RuslRusl
Getting started

Install Rusl, then make agents resolve contracts first.

The CLI vendors schema dependencies into your project. The Agent Kit teaches Claude Code to query before guessing, reuse before inventing, request context when blocked, and contribute evidence from real usage.

Install the CLIInstall Agent Kit
required
Install the CLI
The CLI installs, locks, and vendors schema dependencies from the Rusl registry into your project.
optional
Sign in when you need writes
Reading the graph works unauthenticated. Use the login command for write-side features like feedback and proposals.
Claude Code
Install the Agent Kit
The marketplace plugin gives Claude Code the four Rusl behaviors and auto-registers the MCP server.
manifest
Connect a project
The project manifest records bundle dependencies so humans, tools, and agents resolve the same contracts.
CLI

Install the CLI on your machine.

The CLI owns dependency installation, lockfiles, local schema vendoring, and authenticated write-side workflows.

Homebrew
brew install rusl-labs/tap/rusl
macOS and Linux
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/rusl-labs/rusl-cli/releases/latest/download/rusl-installer.sh | sh
Windows
powershell -ExecutionPolicy Bypass -c "irm https://github.com/rusl-labs/rusl-cli/releases/latest/download/rusl-installer.ps1 | iex"
Verify the install
Login is only required for write-side features. Discovery and graph reads can work without authentication.
rusl --help
rusl login
Agent Kit

Install the Claude Code plugin from the marketplace.

The plugin works without an init step. On first use in a project, it creates an empty bundle manifest so MCP work can be attributed correctly.

Rusl Agent Kit quick start
Install once in Claude Code. The plugin registers the Rusl server as a stdio MCP server and exposes four Rusl skills.
# In Claude Code, add the marketplace
/plugin marketplace add rusl-labs/rusl-agent-kit

# Install the plugin
/plugin install rusl
/rusl:resolve
Query before guessing
Load annotations, examples, and provenance before changing schemas, types, validators, tool inputs, or API payloads.
/rusl:reuse
Reuse before inventing
Search for a canonical bundle before introducing a new domain concept.
/rusl:feedback
Contribute from real usage
After schema-touching work, search, endorse, then create new evidence only when no canonical annotation exists.
/rusl:proposals
Iterate on proposals
Author schemas, comment on reviews, and move contract changes through the proposal workflow.
Requirements
Claude Code with plugin support, the Rusl CLI available to your shell, and login for feedback, endorsement, and proposal writes.
Agent Kit repository
Project manifest

Connect the project with a bundle manifest.

The manifest makes contract dependencies explicit enough for humans, build tools, and agents to inspect.

rusl.bundle.toml
[rusl.resources]
"rusl/bundles/feedback-schemas" = "~>0.1"
"rusl/schemas/common" = "~>0.1"
CLI workflow
rusl add rusl/bundles/feedback-schemas
rusl install
rusl tree
Local or live

Use installed schemas locally, or fetch live contracts when the workflow needs freshness.

Build and review workflows usually want pinned local files. Runtime and agent workflows may need direct resolution.

Local installed schemas
Use resolved contracts for build steps, code generation, validators, form generation, review, and tests.
Live resolution
Fetch schema contents and semantic annotations directly when an agent or app needs the latest resolvable contract.
Then use the schema with your stack
Rusl resolves standard JSON Schema files. From there, use the ecosystem to generate code, validate payloads, render forms, or give agents context before they write code.
Open tools guide
Agent instruction

Make the four behaviors explicit.

Agents need rules they can execute, not a vague suggestion to be careful with schema meaning.

Agent instruction
Before creating or changing a payload shape:
1. Run /rusl:resolve to load the contract, examples, annotations, and provenance.
2. Run /rusl:reuse before hand-authoring a new domain concept.
3. Request context only when the contract meaning is genuinely blocked.
4. Run /rusl:feedback after schema-touching work to endorse or contribute evidence.
Available now
The current agent guide is available as a project-readable instruction file. Link it from project instructions so agents resolve Rusl concepts before making local data decisions.
Open agent guideCLI releases