles for each primitive, allowing you to analyze interactions and update instruction files to refine agent behavior and capabilities.

Create a new agent when existing primitives cannot accomplish the required task. Decision guidance helps determine if the needed functionality can be handled by updating prompts or skills instead of generating an entirely new agent configuration.