v0.20.17 - installation types & harness matrix

One source.Every agent.

Install OpenPack packages, skills, plugins, and MCP sources into every agent runtime.

Browse catalogue Install

Search by meaning

Try:

Runs locally · first search downloads about 50 MB

openpack.json

skillsrulesinstructionscommandssubagentsMCPhookssettingspluginsfragments

OC OpenClaw.openclaw/

CL Claude Code.claude/

CX Codex CLI.codex/

CP GitHub Copilot CLI.github/

HE Hermes.hermes/

+ Your harnessJSONC / JS adapter

Install methods

Start from the CLI, hand the bootstrap markdown to an AI agent, or install only the companion skills you want in your runtime.

CLI Install Agentwheel

Use the package manager on your machine, then initialize a workspace or install directly from the catalogue.

terminal

npm i -g agentwheel
agentwheel init

AI handoff Give install.md to an agent

The markdown tells an agent how to install Agentwheel, verify it, add the companion skill, and use the catalogue.

fetch prompt

curl -fsSL https://raw.githubusercontent.com/NestDevLab/agentwheel/main/install.md

Skill Choose explicit or proactive

Install agentwheel for explicit management. Add agentwheel-discovery when you want automatic semantic suggestions and read-only trials.

codex local

agentwheel doctor --adapter codex --local
agentwheel install github:NestDevLab/agentwheel --adapter codex --local --skill agentwheel
agentwheel install github:NestDevLab/agentwheel --adapter codex --local --skill agentwheel-discovery

Next install a resource

Browse the catalogueFind packages, skills, plugins, and MCP servers with copy-ready Agentwheel commands.
Choose scope explicitlyUse --local, --user, or --fleet <fleet-id>. Named fleets are optional.
Preview when neededAdd --dry-run before broad fleet changes or plugin execution.

resource install example

npx agentwheel install github:NestDevLab/agent-must-have#standard --adapter codex,claude

Runtime matrix

Cells show how Agentwheel can surface each OpenPack capability today: files, directories, config-backed entries, runtime-only behavior, planned work, or unsupported targets.

Adapter	Instructions	Rules	Skills	Commands	Subagents	MCP	Hooks	Settings	Plugins	Fragments
OpenClaw	md user	no	dir	runtime	dir + json	json	runtime	json	dir	compose
Claude Code	md	md	dir	md	md	json	json	json	dir	compose
Codex CLI	md	no	dir	no	toml	toml	json	todo	dir	compose
Hermes	md	no	dir user	runtime	yaml knobs	yaml	runtime	yaml	dir	compose
GitHub Copilot CLI	md	md	dir	md	md	json	json	json	dir	compose

Glossary Open the legend for what each badge means in the matrix.

mdMarkdown surfaceAgentwheel writes Markdown files or managed Markdown blocks the runtime reads directly. md userUser-only MarkdownThe runtime has a supported user-level Markdown surface, but no project-local target is claimed. jsonJSON configThe capability is delivered through a JSON file or JSON deep-merge. yamlYAML configThe capability is delivered through YAML config, usually via deep-merge. tomlTOML configThe runtime consumes TOML files or TOML-backed config entries. dirNative directoryAgentwheel places a package directory where the harness discovers it. dir userUser-only directoryThe runtime supports the directory at user scope; local/project discovery is not claimed. dir + jsonDirectory plus JSONThe runtime needs files plus JSON registration, for example OpenClaw subagent workspaces and allowlists. json settingsJSON settings-backedThe capability exists through settings, for example OpenClaw agents.list[]. yaml knobsYAML knobsThe runtime has config knobs, for example Hermes delegation:, not separate artifact files. runtimeRuntime-onlyThe runtime provides behavior, but there is no managed artifact to install. noUnsupported targetAgentwheel does not currently manage this capability for the built-in adapter. todoPlannedThe target is intended but not implemented yet. composeAgentwheel-onlyFragments are composed into other artifacts; they are never installed directly. Artifact-level matrix

Rules and runtime-only features Behavioral guidance for Codex, OpenClaw, and Hermes belongs in instructions, not rules. Config-backed cells mean the capability is managed through the settings/config merger; mixed cells pair config with native files.

How it works

1 Author packageOpenPack repo

2 Resolve graph.agentwheel lock

3 Install nativeHarness surfaces

What stays aligned

artifacts skills/rules/MCP drift-safe plan/apply adapters built-in/custom sources git/local/registry/MCP selective --skill/--select fleet local/SSH

OpenPack manifest

openpack.json

{
  "schemaVersion": 2,
  "name": "your-org/agent-pack",
  "requires": {
    "core": {
      "source": "github:your-org/core-pack",
      "version": "^1.2.0"
    }
  },
  "provides": [
    { "type": "skills", "path": "skills" },
    { "type": "rules", "path": "rules" },
    { "type": "fragments", "path": "fragments" }
  ]
}

Spec openpack.md Graph design CLI README

Package details

Optional named fleets

User, local, and fleet desired state stay isolated. No fleet has global priority. If another scope owns the same runtime path, preview an explicit normalization transfer and apply only its reviewed plan digest.

named fleet

$ agentwheel install --fleet example-fleet --profile daily --dry-run

Registry entry

index.json

{
  "name": "your-package-name",
  "source": "github:your-org/your-agent-package",
  "type": "package",
  "tags": ["skills", "rules", "instructions"]
}

Asset includes

openpack.json

{
  "type": "skills",
  "path": "skills",
  "assets": [{
    "from": "packages/tmux-bridge/bin",
    "into": "bin",
    "include": ["*.sh"]
  }]
}

Selective install

selective install

$ agentwheel add github:NestDevLab/agent-mesh \
    --skill codex-tmux \
    --adapter codex \
    --installation-type local
$ agentwheel install --skill codex-tmux

MCP registry source

mcp-registry

$ agentwheel install mcp-registry:publisher/server-name \
    --adapter claude \
    --local \
    --dry-run

ClawHub source

clawhub

$ agentwheel install clawhub:@openclaw/whatsapp \
    --adapter openclaw \
    --local \
    --dry-run

Catalogue and registry

Catalogue Registry

publish draft

$ npx agentwheel@latest registry publish https://github.com/owner/repo \
    --description "Reusable skills and rules for coding agents." \
    --tag skills,rules

Advanced usage

Choose how proactive Agentwheel should be.

agentwheel gives your agent explicit management guidance. The separate agentwheel-discovery skill is opt-in: add it only when you want capability gaps to trigger semantic recommendations and non-installing trials.

companion skill

$ agentwheel doctor --adapter copilot --user
$ agentwheel install github:NestDevLab/agentwheel --adapter copilot --user --skill agentwheel --dry-run
$ agentwheel install github:NestDevLab/agentwheel --adapter copilot --user --skill agentwheel-discovery --dry-run
$ agentwheel doctor --adapter codex --local --skill syncwheel --json

Plan/apply loop

full control

$ npx agentwheel init
$ npx agentwheel add github:your-org/agent-pack --adapter codex --installation-type local
$ npx agentwheel plan
$ npx agentwheel install

Trust and offline

trust

$ agentwheel install --trust "github:your-org/*"
$ agentwheel deps tree
$ agentwheel deps why skills/triage
$ agentwheel install --offline

Fleet config

fleet mode

{
  "schemaVersion": 3,
  "fleetId": "example-fleet",
  "packages": [{
    "name": "core-agent-pack",
    "source": "github:example-org/core-agent-pack",
    "driver": "git",
    "adapter": "codex",
    "installationType": "local",
    "mode": "tracking"
  }],
  "agents": {
    "lab-codex": { "adapter": "codex", "installationType": "local", "root": "/workspace/project" }
  }
}

$ npm i -g agentwheel@latest
$ agentwheel fleet register example-fleet --root /srv/agentwheel/fleets/example-fleet --required-package core-agent-pack
$ agentwheel fleet list
$ agentwheel fleet show example-fleet
$ agentwheel install --fleet example-fleet --agent lab-codex --dry-run
$ agentwheel plan --fleet example-fleet --agent lab-codex --recover-legacy-state --force-conflict --replace-conflict
$ agentwheel fleet normalize example-fleet --from user --package core-agent-pack --json
$ agentwheel fleet normalize example-fleet --from user --package core-agent-pack --plan-digest <reviewed-sha256> --apply
$ agentwheel fleet normalize example-fleet --from user --recover

Docs

install.md AI handoff AGENT.md agent guide llms.txt LLM map README quick start Companion skill recommended setup OpenPack spec manifest format Dependency design graphs and locks DESIGN.md architecture LIFECYCLE.md update model Fleet config agents and SSH GitHub repo source Package registry short names