rlaope/oh-my-hermes

What it solves

It acts as an operating layer for the Hermes Agent, providing a structured environment to manage complex tasks. It solves the problem of unreliable agent output by ensuring that work is only marked as "done" when verified by actual tests or CI gates, rather than relying on the agent's own claims.

How it works

It organizes your goals into a graph of small, parallel nodes (using the /dag-ulw command). Each node is assigned a specific model and reasoning effort based on its category (e.g., coding, research, or engineering). It routes tasks to the best-suited model, manages long-term memory, and uses language server protocols (LSP) to ensure code changes are based on structural meaning rather than simple text matching.

Who it’s for

It is for users of the Hermes Agent who want more control, transparency, and reliability in their AI-assisted workflows, particularly those managing large-scale coding projects that require parallel execution and rigorous verification.

Highlights

• Mixture-of-models routing: Automatically selects the best model and effort level for each specific task lane.
• Evidence-based status: Clearly distinguishes between planned, running, and verified work, ensuring you never mistake a claim for a completed result.
• Parallel execution: Breaks large goals into a graph of smaller nodes that run simultaneously without conflicting.
• Native TUI: Provides a live terminal interface to monitor token usage, costs, and agent progress in real-time.
• Expert skills: Includes over a hundred specialized tools for tasks like security review, backend engineering, and testing.