npm package discovery and stats viewer.

Discover Tips

  • General search

    [free text search, go nuts!]

  • Package details

    pkg:[package-name]

  • User packages

    @[username]

Sponsor

Optimize Toolset

I’ve always been into building performant and accessible sites, but lately I’ve been taking it extremely seriously. So much so that I’ve been building a tool to help me optimize and monitor the sites that I build to make sure that I’m making an attempt to offer the best experience to those who visit them. If you’re into performant, accessible and SEO friendly sites, you might like it too! You can check it out at Optimize Toolset.

About

Hi, 👋, I’m Ryan Hefner  and I built this site for me, and you! The goal of this site was to provide an easy way for me to check the stats on my npm packages, both for prioritizing issues and updates, and to give me a little kick in the pants to keep up on stuff.

As I was building it, I realized that I was actually using the tool to build the tool, and figured I might as well put this out there and hopefully others will find it to be a fast and useful way to search and browse npm packages as I have.

If you’re interested in other things I’m working on, follow me on Twitter or check out the open source projects I’ve been publishing on GitHub.

I am also working on a Twitter bot for this site to tweet the most popular, newest, random packages from npm. Please follow that account now and it will start sending out packages soon–ish.

Open Software & Tools

This site wouldn’t be possible without the immense generosity and tireless efforts from the people who make contributions to the world and share their work via open source initiatives. Thank you 🙏

© 2026 – Pkg Stats / Ryan Hefner

@agent-tool-platform/agent-kit

v0.3.0

Published

Host-neutral agent composition, capability resolution, lock generation, readiness planning, and host adapters.

Readme

@agent-tool-platform/agent-kit

The host-neutral composition engine beneath Agent Builder.

Agent Kit converts a canonical agent definition into a validated, reproducible build. It resolves capability releases through a narrow registry reader, selects a compatible six-dimensional capability profile, generates agent.lock, composes bounded instructions, produces a readiness plan, and delegates host files to an adapter.

The package and its sibling first-party registry are prepared for public npm distribution but have not completed their one-time bootstrap. Their checked-in manifests deliberately stay private; release stamping removes those guards only in an ephemeral candidate. After bootstrap, an agent composition repository installs the exact lockstep Platform release:

npm install @agent-tool-platform/[email protected]

Boundary

Agent Kit owns:

  • canonical agent definition validation and normalization;
  • exact capability release and execution-profile resolution;
  • compatibility validation;
  • deterministic lock serialization and SHA-256 identities;
  • bounded agent/capability instruction composition;
  • environment-neutral readiness requirements;
  • host adapter generation;
  • the stable identity seam later Prepare and Agent Instance work can extend.

It does not execute capability tools, deploy providers, store registry data, collect telemetry, or manage running instances.

Canonical definition

{
  "schemaVersion": 1,
  "id": "developer-optimization",
  "name": "Developer Optimization",
  "version": "1.0.0",
  "instructions": "Prefer compact, evidence-backed capability output.",
  "capabilities": [{ "id": "ast-summarizer" }, { "id": "git-optimizer", "version": "0.1.0" }]
}

Capability selections name capabilities, not MCP tools. An omitted capability version resolves to the static registry's pinned current version and is then recorded exactly in the lock. An exact requested version must match that entry; Agent Kit does not forward version requests to or invent a multi-version registry API. An optional profile selects a declared capability profile; otherwise Agent Kit deterministically prefers read-only profiles before mutating profiles, then local, hybrid, and remote bindings.

Registry seam

Agent Kit consumes the Capability Registry package's public reader and canonical entries:

import {
  createCapabilityRegistryReader,
  loadFirstPartyCapabilityRegistry,
} from '@agent-tool-platform/capability-registry';

const registry = createCapabilityRegistryReader(await loadFirstPartyCapabilityRegistry());

Reader results are validated against the Registry's public CapabilityEntry schema and consistency rules, plus Agent Kit's build-safety constraints for public artifact identities and portable references. The six-dimensional profile summaries and explicit bindings come directly from the canonical Registry entry. Capability instructions and per-tool routing remain capability-owned and are delivered by the MCP runtime; Agent Kit composes bounded capability/profile boundaries and the Registry routing summary rather than copying tool schemas. Artifacts, pinned source revisions, and a digest of each complete consumed Registry entry provide reproducible identity.

VS Code build

import { buildVsCodeAgent } from '@agent-tool-platform/agent-kit';

const build = await buildVsCodeAgent(definition, {
  registry,
  readinessSnapshot: {
    schemaVersion: 1,
    availableLocalBindings: [],
  },
});

build.lockText; // deterministic agent.lock JSON
build.adapter.files; // .github/agents/<id>.agent.md and .vscode/mcp.json
build.readiness; // machine-readable preparation requirements
build.instanceIdentity; // stable seam for a later environment-specific instance

The VS Code adapter uses exact package versions for local stdio servers and input placeholders for remote endpoints or named configuration. Declared-but-unpublished packages use offline launch mode and remain a Prepare requirement.

For remote HTTP bindings, Agent Kit carries the Registry's validated header name, configuration name, and safe literal prefix into the resolved execution binding. The VS Code adapter emits one ordinary endpoint input and separate password inputs for required authentication configuration:

{
  "inputs": [
    {
      "id": "example-access-token",
      "type": "promptString",
      "description": "Example: access-token",
      "password": true
    },
    {
      "id": "example-endpoint",
      "type": "promptString",
      "description": "Example MCP endpoint"
    }
  ],
  "servers": {
    "example": {
      "type": "http",
      "url": "${input:example-endpoint}",
      "headers": {
        "Authorization": "Bearer ${input:example-access-token}"
      }
    }
  }
}

This shape follows the authoritative VS Code MCP configuration reference: remote servers use type, url, and string-valued headers, and promptString inputs support substring substitution. Agent Kit uses only the Registry's literal prefix; it does not implement templates, transforms, encoding, or inferred authentication schemes.

Authenticated remote HTTP is host-compatible only when the Registry mapping is complete and the adapter can express it faithfully. Missing or unsupported transport semantics remain incompatible. Unauthenticated remote HTTP remains compatible, and local/hybrid stdio configuration continues to use environment-variable input references.

Lock schema 2 persists the selected client mapping under the binding using only the header name, configuration name, and literal prefix. Registry entry digests also cover the complete source mapping. VS Code adapter schema 2 identifies the new compatibility and generated-header semantics. Configuration values, endpoint values, credentials, absolute local paths, and live provider identifiers never enter agent.lock or generated files.

Readiness

Readiness is a plan, not a liveness probe. It distinguishes:

  • locally available bindings;
  • local setup still required;
  • remote or provider setup still required;
  • missing named configuration;
  • incompatible bindings.

Compatibility answers whether a host can represent the selected binding. Readiness separately answers whether its named configuration, remote connection, local artifact, and provider prerequisites have been prepared. Generating an authenticated HTTP server therefore makes the binding compatible but does not mark a missing token, endpoint, or provider prerequisite ready.

An installed local binding can be ready while its process is inactive. Runtime and provider health remain later Prepare/management concerns.