連理 — intertwined branches. A unified manager for git worktrees and jujutsu (jj) workspaces, driven by teravars-powered TOML configuration with per-host / per-OS overrides and a shared base include mechanism.

Status: 0.1.0 — MVP shipped. All seven verbs (add / list / remove / cd / exec / prune / config) work on git and jj backends. Shell wrappers generated by renri shell-init. See ROADMAP.md for what's next.

Why

The existing crop of worktree CLIs (gwq, wtp, Phantom, claude-squad, Worktrunk, …) all assume git, all use a different config format, and none of them treat jj workspaces as first-class peers. The teams shipping AI-parallel-dev workflows have their config and conventions re-implemented across N tools.

renri aims to be the local CLI that:

• Auto-detects git or jj, dispatches to the right backend, exposes one verb set (add / list / remove / cd / exec / prune / config).
• Configures with TOML + Tera via the teravars library, so a single renri.toml can include = ["team-base.toml"], render {% if system.host == "thinkpad" %} ... for per-machine overrides, and use the standard env(...) / is_windows() helpers.
• Falls back to interactive picker for any required argument that was omitted on the command line — renri cd opens a fuzzy filter, but renri cd feature/foo switches directly. --non-interactive disables the fallback for scripts and CI.
• Treats jj workspaces honestly — surfaces update-stale, ships the prune behaviour that jj workspace forget is missing, handles the colocated-git+jj case where git worktree add requires -b.

Quick example (planned)

# renri.toml
include = ["{{ env(name='RENRI_TEAM_BASE', default='') }}"]

[layout]
worktree_root = "{{ env(name='HOME') }}/wt"
worktree_path = "{{ vcs.owner }}/{{ vcs.repo }}/{{ vcs.branch | replace(from='/', to='-') }}"

[[hooks.post_create]]
type = "copy"
files = [".env.example -> .env"]

[[hooks.post_create]]
type = "command"
run = "pnpm install"

{% if system.os == "windows" %}
[[hooks.post_create]]
type = "command"
shell = "pwsh"
run = "Write-Host 'Welcome to {{ vcs.branch }}'"
{% endif %}

renri add feature/auth          # creates the worktree, runs hooks
renri cd feature/auth           # prints the path (or opens picker if no arg)
renri exec feature/auth -- pnpm test
renri prune                     # gc stale worktrees / forgotten jj workspaces

For renri cd to actually change the parent shell's directory, source the snippet from renri shell-init:

# bash / zsh
eval "$(renri shell-init bash)"

# fish
renri shell-init fish | source

# pwsh — add to $PROFILE:
Invoke-Expression (& renri shell-init powershell)

AI / agent integration

renri ships as an APM package. One command gives the renri skill to every detected agent client (Copilot / Claude Code / Cursor / OpenCode / Codex / Gemini):

apm install yukimemi/renri

The skill source lives at .apm/skills/renri/SKILL.md; APM compiles it into the right format for each agent on install.

An MCP server is planned for v0.2 so MCP-compatible AI clients can drive renri verbs directly without going through bash.

Cargo features

(none yet — added as needed.)

License

MIT.