Concepts
The five systems that make up the Sherpa framework — behavioral agents, governance, execution, conventions, and Studio.
Sherpa is built from five systems that each solve a distinct problem in Human+AI collaboration. Understanding these systems and how they connect is the fastest way to evaluate whether the framework fits your workflow.
The Five Systems
| System | What it solves | Learn more |
|---|---|---|
| Behavioral Agents | Agent roles that produce consistent, measurable behavior through constraints — not identity claims | Role schema, the 11 roles, evidence base |
| Governance | Coordination so multiple agents and humans can work on the same codebase without corrupting shared artifacts | Three-layer model, initiative lifecycle, conflict prevention |
| Execution Pipeline | Decomposing approved work into tasks, dispatching them across backends, and reviewing the output | Planner/Worker/Judge, 9 backends, 3 dispatch modes |
| Conventions and Config | Encoding process knowledge as executable artifacts — rules, skills, and project configuration | Auto-loading rules, 16 skills, sherpa.json |
| Studio | A centralized governance dashboard for visualizing and managing everything across multiple projects | Package architecture, multi-project federation, routes |
How the systems reinforce each other
These five systems are not independent features — they form a feedback loop. Behavioral agents define the roles that the execution pipeline dispatches. The governance engine tracks the initiatives and tasks that agents work on, using conventions to enforce quality and process constraints automatically. Studio makes the entire system visible: initiative lifecycles, agent missions, dispatch status, and convention compliance rendered in a single dashboard. When an agent completes work and a judge reviews it, the output feeds back into governance state, which updates Studio's view, which informs the next planning cycle. The result is a system where adding structure in one layer makes every other layer more effective.