ARDS v4.0 Specification

The open standard for agent-ready documentation and workspaces

One format. Every AI coding agent. Define your project context once — all major AI coding agents understand it automatically. New in v4.0: the same standard now extends into the cloud workspace, so the documents your team shares and the context your agents see are the same objects.

5 AI tools. 5 config formats. 0 standards.

Every AI coding agent invented its own way to read project context. Your team maintains duplicate files that drift out of sync.

Before

Claude CodeCLAUDE.md
Codex CLIAGENTS.md
Cursor.cursorrules
GitHub Copilot.github/copilot-instructions.md
Windsurf.windsurfrules
replaced by

After

SurfContextCONTEXT.md + .context/ + surfcontext.json

v4.0: from file standard to workspace standard

Through v3, context was something a repository presented to an agent. In v4.0, knowledge and context unify: the documents a team shares and the context an agent sees are the same objects, living in a cloud workspace the agent reaches through MCP. The file layer is unchanged and fully valid on its own — every workspace feature is additive and optional, and a valid v3 project is a valid v4 project.

Workspace Model & Scopes

Five scopes — personal, workspace-private, workspace, repo, public — with hard, credential-enforced isolation boundaries.

Workspace as Command Center

Knowledge lives and agents launch in the workspace. The git repo becomes tooling shell and archive, with a file-layer offline fallback and mandatory export.

Tasks & Stages

An 8-stage lifecycle from todo to done, with stage-bound artifacts that serve as agent memory across sessions.

Agent Launch & Playbooks

Declared tool allow/deny lists, stage-authority ceilings, heartbeats, and metering for every launched agent.

Skill Runtime & Executable Plans

The dated plan doc becomes the runtime's input — journaled deterministic resume with budget ceilings.

MCP Integration

A versioned, append-only tool vocabulary with capability tiers, gated availability, workspace-pinned credentials, and OAuth 2.1.

How it works

Three files. Universal compatibility. Zero lock-in.

  1. 1

    Write your context5 min

    Create CONTEXT.md and a .context/ directory with your agent definitions and knowledge docs.

  2. 2

    Target platforms2 min

    Add surfcontext.json to declare which AI platforms to generate config files for.

  3. 3

    AI auto-discovers0 min

    Open your project in any supported AI tool. It reads your context automatically — no copy-paste, no per-tool config.

  4. 4

    Connect a workspaceOptional

    Adopt the v4 workspace layer when you're ready. Agents reach the same knowledge through MCP, and the file layer remains your export and offline fallback.

Platform compatibility

SurfContext generates the right config file for each platform from a single source of truth.

Generated FileStatus
CLAUDE.md
AGENTS.md
.cursorrules
.github/copilot-instructions.md
.windsurfrules

Ready to standardize your AI context?

Get started in under 5 minutes. No dependencies, no build step, no lock-in.

Get Started Read the Full Spec