harness-ai-kit

A package manager and composition layer for AI agent assets.

harness-ai-kit installs, resolves, locks, validates, and synchronizes Skills, CLIs, MCPs, plugins, hooks, subagents, and loops across Codex, Claude Code, Cursor, Kiro, and DeepSeek Harness (dsh).

Quickstart · Asset Catalog · Usage Scenarios · Concepts · Roadmap · Changelog · 中文文档

Why

AI teams collect useful prompts, Skills, CLIs, and MCP servers quickly. The hard part is making that collection reproducible: knowing what is installed, which versions work together, where it is materialized, and how a teammate can get the same working state without copying runtime directories.

harness-ai-kit makes the project manifest the source of truth:

harness-ai-kit.yml  ->  resolve  ->  harness-ai-kit.lock  ->  runtime materialization
       intent             plan          integrity snapshot       Codex / Claude / Cursor / Kiro / dsh

The lock records resolved sources and checksums. Installation uses a staging directory and only replaces the runtime target after the selected assets are ready. The result is useful for one developer, and essential when a team needs the same AI engineering environment across projects and runtimes.

The REMIX Method

This project is a composition layer, not another vertical AI toolkit. When a focused community Skill, CLI, or MCP already solves a problem, the preferred path is to compose it, pin it, and make it reproducible instead of rebuilding it.

• Reuse proven assets from the community or your own repositories.
• Extend an asset when a small, explicit adaptation is enough.
• Mix Skills, CLIs, MCPs, and runtime assets into one workflow.
• Integrate the selected assets through a typed manifest and lockfile.
• eXecute the same declared environment across supported runtimes.

No Lock-In

An installed Skill is still a normal directory of Markdown and metadata. You can read it, copy it, or install it directly from its Git repository without using this tool. harness-ai-kit is not a gatekeeper for content; it adds the reproducible inventory, dependency resolution, checksums, and runtime sync around that content.

Quick Start

Requirements: Python 3.10+ and Git.

python -m pip install --upgrade harness-ai-kit==0.18.18
harness-ai-kit init

mkdir my-agent-project
cd my-agent-project
harness-ai-kit init-project
harness-ai-kit add skill https://github.com/OWNER/REPO/tree/main/path/to/skill
harness-ai-kit sync
harness-ai-kit doctor

init creates or updates the shared configuration at ~/.harness-ai-kit/config.yaml. init-project creates the project manifest; sync resolves it, writes the lockfile, and materializes the selected assets for the configured runtime. See the quickstart for runtime-specific installation details.

Team Workflow

Commit the declaration and lockfile, not copied runtime directories:

Maintainer                         Teammate
----------                         --------
add selected assets                clone or pull the project
review the lockfile                harness-ai-kit sync
commit manifest + lockfile         receive the same resolved asset set

This keeps local customizations out of version control while allowing a shared, auditable AI asset baseline. sync reconciles managed assets; it is not a blind wipe of unrelated local files.

What It Provides

Capability	Outcome
Typed asset model	One dependency contract for Skills, CLIs, MCPs, plugins, hooks, subagents, and loops
Resolution and lockfiles	A reproducible selection of versions, sources, features, and SHA-256 checksums
Multiple runtime adapters	Project or global installation for supported AI coding runtimes
Git-based sources	Install a reviewed Skill directly from a public Git repository
Safe materialization	Staging, verification, replacement, and rollback-aware installation flow
Configuration boundary	User-specific endpoints and credentials live in ~/.harness-ai-kit/config.yaml, not in assets
Curated public assets	Reusable engineering, diagnostic, and AI-development assets listed in the catalog

Architecture

                     harness-ai-kit CLI
 init | add | install | sync | lock | doctor | validate | upgrade
                              |
                 manifest + dependency resolver
                              |
                     harness-ai-kit.lock
                              |
         source adapters + cache + checksum verification
                              |
      runtime adapters and asset bundles for AI coding environments

The public project deliberately separates portable product behavior from private operating context. Public packages must work with a user's own configuration and public dependencies; private endpoints, credentials, and deployment topology do not belong in the published tree.

Usage Paths

• Adopt a public Skill: install from a Git repository, then sync it into a project runtime.
• Share an engineering baseline: commit the manifest and lockfile so the team resolves the same assets.
• Author an internal or public asset: use the metadata contract, validate it locally, and publish only through an explicit reviewed release path.
• Run dsh: install Skills or the bundled plugin through the dsh runtime adapter. See dsh integration.

The usage scenarios explain when to use a Skill, when a loop is appropriate, and how a spec-driven workflow can route both.

Roadmap

The current public product focuses on portable asset management, reproducible installation, and a reviewed public catalog. Future work expands authoring and automation first; registry, browser, and organization administration remain separate platform milestones rather than hidden dependencies of the core CLI.

See ROADMAP.md for scope, milestones, and non-goals.

Documentation

• Quickstart
• Core concepts
• CLI reference
• Asset catalog
• Asset authoring contract
• Troubleshooting
• OSS release process

Contributing And Security

Use CONTRIBUTING.md for contribution expectations and SECURITY.md for responsible disclosure. Issues and feature requests belong in the GitHub issue tracker; open-ended design discussion can use GitHub Discussions.

License

Apache-2.0 © 2026 SeedForge.