claudeRig

Sync your Claude Code setup across machines — config, skills, and session history — to your own private git repo, and restore it on any computer with paths corrected across OSes and secrets never leaked. Pick up where you left off on a different machine.

The fourth rig: a single statically-linked Go binary, zero runtime deps, installable by curl | sh / Homebrew / winget / Scoop on macOS, Linux, and Windows.

clauderig init                 # wizard: create/choose a PRIVATE repo, machine name, hooks
clauderig sync                 # snapshot → redact secrets → rewrite paths → commit → push
clauderig restore              # pull → rewrite slugs for this OS → merge (keeps local secrets)
clauderig restore --dir /tmp/x # restore the CLI payload into a folder (inspect, don't touch ~/.claude)
clauderig status               # remote reachability, last sync, per-root counts, hooks
clauderig reroot <id> ~/Git/p  # re-file a session under the directory it really belongs to
clauderig repo                 # what the sync repo costs: size, files, commits, history ratio
clauderig repo gc              # repack: reclaims space, keeps every commit — try this first
clauderig repo prune --before 2026-08-01  # fold everything before a date into one commit
clauderig recent                  # sessions you actually worked on, newest first
clauderig search "auth refactor"  # find a session by title/content, with a resume command
clauderig pull                 # fetch latest into the staging repo (SessionStart hook target)
clauderig account list         # show stored Claude Code logins (alias: ls / status)
clauderig account run me@x.com # launch Claude Code as another account, isolated session
clauderig account prepare work # ready an account's profile and print its CLAUDE_CONFIG_DIR (for launchers; --json)
clauderig mcp add ctx7 npx -y @upstash/context7-mcp   # manage MCP servers (list/add/remove/enable)
clauderig desktop open work    # a Claude Desktop window per account, each its own profile
clauderig desktop main         # back to the plain Claude Desktop, which a profile window hides
clauderig desktop prune --vm   # reclaim the Cowork VM image + caches; keeps login and history
rig worktree new feat/x        # sibling worktree + review window; never moves this session
clauderig doctor               # health-check env + sync + worktree discipline + ignored settings (--fix repairs what it can; ignored settings are advisory)
clauderig hooks install        # SessionStart→pull, Stop→sync, SessionEnd→sync --flush
clauderig ui                   # interactive dashboard

What it does

• Cross-OS path correction. A session captured at /Users/john/Git/x resumes at C:\Users\John\Git\x. Project directory slugs and path values inside config are re-derived for the target machine (core/pathmap), including paths embedded in Desktop’s stored permission reasons.
• Secret redaction and publication checks. Secret-bearing config fields are stripped, and complete staged-text scanning refuses recognized credentials in transcripts and other staged files, regardless of size. Set redactTranscripts to true to scrub supported signatures from the staged copy of a conversation — the transcript, the tool results beside it, and memory notes. Live files are never edited. These checks do not sanitize existing Git history or identify every possible secret.
• Transcript chunking. On for new configurations; existing configs without the key use auto and follow the repository. Large backups use reusable 4 MiB chunks while restore writes native JSONL. Existing backups migrate on the next sync. Upgrade every participating client before enabling it. See enabling chunking.
• Private repo, no exceptions. The remote must be a GitHub repo that gh confirms is private — created with gh repo create --private or an existing one verified via gh repo view.
• Allowlist, default-deny. Only curated files sync; the ~12 GB Desktop cache tree is pruned, never descended.
• Bounded repo, unbounded memory. 90-day retention on transcripts + a size-based history squash — but every session sync ever staged keeps a permanent row in the ledger, so an aged-out chat is still findable by title, project and date. Project memory is exempt — it's durable state, not a dated record, so it never ages out of the sync.
• Find your sessions. clauderig recent lists them newest first — dated by what each transcript says, not by a file mtime that a restore or a repo checkout rewrites — and clauderig search locates one by title or content across live and synced history. Both hand you a claude --resume command. See Commands.
• Worktree discipline. A guard hook plus rig worktree make branches + PRs the default for Claude Code and keep a session from scrambling its chat history by moving the working directory. See Commands.
• Multiple accounts. clauderig account captures, lists, and switches between Claude Code logins, and can run a session as another account in an isolated environment without touching your machine-wide login.
• MCP server management. clauderig mcp adds, removes, lists, and toggles MCP servers across user / project / local scopes — args-driven or via an interactive screen (mirrors claude mcp).

Install

curl -fsSL https://rigsmith.sh/clauderig | sh    # once the release exists
# or build from source (single Go module):
go build -o clauderig ./cmd/clauderig

Requires git and the GitHub CLI (gh, authenticated) for the private-repo gate.

• All commands →