c

Pick a Claude Code account and launch it, showing each account's 5-hour and weekly usage with the time until reset. One command in place of a drawer full of CLAUDE_CONFIG_DIR=... claude aliases.

$ c

   account   plan       5h  week  resets in
→1 personal  max 20x   26%    3%  2h 27m
 2 work      max 5x    71%   44%  41m
 3 team      pro        4%    1%  3h 12m

  yolo on · worktree off  ·  [enter] personal   ↑↓ move   1-3 pick   y yolo   w worktree   r refresh   q quit

Install

npm install -g @tonoid/c

Or run it without installing: npx @tonoid/c. Node 18 or newer, no dependencies. Linux; see Limitations for macOS.

npm is the only distribution channel: no Homebrew tap, no apt repo, no install script. It is one file, so a checkout on your PATH works just as well (see Development).

c, not cc, because cc is the C compiler. A real command on PATH rather than a shell alias, so it works from scripts and from any shell, not just an interactive one. If the single letter collides with something of yours, rename the installed binary and set CMD at the top of c.mjs to match: every message reads from it.

Commands

| Command | Description | |---|---| | c | Usage table, pick an account. Enter takes the top row, which is the one used last. | | c <args...> | Skip the menu: last-used account, everything passed through to claude. c --resume, c --worktree, c "fix the failing test". | | c -a <id> [args...] | A specific account, by id or display name. | | c status | Table only. Also c ls. | | c add <id> | Create ~/.claude-<id> and log in to it. | | c yolo [on\|off] | Toggle the --dangerously-skip-permissions default. On. | | c worktree [on\|off] | Toggle the --worktree default, a git worktree per session. Off. | | c version, c help | |

In the menu: ↑↓ move the arrow (wrapping at both ends), enter launches the marked account, 1-9 pick a row outright, y toggles yolo, w toggles worktree, r refetches usage, q quits. Every key redraws the block in place instead of printing another copy of it.

Both defaults are remembered, and both are skipped when you type the flag yourself, so c --worktree with the default already on still passes it once.

How it works

Every ~/.claude* directory holding a .credentials.json is an account, so there is nothing to register: c add work, or any manual CLAUDE_CONFIG_DIR login, just shows up. Directories without credentials (plugin caches such as ~/.claude-mem) are ignored.

Rows sort most-recently-used first, so c then enter is always the account you were just in.

Usage comes from GET https://api.anthropic.com/api/oauth/usage, authenticated with the OAuth token Claude Code already wrote into each config directory. Same data as /usage inside a session. It is an undocumented endpoint: if Anthropic changes it, the usage columns show an error and picking and launching still work.

Launching sets CLAUDE_CONFIG_DIR for the child process only, so two terminals can run two accounts at once.

State

~/.config/c/db.json: order (most-recently-used ids), the yolo and worktree flag defaults, and cached usage. Refetched when older than 60s, or on r. Delete the file to reset. Nothing else is written, and no credential ever leaves the machine except as the authorization header on the usage call.

Another remembered claude flag is one line: add it to FLAGS in c.mjs and it gets a c <name> [on|off] subcommand and a db field for free.

Limitations

• Subcommand names shadow prompts. c status prints the table; to send that word as a prompt use c -a main status.
• macOS keeps Claude Code credentials in the Keychain rather than in .credentials.json, so discovery finds nothing there. Linux only for now.
• The menu assumes its footer fits on one terminal line; a very narrow window can leave a stale line behind on redraw.

Development

git clone https://github.com/tonoid/c
cd c
npm test

The suite is node:test and node:assert only. It builds throwaway HOME directories with fake config dirs, so it never reads or writes real accounts and never calls the network.

To run the checkout as the real command, so git pull updates it:

ln -s "$PWD/c.mjs" ~/.local/bin/c

Releasing

Versions come from the commit messages, so land work on main with conventional commits:

| Prefix | Effect | |---|---| | fix: ... | patch, 1.0.0 to 1.0.1 | | feat: ... | minor, 1.0.0 to 1.1.0 | | feat!: ... or a BREAKING CHANGE: footer | major, 1.0.0 to 2.0.0 | | docs:, chore:, test:, refactor: | no release |

The release workflow runs release-please on every push to main. It keeps a single release PR open ("chore(main): release X.Y.Z") holding the version bump in package.json and the new CHANGELOG.md section. Nothing publishes while that PR sits there. Merging it tags vX.Y.Z, cuts the GitHub release, and triggers the publish job, which runs the suite and then npm publish with the NPM_TOKEN repository secret.

release-please-config.json and .release-please-manifest.json hold the release type and the current version. The manifest is the source of truth for what ships next, so let release-please edit it rather than bumping package.json by hand.

Two one-time prerequisites: the NPM_TOKEN secret, and "Allow GitHub Actions to create and approve pull requests" enabled in the repository's Actions settings, without which release-please cannot open its PR. The very first publish of a scoped package also has to be done by hand (npm publish) so npm records it as public.

License

MIT