Agent System makes an OpenClaw agent workspace self-onboarding: run openclaw agent-system install there to register and identify the agent, reconcile its supported configuration, and equip its managed tools to operate with that agent's own environment and credentials instead of a shared global identity.
Note
Requires OpenClaw 2026.7.1-2 or newer. CI covers macOS 26 and Ubuntu 24.04.
Warning
Agent System is still a work in progress. Check back regularly for updates.
Today, Agent System:
- registers an agent workspace with OpenClaw and reconciles its public identity
- assembles environment variables and credentials per agent from declared dotenv, inline, and 1Password sources
- wraps supported tools with the active agent's declared configuration, environment, credentials, and workspace boundaries
- applies each tool's operation-specific
allow,ask, ordenypolicy before resolving credentials or executing the operation - validates manifests, installs configured components, projects executable paths, and reports installed-state drift
Install the current release from npm:
openclaw plugins install npm:@tanaab/openclaw-agent-system
openclaw plugins enable agent-systemFor a development checkout, follow Install from source.
Add .agent-system/agent.yaml to the workspace you want Agent System to manage. A root-level agent.yaml is also supported as a shorthand.
schema-version: 1
agent:
id: tanaabot
name: Tanaabot
email:
from-environment: AGENT_EMAIL
environment:
# import this agent's identity and tool credentials from 1password.
op: z7q4m2n9v6k3p8r5t1w0x4c2ba
set:
SSH_KEY:
from-op: 'op://v4u7l2t9n5p8r1c6x3z0m4q7da/ssh-key/private key?ssh-format=openssh'
required:
- AGENT_EMAIL
- GH_TOKEN_TANAABOT
- SSH_KEY
github:
username: tanaabot
token: GH_TOKEN_TANAABOT
git:
ssh:
private-keys:
from-environment: SSH_KEYFrom that workspace, store the 1Password bootstrap credential when needed, then validate and install the agent:
# persist the current 1password service account token for this agent.
openclaw agent-system credentials set op --from-env
# validate the manifest, then reconcile the agent and its configured components.
openclaw agent-system validate
openclaw agent-system install
# inspect managed state without changing it.
openclaw agent-system doctor
# verify the github identity supplied by this agent's environment.
openclaw agent-system tool gh -- api user --jq .logininstall is explicit and repeatable: it adds the OpenClaw agent when needed and reconciles only the state declared by the workspace. See Advanced for the complete manifest and CLI references.
Agent System currently ships wrappers for:
Tip
For managed agents, disable competing Git and GitHub skills, plugins, and tool wrappers so requests consistently use Agent System's agent-scoped identity and policy boundaries.
Each wrapper uses the shared Agent System runtime for trusted agent binding, environment and credential resolution, operation policy, execution, redaction, and auditing. A public Tool API is planned so other OpenClaw plugins can add compatible wrappers through the same runtime.
Important
Model-facing agent_system_* tools are the agent-bound execution surface.
openclaw agent-system tool, credentials, and the packaged command shims are
trusted operator interfaces: a process with unrestricted host command access
can select another installed agent or workspace. Run doctor to inspect this
exposure, and use native tools for agent work.
See Development for source installation, the recommended DevGuard workflow, validation, and coding standards.
Use the GitHub issue queue for bugs and feature requests.
See CHANGELOG.md for implemented changes and GitHub releases for published artifacts.
Made with contrib.rocks.
Agent System is licensed under the MIT License.