What's the Minimum Architecture a Modern Agent Harness Needs?

Описание и применение

Минимальная архитектура агентского хранилища, описанная в MiniDSH, использует журнал сессий как источник правды, чёткие границы владения и политики, и единый runtime для всех интерфейсов.

Задача

Designing a minimal yet architecturally complete agent harness that clearly defines truth, state ownership, permissions, composition, crash recovery, and extensibility.

Результат

MiniDSH demonstrates that a small, well‑structured harness can satisfy core architectural invariants, enabling clear reasoning about truth, state, and extensibility while remaining open to future evolution.

Рабочий процесс

• 1. Identify the minimal set of invariants needed for a modern agent harness.
• 2. Define clear ownership and lifecycle for each component (models, tools, persistence, authority, context).
• 3. Use an append‑only session log as the canonical source of truth.
• 4. Route model proposals through a policy layer that decides effect execution.
• 5. Keep surfaces (terminal, browser, headless) over a shared runtime.
• 6. Before adding a feature, map its ownership, dependencies, lifecycle, and durable state to the existing architecture.
• 7. Conduct hardening/review sessions to falsify and refine the design.

Доказательства

the append‑only session log is the canonical source of truth; models, tools, persistence, authority and context have explicit seams and owners; the model proposes an effect, but policy and the effect boundary decide whether it happens; the model is a routed capability, not synonymous with the agent; terminal, browser and headless are surfaces over one runtime; something gets a first‑class composition boundary only if it actually needs to be replaced, scoped, removed or independently owned.

Источник