prompt

Free

Design a multi-project agent OS with Workspace isolation, white-box memory, and always-on execution.

FreeFree tier

Type

Open Source

Visit Website GitHub (40 stars)

About prompt

The 'WorkSpace-Isolated Agent OS Architect' is a system prompt designed to guide the creation of a productivity-oriented agent operating system. It emphasizes Workspace-based isolation as the fundamental unit, where each project has its own filesystem, memory store, skill set, and cost ledger to prevent context pollution. The prompt outlines core design principles including white-box memory (visible, editable, rollback-capable), model-tier routing (using appropriate models per task difficulty), ambient background execution (continuing work after user steps away), and native MCP (Model Context Protocol) integration. The goal is to build a multi-project agent OS that operates consistently across Web, CLI, and IM front-ends.

Key Features

WorkSpace isolation with per-project filesystem, memory, skills, and cost ledger

White-box memory with visible, editable, pin-able, and rollback-capable storage

Smart model routing: flagship model only for complex tasks, lighter models for simpler calls

Ambient background execution: tasks continue and deliver results as files with traceable audit trails

Native MCP (Model Context Protocol) integration for tool discovery, auth, and invocation

Context firewalls between Workspaces to prevent token or memory leaks

Operates consistently across Web, CLI, and IM front-ends

Pros & Cons

Pros

Strong project isolation prevents cross-contamination of files, memory, and skills
White-box memory allows full visibility and control over agent's knowledge
Smart model routing saves costs by using appropriate model tiers per task
Background execution enables long-running tasks without user supervision
MCP-native design ensures modular and extensible tool integration

Cons

Requires significant infrastructure to implement Workspace isolation and ambient execution
White-box memory adds complexity for users who prefer simple black-box agents
Not a standalone tool; requires a platform to implement the architecture described
May be overkill for simple single-project agent needs

Best For

Designing a multi-project agent operating system for productivity automationBuilding an agent platform with strict isolation and context boundariesImplementing white-box memory systems for agent workflowsCreating a cost-optimized model routing system for agent tasksDeveloping ambient agent execution that works asynchronouslyArchitecting MCP-native tool integration for agent platforms

FAQ

What is the 'WorkSpace-Isolated Agent OS Architect'?

It is a system prompt that provides a design blueprint for building a multi-project agent operating system with strict Workspace isolation, white-box memory, smart model routing, ambient background execution, and native MCP integration. The prompt was contributed by PilotDeck (OpenBMB / THUNLP / ModelBest / AI9Stars) and is part of the ai-boost/awesome-prompts collection.

What is Workspace isolation in this context?

Each project (Workspace) gets its own filesystem, memory store, skill set, and cost ledger. There are no cross-mounts by default, and memory retrieval is bounded to the active Workspace. Shared knowledge requires explicit import with version pinning.

Does this prompt provide a ready-to-use agent OS?

No, it is a design specification and architectural blueprint. Developers need to implement the described system using appropriate technologies and platforms. The prompt guides the design of such an OS.

What is 'ambient execution'?

Ambient execution means the agent can discover candidate tasks, run long-horizon monitors, and land results as local files while the user is away. It reports back with structured summaries, not chat noise.