gemini-cli-commands-demo

This project demonstrates a sub-agent orchestration system built within the Gemini CLI, using its native features to manage complex, asynchronous tasks performed by specialized AI agents. It showcases how prompt-driven commands can replace traditional programming logic to create a transparent and debuggable multi-agent workflow.

Key Features

• Filesystem-as-State — The entire system state (tasks, plans, logs) is stored in structured directories, eliminating external databases and making debugging straightforward.
• Prompt-Driven Commands — Orchestrator logic is defined in .toml prompts that create custom Gemini CLI commands (e.g., /agents:start, /agents:run), avoiding traditional code.
• Asynchronous Agent Management — Sub-agents run as background processes, with completion tracked via PID and .done sentinel files for status reconciliation.
• Specialized Sub-Agents — Includes agents like coder-agent and reviewer-agent, each with unique personas and constrained capabilities for specific tasks.
• Transparent Workspace — Agents operate in a dedicated .gemini/agents/workspace/ directory, where all outputs and modifications are visible and organized.

Philosophy

The system embraces simplicity and transparency by leveraging the filesystem as a state store and using prompts as the primary configuration method, making complex agent orchestration accessible without deep programming expertise.