Commands

Command	What
init	First-run wizard: remote (private), machine identity, roots, hooks
sync	Walk → redact → manifest → tripwire → commit → push (--dry-run)
pull	Fetch latest into the staging repo (no write to ~/.claude)
restore	Restore here, rewriting paths (--dir, --backup, --force, --prune); nudges a Desktop restart when Code sessions come back
status	Sync state: remote, last sync, roots, hooks
search	Find a Claude Code session by title or content across live + synced history (alias grep); --raw grep lines, --all every file, --live/--repo scope, -s case-sensitive
global	install / uninstall / status the global sync hooks in ~/.claude (alias hooks)
project	install / uninstall / status this repo's guard hook + CLAUDE.md guide (committed)
local	same as project, but gitignored (.claude/settings.local.json)
guard	The PreToolUse hook that enforces worktree/PR discipline — invoked by Claude Code, not run by hand (wired in by project/local)
guide	install / uninstall / status / show the CLAUDE.md guide block standalone (--global targets ~/.claude/CLAUDE.md, --path overrides; install previews in a scrollable UI, skipped with -y or off a TTY)
mcp	list (alias ls) / get / add / remove (alias rm) / enable / disable MCP servers (--scope user｜project｜local, --transport stdio｜http｜sse, --env, --header); bare mcp on a TTY opens an interactive screen (mirrors claude mcp)
account	Manage multiple Claude Code logins: add / list (alias ls/status) / run <id｜email> [-- claude args] / switch / sessions (alias ps) / remove (alias rm) / purge. run --no-share isolates a session; switch takes --dry-run / --force / --kill
config	get / set / show / path / edit (~/.clauderig/config.json)
doctor	Health-check environment + sync + worktree discipline (--fix repairs)
ui	Interactive dashboard

The worktree and prune verbs (rig worktree, rig prune) live in rig — claudeRig wires the guard that makes them the default path. See Worktree discipline below.

The sync → restore loop

sync snapshots your ~/.claude setup, redacts secret-bearing fields, rewrites machine-specific paths into a portable form, and commits/pushes to your private repo. restore does the inverse on another machine: it pulls, rewrites the portable paths into this OS's slugs, and merges — keeping any local secrets in place so a new machine simply re-authenticates.

When a restore brings back Claude Code sessions, it reminds you to fully quit and reopen Claude Desktop — Desktop only rebuilds its Code-tab list from the restored session sidecars on startup, so a running app won't show them until it restarts.

Finding a session

Lost track of which chat had that work? search finds it by title or content:

clauderig search "auth refactor"

It scans your Claude Code transcripts (projects/**.jsonl) across this machine's live ~/.claude and the synced staging repo — which may hold sessions from other machines or older ones no longer live here — then groups the hits into named sessions:

● Refactor the auth middleware
  a1b2c3d4 · 2026-07-02 · opus-4-8 · ~/Git/acme-api · desktop+repo
  17 match(es)   resume: cd ~/Git/acme-api && claude --resume a1b2c3d4-…
    …split the token check out of the request handler…

Each result carries the session's Desktop title, project, last-used date, and a ready claude --resume command. Titles are matched too, so a topic word finds the chat even when it isn't in the body, and results rank by relevance (title hit, then match count, then recency). Matches inside injected skill-listing and system records are ignored — so a word that appears in the skill catalog (like a skill name) doesn't light up every session.

• --raw — grep-style line output instead of grouped sessions
• --all — search every file (config, skills, file-history, the Desktop dir), not just transcripts (implies --raw)
• --live / --repo — restrict to this machine or the synced repo
• -s / --case-sensitive

On a long search a live scanning… N transcripts, K matches status shows on the terminal (and stays out of piped output).

Desktop "Chat" tab

Desktop Chat-tab conversations live server-side, not on disk — they never appear here. search covers your Claude Code transcripts; a miss doesn't prove a Chat-tab conversation is gone, so check claude.ai for those.

Hooks

clauderig hooks install

Wires two Claude Code hooks: SessionStart → pull (so each session starts from the latest synced state) and Stop → sync (so your work is captured when a session ends). Both are portable across OSes and idempotent.

Worktree discipline

clauderig guard (a PreToolUse hook) and rig worktree make worktrees and PRs the default path for Claude Code, and stop a session from scrambling your VS Code chat history by moving its working directory. Chat history is keyed to the folder path, so the model edits from one pinned window while worktrees open in their own window for review only.

rig worktree new <branch>   # sibling checkout off mainline (prints the path)
rig worktree new <branch> --open    # …and open a review window for this run
rig worktree new fix/x --base release-1
rig worktree list           # this repo's worktrees (alias: ls)
rig worktree open <branch>  # (re)open a worktree's review window (branch or path)
rig worktree rm <branch>    # remove the worktree, keep the branch (-f if dirty)

Worktrees live at <parent>/<repo>-worktrees/<branch> — a sibling of the repo, so they never clutter the primary checkout and each gets its own review-window history. new never moves the session's cwd; it prints the path and, when opted in, opens a separate window.

Configuring the review window

Because worktree is a rig command, the review-window behavior is configured in .rig.json via rig config set, not in claudeRig. By default new does not open a window — opt in per run with --open, or always with the worktree.autoOpen key:

rig config set worktree.autoOpen true       # always auto-open (like --open)
rig config set worktree.openCmd "cursor -n"  # open Cursor instead of VS Code

• autoOpen (default false) — whether new opens a window at all. --open/--no-open override it per run; worktree open always opens.
• openCmd (default code -n) — the program plus any flags; the worktree path is appended as the final argument and run directly (no shell). Examples: code -n, cursor -n, code-insiders -n, subl -n, idea.

See rig → Configuration for the full details. When the opener isn't on PATH, new/open print the command to run instead.

TIP

See the worktree-discipline doc for the guard rules and the full model.

TIP

See the design doc for the full picture.