Crate oneharness_core 

oneharness-core: the reusable engine behind the oneharness CLI, and a depend-able library for tools that install hooks into — or sync configs for — the supported agentic coding harnesses (Claude Code, Codex, OpenCode, Goose, Qwen Code, Crush, Copilot CLI, Cursor).

• domain is pure: it builds argv, parses output, shapes the report, holds the harness registry, renders hooks, layers config, and computes the sync merge — with no process / filesystem / env / clock I/O.
• io performs the real I/O: PATH resolution, version probing, spawning, reading config, and writing harness config files and hooks.

The cross-harness capability a consumer most often wants: build a domain::hooks::HookSpec and call io::hooks::install to write a pre-tool hook into any harness in that harness’s native shape (a shared config file, a dedicated hooks file, or a plugin). It is generic — the command and plugin identity are caller-supplied, never hardcoded.

Modules

domain

Pure core: no process, filesystem, env, clock, or terminal I/O lives here. These modules build commands, shape the report, and parse output; all real I/O stays behind the io and commands boundaries.

errors

Typed errors surfaced only at the application boundary (see dispatch).

io

I/O boundary: PATH resolution, version probing, and subprocess execution. Everything that touches the process, filesystem, env, or clock lives here or in commands, never in domain.