Skip to content
bordoniPublic

About

Run Claude Code under separate accounts on one machine (work + personal) with isolated logins and shared config

Topics

Resources

Stars

11 stars

Watchers

0 watching

Forks

Repository files navigation

claudep

CI

Run Claude Code under separate accounts on one machine, for example a company Enterprise org and a personal Max plan, with fully isolated logins and shared configuration.

Claude Code has no account switching of its own. claudep is one bun script with no dependencies. It manages named profiles and points Claude Code at them through CLAUDE_CONFIG_DIR.

claudep init enterprise --sso --email you@company.com --alias eclaude
eclaude          # Claude Code as the enterprise account
claude           # Claude Code as whatever ~/.claude is logged in as

Not affiliated with Anthropic

claudep is an independent, unofficial tool written by me, Gustavo Bordoni. I have no affiliation with Anthropic. Anthropic has not endorsed, sponsored, reviewed or supported this project in any way. "Claude" and "Claude Code" are Anthropic's names and trademarks and appear here only to describe what the tool works with. Use it at your own risk and under the terms of your own Claude account.

Install

Requires bun and an existing Claude Code install (claude on your PATH).

bun add -g @bordoni/claudep        # or: npm install -g @bordoni/claudep  (bun is still what runs it)
bunx @bordoni/claudep --help       # try it without installing

The package name is @bordoni/claudep and the command it installs is claudep. The same package is mirrored to GitHub Packages, which needs a token with read:packages even for public installs; .ref/releasing.md has the .npmrc lines.

Or clone and symlink it anywhere on your PATH:

git clone git@github.com:bordoni/claudep.git ~/workspace/claudep
ln -s ~/workspace/claudep/claudep.ts ~/.local/bin/claudep

Windows

claudep runs natively on Windows from PowerShell 5.1, PowerShell 7 and Git Bash. Install it with bun, which writes claudep.exe into %USERPROFILE%\.bun\bin:

bun add -g @bordoni/claudep

Two things differ from macOS:

  • Profiles share configuration through symlinks, and Windows lets a normal user create symlinks only with Developer Mode on (Settings > For developers > Developer Mode). claudep init stops and says so when it is off. Turn it on, open a new terminal and run the same command again.
  • There is no keychain. Claude Code keeps each profile's login in <profile>\.credentials.json, and claudep doctor checks that the file is there.

claudep env prints PowerShell syntax when run from PowerShell and sh syntax from Git Bash, so claudep env work | Invoke-Expression and eval "$(claudep env work)" both work. Git Bash otherwise behaves like bash elsewhere: put eval "$(claudep shell-init bash)" in ~/.bashrc. An npm-installed claude.cmd runs through cmd.exe; the native installer's claude.exe avoids that hop and is what claudep doctor recommends. cmd.exe itself is not supported.

Usage

claudep <name> [claude args…]        run claude with profile <name>
claudep init <name> [options]        create/update a profile; logs in when it has no login yet
claudep list [--json] [--live]       every profile and who it is logged in as
claudep status <name> [--json] [--live]   login state for one profile ("default" = ~/.claude)
claudep env <name> [--shell <sh>]    print the CLAUDE_CONFIG_DIR pin for eval, source or Invoke-Expression
claudep vars <name> [KEY=VALUE…]     list or set the variables a profile exports (see below)
claudep alias <name> <command>       write a shim so "<command>" == "claudep <name>"; rewrites one claudep wrote
claudep alias [--json]               list every shim on PATH and its state
claudep settings <name> [key=value…] settings one profile gets on top of the shared settings.json (see below)
claudep handoff <name> [session-id]  carry this directory's latest session on under another profile
claudep usage [name…] [--json]       five-hour and weekly usage for every profile
claudep statusline                   statusline segment with the profile and its usage (reads stdin)
claudep current [--json|--name]      which profile this shell is on, and why
claudep doctor [name] [--json]       verify symlinks, keychain entry, unclassified files, shims, Claude Code version
claudep doctor --accept <name> <item>   stop counting one shared item as a problem; --unaccept undoes it
claudep ps [name…] [--json] [--stale]   live Claude Code sessions in every profile; --stale only older versions
claudep rm <name> [--keep-login]     log out and delete a profile (base is never touched); refuses
                                     one with live sessions unless --force
claudep local <name> | --remove      pin the current directory tree to a profile (see below)
claudep resolve [dir]                print the profile pinned for a directory
claudep shell-init [zsh|bash|fish|powershell]   print the hook that applies pins on cd
claudep completion [zsh|bash|fish|powershell]   print tab completions for commands, flags and profile names
claudep --version

init options: --sso, --email <addr>, --console, --copy-mcp, --alias <command>, --env KEY=VALUE (repeatable), --no-login, --force. Run claudep help for the full text.

claudep init on a profile that is already signed in does not start a login. Pass --sso, --console or --email to log in again, or run claudep <name> auth login. Login flags are saved in <profile>/claudep.login, even with --no-login, so the next init logs in the same way and claudep status, doctor and init --no-login print the full command.

Keeping profiles in sync

claudep <name>, claudep run and alias commands link any shared file or directory the profile is missing before they start claude, and say so on stderr: claudep: linked CLAUDE.md into work. That covers CLAUDE.md, settings.json, keybindings.json, statusline-command.sh and the shared directories when they appear in ~/.claude after init. Other top-level *.md files wait for claudep init <name>, because a new one could be per-instance state. A launch never replaces a real file or directory, and a failed link never stops claude from starting. It adds no links when your shell sets CLAUDE_CONFIG_DIR to a directory outside the profiles root, or when every shared link in the profile points at another base.

claudep doctor explains a real item that hides the shared one. For a directory it counts what links into the base and names what only the profile has. For settings.json it lists the top-level keys that differ, by name only, and which ones claudep settings can carry. Both print the commands that share it:

mv ~/.claudep/work/skills ~/.claudep/work/skills.old && claudep init work --no-login

When the difference is what you want, say so and doctor stops counting it. The choice lives in <profile>/claudep.doctor:

claudep doctor --accept work settings.json     # shown as info, exit 0
claudep doctor --unaccept work settings.json   # a problem again

A launch and claudep init leave an accepted item as it is; claudep init --force links it anyway.

claudep doctor --json prints {schemaVersion, claude, base, profiles, aliases, problems}, each check as {item, state, level, message, fix?, problem}, with the same exit code as the text output: 1 when problems is above zero.

claudep alias lists every shim on PATH with its profile, target and state: ok, missing-target, other-target, missing-profile or shadowed. claudep alias <name> <command> rewrites a shim claudep wrote, in the directory it already sits in, and leaves any other file alone, including a symlink. The names claude and claudep are refused. claudep doctor fails on a shim that cannot start and warns on one that runs another claudep, with the command that fixes it. After switching how claudep is installed, run the claudep alias line it prints for each one.

Pin a profile to a directory

A file named .claudep at the top of a repo names the profile every hooked shell should use inside that tree, the way .nvmrc names a Node version. Commit it and the whole team inherits the pin.

cd ~/work/acme
claudep local enterprise          # writes ./.claudep containing "enterprise"

Shells apply pins through a hook. Add one line to ~/.zshrc (or ~/.bashrc with bash, including Git Bash on Windows):

eval "$(claudep shell-init zsh)"

In fish the line goes in ~/.config/fish/config.fish, and in PowerShell in $PROFILE:

claudep shell-init fish | source
claudep shell-init powershell | Out-String | Invoke-Expression

From then on, cd into a pinned tree sets CLAUDE_CONFIG_DIR for that profile, plus the profile's own variables, and cd out of it returns the shell to ~/.claude. The hook is pure shell with no subprocess (builtins in fish, cmdlets only in PowerShell), so it costs nothing at the prompt. The rules:

  • The nearest .claudep file upward from the current directory wins. An empty one cancels a parent pin.
  • The hook only changes a CLAUDE_CONFIG_DIR it set itself. It tracks that in CLAUDEP_AUTO, so a manual pin from eval "$(claudep env work)" (claudep env work | source in fish, claudep env work | Invoke-Expression in PowerShell) or a plain export stays put until you eval "$(claudep env --unset)".
  • claudep <name>, claudep run, claudep handoff and alias commands put CLAUDEP_RUN=<profile dir> (none for default, which has no profile dir) in claude's environment and drop an inherited CLAUDEP_AUTO, so inside that session claudep current says set by: claudep run and the hook leaves the session's profile alone.
  • A pin that names a profile you have not created prints one warning per directory change and sets nothing.

claudep env prints the syntax of the shell it runs in: it looks at its parent process (fish, pwsh, or an sh-like shell) and falls back to your login shell. Pass --shell sh|fish|powershell when neither answer fits, for example from a script or a Makefile.

claudep current tells you which profile the shell is on and how it got there (hook, manual pin, or nothing). claudep list adds the same line at the bottom.

Per-profile variables

A profile can carry environment variables that go with its account, such as a Bedrock or Vertex setup, or the Console sign-in Claude Code should use:

claudep vars bedrock CLAUDE_CODE_USE_BEDROCK=1 AWS_PROFILE=bedrock AWS_REGION=us-east-1
claudep vars work ANTHROPIC_PROFILE=work      # pick a Console sign-in from ~/.config/anthropic
claudep vars work                             # list them
claudep vars work --unset ANTHROPIC_PROFILE   # remove one
claudep init vertex --env CLAUDE_CODE_USE_VERTEX=1 --env CLOUD_ML_REGION=us-east5

They live in ~/.claudep/<name>/claudep.env, one KEY=VALUE per line. claudep <name> and alias commands pass them to Claude Code, claudep env <name> prints them with the pin, and the shell hook exports them while you are inside a pinned tree and clears them when you leave. The file is read as data and never run, by claudep and by every hook.

  • A variable you set yourself always wins. The hook and claudep env leave it alone, and claudep <name> passes your value.
  • CLAUDEP_ENV_KEYS lists what a claudep.env put in the shell, so moving to another pinned profile or running claudep env --unset clears exactly those.
  • CLAUDE_CONFIG_DIR, CLAUDE_SECURESTORAGE_CONFIG_DIR, CLAUDEP_* and the credential variables in CREDENTIAL_ENV are refused, including ANTHROPIC_API_KEY, ANTHROPIC_AUTH_TOKEN, ANTHROPIC_CUSTOM_HEADERS, CLAUDE_CODE_OAUTH_TOKEN, CLAUDE_CODE_OAUTH_REFRESH_TOKEN, MCP_CLIENT_SECRET, AWS_SECRET_ACCESS_KEY, AWS_SESSION_TOKEN and AWS_BEARER_TOKEN_BEDROCK. claudep does not store Claude Code's credentials. General-purpose tokens such as GITHUB_TOKEN are not on the list. Variables that hold a path or a profile name, such as AWS_PROFILE or ANTHROPIC_IDENTITY_TOKEN_FILE, are fine. A refused line in an existing file is skipped, and claudep <name>, claudep env, claudep vars and claudep doctor name its line number.
  • Values are one line, not empty, with no spaces at either end. claudep vars edits the file in place and keeps your comments. claudep doctor lists each profile's variable names and reports lines nothing reads.
  • A shell that is already inside a pinned tree picks up a change on the next cd.

Per-profile settings

settings.json is shared. For the few settings one account needs differently, give the profile its own:

claudep settings work model=opus                 # values are JSON when they parse, otherwise a string
claudep settings work syncClaudeAiSkills=false   # stop syncing this account's claude.ai skills
claudep settings work                            # list them
claudep settings work --unset model              # remove one

They live in ~/.claudep/<name>/claudep.settings.json. claudep <name>, claudep run and alias commands pass them to Claude Code with --settings, which ranks above your user and project settings and below managed ones. Plain claude in a shell the hook or claudep env set up does not see them, and neither does a run where you pass your own --settings. env is refused here; use claudep vars. claudep doctor lists the keys and fails on a file that is not a JSON object.

Tab completion

One line in your rc file completes subcommands, their flags, the values --shell, shell-init and completion take, and your profile names:

eval "$(claudep completion zsh)"      # ~/.zshrc, after compinit
eval "$(claudep completion bash)"     # ~/.bashrc
claudep completion fish | source      # ~/.config/fish/config.fish
claudep completion powershell | Out-String | Invoke-Expression   # $PROFILE, pwsh or Windows PowerShell 5.1

Windows PowerShell 5.1 does not hand a bare - or -- to a completer, so type the first letter of a flag there: claudep init --e<Tab>. pwsh completes --<Tab> too.

Profile names come from a directory listing of ~/.claudep at the moment you press Tab (Get-ChildItem in PowerShell), so the script never runs claudep and a new profile shows up without reloading anything. In zsh the same output also works as a file: claudep completion zsh > ~/.zfunc/_claudep with ~/.zfunc on fpath before compinit, if you prefer autoloading over eval.

Show the profile in your prompt or statusline

In a shell prompt use the variable itself; no command runs:

# zsh or bash: the last path segment is the profile name when a profile is active
PROMPT='${CLAUDE_CONFIG_DIR:+[${CLAUDE_CONFIG_DIR##*/}] }'"$PROMPT"
# fish
function fish_prompt; set -q CLAUDE_CONFIG_DIR; and echo -n "[$(string replace -r '.*/' '' -- $CLAUDE_CONFIG_DIR)] "; ...; end

In scripts, and in the shared statusline-command.sh, claudep current --name prints exactly one word: the profile name, default for ~/.claude, or custom for a CLAUDE_CONFIG_DIR outside the profiles root. The statusline script inherits Claude Code's environment, so it sees the profile the session was started with.

To show usage too, pipe the statusline input to claudep statusline. It prints work · 5h 23% · 7d 41% from the rate_limits Claude Code hands every statusline, and remembers the numbers for claudep usage:

# in statusline-command.sh
input=$(cat)
segment=$(printf '%s' "$input" | claudep statusline)

It starts bun on every render, about 30 ms here, and writes ~/.claudep/.usage/<name>.json only when the numbers change. API-key, Bedrock and Vertex sessions get no rate_limits, so the segment is just the name.

Usage limits per profile

claudep usage                # every profile: five-hour and weekly use, when each resets, how fresh
claudep usage work --json    # one profile, with the per-model weekly windows

From Claude Code 2.1.301 on, claudep usage asks each profile with claude -p /usage, which reports usage without running the model and, with --no-session-persistence, writes no transcript. Older versions can sign a profile out on a short command like that (anthropics/claude-code#95822), so there it shows the numbers last seen by claudep statusline or an earlier run instead. --live asks anyway, --cached never asks. claudep never reads a token: Claude Code makes every request.

Live sessions

claudep ps              # every live session in ~/.claude and each profile
claudep ps work --json  # one profile, as JSON
claudep ps --stale      # only sessions on a Claude Code older than the one installed

Each running Claude Code session writes sessions/<pid>.json in its config dir. claudep ps reads those files and nothing else in the folder, so the .key files next to them, which hold a peer token, are never opened. It shows the profile, pid, name, status and how long it has held it, kind, version and working directory, and skips a record whose process is gone. On Linux it also skips one whose pid now belongs to another process or another pid namespace. claudep rm refuses a profile with live sessions, because its logout revokes the token under them; --force removes it anyway.

MCP servers per profile

claudep mcp                      # server × profile table from the files, no subprocess
claudep mcp work --live          # ask claude mcp list in that profile
claudep mcp --server nx1 --json  # one server everywhere; exit 77 if it needs auth

By default (--cached) claudep mcp reads only names: the servers in each .claude.json mcpServers (shown as unknown) and the entries of mcp-needs-auth-cache.json (needs-auth). It never reads a value, so URLs, env and headers stay where they are. claude.ai connectors show up only once Claude Code has cached them as needing auth, and an entry can outlive the problem until Claude Code connects again. --live runs claude mcp list in each profile, from an empty scratch folder so no project .mcp.json is read and a stdio server that writes relative files leaves nothing behind, and keeps the name and one word per server: connected, needs-auth, failed, off or unknown. A line it cannot read is counted, never printed, because those lines carry stdio arguments and URLs. --live connects to every server, which starts stdio servers and can refresh MCP logins, so it is never the default; below Claude Code 2.1.301 it falls back to the names. It exits 77 when a server named with --server needs auth, or with none named, any server. claudep doctor lists the servers waiting for auth.

Hand a session to another profile

When one account hits its limit, carry on in another:

claudep handoff personal                # newest session in this directory, resumed under "personal"
claudep handoff personal <session-id>   # a specific one
claudep handoff work -- -p "go on"      # anything after -- goes to claude

projects/ is shared, so every profile can read every transcript. handoff finds the newest one for the current directory and runs claude --resume <id> --fork-session under the other profile. The fork gets a new session id, so the original session can keep running without two processes writing one file. The first reply is slower and costs more than usual: the prompt cache belongs to the account that built it.

Exit codes

claudep <name>, claudep run, claudep handoff and alias commands exit with claude's own code. For the cases below claudep uses a code from sysexits.h instead, so a script can tell them apart. Any other stop before claude starts, such as handoff finding no session, exits 1.

Code Meaning
64 Usage error or an invalid profile name
67 The profile does not exist
69 claude is not on PATH, is still missing after an update, or only a .ps1 was found
75 In print mode (-p), claude stopped on a usage limit or billing error: rate_limit, billing_error
77 In print mode, claude stopped because the login no longer works: authentication_failed, oauth_org_not_allowed, account_on_hold, verification_required, cloud_credential_error
78 The profile's claudep.settings.json cannot be read, or CLAUDE_CONFIG_DIR is set but empty

Other commands exit 0 on success and 1 on failure, except that an invalid profile name is 64 everywhere.

Login markers

Every claudep <name>, run, handoff and alias command adds StopFailure hooks to the --settings it passes, one per error above. When a turn ends on one, the hook writes the error's name and the run's id to ~/.claudep/.health/<name>. Each hook is a single shell printf into that file: Claude Code does not wait for StopFailure hooks, and on 2.1.323 one that took 50 ms was killed before it wrote anything. It reads no token and prints nothing. From then on:

  • a print-mode run that recorded a login error exits 77, and a limit 75, instead of claude's 1, with the error and the login command on stderr;
  • after a login error, claudep list and status show needs login with when it was seen, and doctor fails the profile with the login command; after a limit they print hit a limit and doctor notes it. --json adds failure: {error, at} for both;
  • the marker clears when a later print-mode run or claudep <name> auth login exits 0, and when the profile is removed. Other clean exits, auth status included, leave it, and so does an interactive session that ends cleanly after recording it.

claude auth status reports a login whose refresh token has stopped working as logged in, so the marker is the only place that failure shows. Passing your own --settings leaves the hooks out. Plain claude records nothing. The hooks need a POSIX shell, which Claude Code uses on Windows through Git Bash; that path has not been checked on Windows yet.

Claude Code updates

Install Claude Code with the native installer when you use profiles. Each profile is its own config dir, so each runs its own auto-updater. With the native installer an update unpacks a new version and swaps one symlink, so claude is always there. With an npm or bun global install every profile reinstalls the same package, and claude is gone while the reinstall runs (anthropics/claude-code#93197). Move with claude install.

  • claudep <name>, claudep run and alias commands wait when claude is missing: up to five minutes while any config dir holds a fresh .update.lock, five seconds otherwise.
  • claudep doctor names the install method, every claude on PATH, whether each config dir runs the updater and why, including env in managed settings (managed-settings.json and managed-settings.d/), and each one's last update. On npm or bun it warns when more than one config dir updates. Leave one on and turn the rest off with claudep vars <name> DISABLE_AUTOUPDATER=1.
  • DISABLE_AUTOUPDATER=1 in the shared settings.json env turns the updater off in every profile at once. claude update still works by hand.

How it works

~/.claude is left exactly as it is and stays the default profile. Each named profile is a thin directory under ~/.claudep/<name>. Inside it, shared configuration is a symlink back into ~/.claude:

Shared (symlinked) Per profile (real files)
CLAUDE.md and other top-level *.md .claude.json (login identity, user-scope MCP servers, folder trust)
settings.json, keybindings.json, statusline-command.sh org-pushed remote-settings.json, policy-limits.json
hooks/, skills/, commands/, agents/, rules/, output-styles/, themes/, workflows/ history.jsonl, todos/, sessions/, teams/, caches, telemetry
plugins/, plans/, projects/ (session transcripts and auto-memory) credentials

On macOS, credentials never touch the profile directory: Claude Code stores them in the Keychain under Claude Code-credentials-<sha256(CLAUDE_CONFIG_DIR)[0:8]>, so every profile has its own login and refresh token and they cannot overwrite each other. I verified this against Claude Code 2.1.259. An older bug where every config dir shared one Keychain entry no longer applies. On Linux and Windows the login is <profile>/.credentials.json, a real file inside the profile that is never shared.

The shared list is an allowlist, so an account-specific file cannot leak across profiles by accident. claudep doctor reports any base file that is neither shared nor known-private. That is how you notice when a new Claude Code version adds something.

Notes

  • The first time you open a repository under a new profile you will re-accept folder trust, and claude.ai connectors need their OAuth redone in that profile.
  • Two profiles running at the same time write the same settings.json and plugins/. That is the same situation as two terminals today.
  • In Claude Code 2.1.311, background sessions (claude --bg) under a profile get their own on-demand supervisor, which keeps its files in the profile. claude daemon install still refuses a profile, and a server flag currently turns it off for every config dir, ~/.claude included.
  • Since Claude Code 2.1.280, every auto-memory write in a profile asks for approval, and auto mode cannot give it. The memory directory sits under the shared projects/ symlink, and Claude Code now checks a write at the path it lands on, ~/.claude/projects/..., which it treats as sensitive. This is anthropics/claude-code#98044, open upstream. claudep <name>, claudep run and alias commands avoid it: they pass the real memory path with --settings, unless you pass your own --settings. Plain claude in a shell the hook or claudep env set up still asks. claudep doctor warns while your Claude Code is affected. The default profile is not.
  • Skills and plugins synced from claude.ai land in the shared skills/synced/ and plugins/synced/, in one folder per account, so profiles do not mix them. The syncClaudeAiSkills and syncClaudeAiPlugins switches live in the shared settings.json; to change one for a single profile, use claudep settings <name> syncClaudeAiSkills=false.
  • Console sign-ins made without an API key live in ~/.config/anthropic, outside every config dir, so all profiles see the same ones. ANTHROPIC_PROFILE picks which one Claude Code uses; set it per profile with claudep vars <name> ANTHROPIC_PROFILE=<sign-in>.
  • In VS Code, CLAUDE_CONFIG_DIR in claudeCode.environmentVariables must be an absolute path. The session list ignores it when claudeProcessWrapper is set.
  • claudep list, status, usage, doctor and mcp --live give each claude they start 30 seconds. One that has not answered by then is stopped (SIGTERM, then SIGKILL three seconds later; on Windows the process tree is killed at once with taskkill /T /F) and its row shows timeout with what .claude.json or the last usage run recorded; --json adds "timedOut": true. Set CLAUDEP_TIMEOUT=<seconds> to change it, 0 to wait forever. claudep <name>, run and handoff are never stopped.
  • Set CLAUDE_PROFILES_DIR to move the profiles root. Keep it out of iCloud or Dropbox; .claude.json is rewritten constantly and sync tools create conflict copies.
  • Never put CLAUDE_CONFIG_DIR in a settings.json env block. Claude Code detects that mismatch and disables features; claudep doctor reports it.
  • ANTHROPIC_API_KEY, ANTHROPIC_AUTH_TOKEN and CLAUDE_CODE_OAUTH_TOKEN override the login in any config dir, always in -p mode. claudep <name> and claudep doctor warn when one is set and leave it alone. claudep doctor also names the key source Claude Code reports, and fails when Claude Code reports a config dir other than the profile's.
  • CLAUDE_SECURESTORAGE_CONFIG_DIR moves Claude Code's credential store away from the config dir, so one inherited value puts every profile on one login. claudep <name>, claudep run, alias commands and claudep doctor warn when it is set to anything but the profile dir.
  • An empty CLAUDE_CONFIG_DIR makes Claude Code use the current directory as its config dir (anthropics/claude-code#104629). claudep current, claudep run default and claudep doctor stop on it and tell you to run unset CLAUDE_CONFIG_DIR. Running a named profile still works, because claudep sets the variable itself.
  • With --add-dir ~/.claude/projects, Claude Code also loads ~/.claude/CLAUDE.md as project instructions through the parent walk. Once CLAUDE.md is linked into a profile, that session loads it twice. Narrow the --add-dir to the folder you need.
  • claudep doctor fails on macOS when Claude Code is older than 2.1.144, the first build whose Keychain item is namespaced per config dir.
  • Before Claude Code 2.1.301 a short claude auth status can spend a profile's refresh token and sign it out (anthropics/claude-code#95822). On those versions claudep list, status and doctor show the account recorded in the profile's .claude.json, marked *, and do not ask Claude Code. --live asks anyway. Update Claude Code to get live answers back.

Releases

Versions, dates and changes are in CHANGELOG.md. Releases are tagged with the bare version, 0.1.0, and published to npm with provenance by GitHub Actions.

Development

bun install        # dev tooling only; nothing ships with the tool
bun run check      # typecheck, lint, tests. Same as CI

To work on the script, start with AGENTS.md. The detail lives in .ref/.

License

MIT

About

Run Claude Code under separate accounts on one machine (work + personal) with isolated logins and shared config

Topics

Resources

Stars

11 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages