Hopper is a structured sequential task orchestrator for human-agent collaboration. It is designed to standardize the "hand-off" between messy developer intent and precise AI agent execution.
The primary bottleneck in human-agent collaboration is execution latency. Humans think 10x faster than agents can execute. While an agent is fixing a single bug, the developer has already processed the next three steps (add tests, document, refactor).
Today, those thoughts are either lost, stored in disparate notes, or dumped into a chat window where they overwhelm the agent's context and focus. Hopper allows you to offload these thoughts into a DAG-aware queue immediately, ensuring agents only see one actionable objective at a time while you maintain the "Thought Queue."
Hopper is designed to be interface-agnostic. It serves as the single source of truth for:
- IDE Agents: Cursor, Windsurf, Copilot, etc.
- Terminal Agents: Aider, OpenDevin, and CLI-based automation.
- Manual Verification: Humans reviewing the structured intent before execution.
This is an exploratory project in its early stages. It is designed to pave the way for a fully autonomous development environment by standardizing the hand-off between human intent and agent execution.
Note: This project was built using agent-led development, with a human orchestrator directing the build process through the Hopper protocol itself.
-
Clone
git clone https://github.com/Mohamed1756/Hopper.git
-
Install
cd Hopper ./scripts/install.sh -
Start Run the following command inside any destination project directory:
hopper start
Hopper utilizes a two-pane Mission Control architecture:
- Orchestrator Pane: Capture and groom intent bursts.
- Monitor Pane: Real-time visibility into the active intent mirror (@.hopper/ACTIVE_GOAL.md).
Protocol specifications and design docs are located in the Project/ directory.