← Back to Discover

harness-ai-kit

Python command-line tool that installs, locks and syncs the skills, CLIs, MCPs and plugins used by AI coding agents (Codex, Claude Code, Cursor, Kiro, DeepSeek Harness) from a single project manifest.

Command-line toolsCoding agents
Visit website ↗

What it does

harness-ai-kit is a package manager for the extras that AI coding agents rely on: skills, CLIs, MCP servers, plugins, hooks, subagents and loops. A project lists what it wants in a manifest file. The tool resolves those entries, writes a lockfile that records sources and SHA-256 checksums, and then places the chosen assets where the configured runtime expects them. The runtimes it targets are Codex, Claude Code, Cursor, Kiro and DeepSeek Harness (dsh). Installs go through a staging directory, and the runtime target is replaced only once the selected assets are ready. Skills can be added straight from a public Git repository, and an installed skill remains an ordinary directory of Markdown and metadata. It is installed with pip and needs Python 3.10+ and Git. The repository describes a team workflow in which the manifest and lockfile are committed, so a teammate who runs sync ends up with the same set of resolved assets. Its catalog is described as 42 skills, 5 CLIs and 1 plugin.

Capabilities

  • One manifest and lockfile for skills, CLIs, MCPs, plugins, hooks, subagents and loops
  • SHA-256 checksums recorded in the lockfile
  • Runtime adapters for Codex, Claude Code, Cursor, Kiro and dsh
  • Install skills directly from public Git repositories
  • Staged, rollback-aware installation

When to use it

  • Keeping a team on the same set of agent skills and tools across projects
  • Pinning and verifying skills pulled from public Git repositories
  • Using one declared setup with several AI coding tools