@claxedo/agent-extensions
@claxedo/agent-extensions lets a host install one package and materialize the supported components into runner-native locations: skills, MCP server configs, and local runner plugins. The package owns the deterministic mechanics (source parsing, fetch/cache, lock files, install lifecycle, materialization, and owned-artifact cleanup); product hosts own authorization, catalog allowlists, and fan-out to workspace runtimes.
Install
Section titled “Install”npm install @claxedo/agent-extensionspnpm add @claxedo/agent-extensionsbun add @claxedo/agent-extensionsQuickstart
Section titled “Quickstart”import { createAgentExtensions } from "@claxedo/agent-extensions"
const extensions = createAgentExtensions({ projectDir: process.cwd(),})
await extensions.install({ source: "acme/review-tools", targets: ["codex", "claude", "opencode"],})
console.log(await extensions.list())The source uses supported owner/repo GitHub shorthand. For local project packages, use extensions.installCached({ packagePath, id, targets }) instead.
agent-extensions install acme/review-tools --targets codex,claudeagent-extensions list --jsonagent-extensions materialize --targets codex,claude,opencode,cursoragent-extensions doctor--cache-dir controls durable package data; --runtime-dir controls generated runtime state (default <project>/.agent-extensions).
Key capabilities
Section titled “Key capabilities”- Targets —
["opencode", "claude", "codex", "cursor"]. - Package shape — an extension bundles
SKILL.md,mcp.json, and/or.cursor-plugin/plugin.json; conventionalskills/,mcp/,plugins/directories are also supported. - Snapshot & replay —
extensions.snapshot({ policyOverrides })thenapplyRuntimeAgentExtensions(snapshot, dir)from/replay. - Safety model — ownership records in
.agent-extensions/materialized.json; refuses to overwrite unmanaged paths; GitHub packages locked by resolved SHA and verified against recorded digests.
Full API
Section titled “Full API”See the README on GitHub for the complete CLI reference, package-shape conventions, host-integration boundaries, and the safety model.