harness-alchemist

Maintaining separate agent plugins for Claude Code, Codex, OpenCode, Google Antigravity, and DeepSeek Harness typically means more adapter boilerplate than actual skill logic. Harness Alchemist collapses that into a single scaffold: one generated TypeScript repository with all five harness entrypoints pre-wired, so you write your skill logic once and it runs natively on every target platform.

The architectural principle is that skills own runtime behavior while adapters stay thin. Each product skill ships as a pair of behavioral .mjs and .py twins that carry all business logic and tool contracts; every harness entrypoint is a lightweight adapter that delegates to those scripts via a stdin/stdout JSON protocol. The scaffold ships with built-in quality gates out of the box including TypeScript checks, runtime delegation tests, structural validation, and an npm-payload audit. For existing monorepos, an alchemy.json manifest enables adapted-package mode; for teams whose stack is Python, Go, Rust, or another non-JS language, the skills-only runtime mode exposes workflows to all five platforms without adopting a JavaScript runtime.

Built for developers and engineering teams that need to distribute the same set of skills or workflows across multiple AI coding agent platforms. Zero runtime dependencies, a CI pipeline with tag-triggered publishing, and install-level verification via install-check that drives each local harness CLI to confirm plugin discovery keep the entire path from initial scaffold to user-facing release backed by repeatable, automated checks.