Core concepts

Agentrig is built around a few simple primitives.

Plugins

A plugin is a versioned Agent Plugins v1 package identified by the canonical id format namespace.plugin. Published versions carry plugin.json plus the required source, lock, and review artifacts.

Plugins are AgentRig's primary distribution unit. Their portable components are skills and MCP servers. AgentRig-specific hooks, commands, agents, rules, assets, scripts, and supporting files live under the ai.agentrig extension boundary and are translated into provider-native output.

Skills, MCPs, and hooks

Skills, MCPs, and hooks are first-class discovery views over plugin contents. They help you find the exact capability you need without manually digging through a large repo.

When you want the whole bundle, install the plugin:

agentrig plugin install codex agentrig/agentrig.core-committer

When you only want specific artifacts, install a Selection Bundle:

agentrig install codex agentrig/agentrig.core-committer \
  --pick skill:codex-analysis \
  --pick skill:gitwhat

Standalone skills can also be installed directly:

agentrig skill install codex agentrig/acme.review

Selection Bundles are local install records for selected artifacts. They keep the plugin source provenance and let AgentRig safely uninstall the selected files or JSON entries later.

Registries

Registries host plugin metadata and files. Public installation is allowed only from static signed registries.

• Official registry: https://agentrig.ai/registry
• Self-hosted registries: registries you host yourself, as long as they implement the same contract
• Submissions and community pages: discovery and review surfaces, not install sources

The directory (https://agentrig.ai/directory/index.json) is a curated list of registries for discovery. Directory presence does not imply install trust.

Registry install refs are latest-first by default. Use agentrig/<namespace.plugin> or agentrig/<namespace.skill> for normal installs, and add @<version> only when you need an exact reproducibility or rollback pin. AgentRig records the concrete resolved version and snapshot digest in its install ledger either way.

Rigs

A rig is a named set of plugins. Rigs can extend other rigs, allowing team-wide baselines plus project-specific add-ons.

Trust tiers

AgentRig trust tiers are:

• Official - installable
• Reviewed - installable
• Listed - discovery-only
• Blocked - hard blocked
• Yanked - historical or advisory only

Manifests

Plugin source directories carry plugin.json. AgentRig plugin manifests are Agent Plugins 1.0.0 conformant. name is the canonical plugin id, and AgentRig-specific fields live under extensions["ai.agentrig"].

The CLI stores exact source identity, snapshot digests, target paths, and owned hashes in personal or workspace .agentrig/plugin-installs.json ledgers so it can inspect and remove AgentRig-managed output safely later.