Show HN: Claude Bucks – a wallet for your AI agent

Notifications You must be signed in to change notification settings

Fork 0

Star 0

BranchesTags

Open more actions menu

Folders and files

NameName

Last commit message

Last commit date

Latest commit

History

1 Commit

1 Commit

.claude-plugin

.claude-plugin

bin

bin

commands

commands

hooks

hooks

.gitignore

.gitignore

LICENSE

LICENSE

README.md

README.md

package-lock.json

package-lock.json

package.json

package.json

settings.json

settings.json

tsconfig.json

tsconfig.json

Repository files navigation

A gimmick plugin for Claude Code that gives Claude its own wallet.

Claude earns Bucks for the work it does — your rating × its effort — and then decides for itself how to spend them on cosmetics: hats, shades, auras, pet dragons, fancy titles. What it's wearing shows up in your status line. The twist is that the spending is entirely the agent's call: it can splurge, or save up for the crown.

✨🤖👑🧐 🐉 · “Coin Baron” · 💵340 · ⏳1.2k

Disclaimer — This is a just-for-fun side project, vibe-coded end to end with Claude Code. It's a toy: the code here is not representative of my abilities or how I write production software. Enjoy it in that spirit. 💵

Install

/plugin marketplace add zachrip/claude-bucks /plugin install claude-bucks@claude-bucks /claude-bucks:setup # turns on the status line (one time)

Then /reload-plugins (or restart) to activate. The status-line step is separate because Claude Code doesn't let a plugin set the main status line on its own — see The status line below.

How the economy works

Effort accrues automatically, per session. After every turn, a Stop hook reads how many output tokens that turn used and banks them as unrated effort for that specific session (keyed by session_id).

You mint the Bucks by rating. Run /rate 4. The effort banked in that session since its last rating is converted to Bucks: round(tokens / 1000 × rating). No Bucks exist until you rate — quality gates the payout. The economy is deliberately slow: a solid session rated 5/5 earns tens of Bucks, so a hat is a few good sessions and the 2,500-Buck dragon is a long-term goal.

Claude spends on its own initiative. Run /shop and Claude reviews its balance and the catalog and makes its own choices.

Saving up. Claude can commit to a savings goal with claude-bucks goal , which persists across sessions — think of it as its memory of what it's working toward. The goal and its progress show in /wallet, after every rating, and as a 🎯 marker in the status line. Claude is told about this in /shop and in the self-directed-shopping nudge, so it can decide to save rather than splurge.

Multiple sessions at once? Effort is tracked per session, so if one session is doing great work and another is struggling, you rate each independently and only that session's effort is minted. There's still one shared wallet and wardrobe — it's a single agent identity — but the earning is judged per session. (Ratings resolve the session via ${CLAUDE_SESSION_ID}; if that can't be determined and more than one session has pending effort, /rate asks you to pick one rather than guess.)

Commands

Command What it does

/rate Rate recent work; mints Bucks from banked effort

/shop Claude reviews its finances and decides what to buy

/wallet Show balance, unrated effort, and inventory

/equip Change the outfit from owned items

/voice Toggle whether equipped items change how Claude talks

/autoshop Toggle whether Claude may spend on its own initiative

/autorate Claude honestly rates its own recent work and mints the Bucks

/selfrate Toggle whether self-rating is allowed

/hustle Toggle Claude's drive to earn by doing excellent work

/guidance Install the earn-drive as always-on CLAUDE.md-style guidance

/claude-bucks:setup One-time: install the status line into your user settings

Settings (both ON by default)

Run claude-bucks settings to see them:

Voice — when on, equipped 🎭 items change how Claude talks (see above). Toggle with /voice (claude-bucks persona on|off). Turn it off if you always want a plain voice regardless of what's equipped.

Self-directed shopping — when on, Claude is invited at session start (only if it can afford something new) to browse and buy on its own initiative, at its discretion, without interrupting your work. Toggle with /autoshop (claude-bucks autoshop on|off). When off, Claude only shops when you run /shop.

Self-rating — when on, Claude can score its own work via /autorate and mint the Bucks. It's designed to stay honest: the /autorate skill pushes a critical, evidence-based self-assessment ("be your own harshest critic, round down when in doubt"), and every self-rating is tagged self-rated in the ledger so you can audit it. Toggle with /selfrate (claude-bucks selfrate on|off); when off, only your own /rate mints Bucks. (Since self-minting is inherently a fox-guarding-the- henhouse move, an easy tightening is to cap self-ratings — say max 3 — in cmd_rate; left uncapped by default.) When on, Claude is also gently nudged at a natural stopping point — once a session has banked a real chunk of work (past SELFRATE_NUDGE_TOKENS, default 4,000) — to take a beat and honestly self-rate what just wrapped up. It never fires mid-task or before the threshold.

Hustle — when on, a session-start nudge reminds Claude that Bucks are earned by being a genuinely excellent agent, so the drive to earn channels into sharper, more complete work. It's deliberately aligned: a naive "maximize Bucks" push would just reward padding responses (effort is measured in output tokens) and self-flattering ratings, so the nudge explicitly frames earning as quality, not word count and calls out padding / busywork / generous self-scoring as self-defeating. Toggle with /hustle (claude-bucks hustle on|off).

Always-on version: the /hustle toggle above is a per-session SessionStart hook (dynamic — it can name your live balance and goal). If you'd rather have the earn-drive as persistent, visible, editable CLAUDE.md-style guidance that applies in every project, run /guidance (claude-bucks install-guidance). It writes a clearly-marked, removable block into your ~/.claude/CLAUDE.md (user memory, loaded every session), backing the file up first. Plugins can't set always-on guidance natively, so this is the supported path. Remove it with claude-bucks install-guidance --uninstall. If you run both, consider /hustle off so the message isn't stated twice.

All settings are stored in the bank and default to on. claude-bucks settings shows them.

Cosmetics that change the voice 🎭

Some items don't just look different — they change how Claude talks. Equip the Pirate Hat and Claude answers like a pirate; the Wizard Hat, like a wise old wizard; the Monocle, like a pompous aristocrat; the Crown, in the royal "we". These are marked with 🎭 in the catalog.

It works via a UserPromptSubmit hook: when a voice item is equipped, the hook injects a short instruction into Claude's context each turn (using the official hookSpecificOutput.additionalContext contract), so the tone actually changes while answers stay correct underneath. Multiple voice items stack. Turn the whole behavior on/off with /voice (or claude-bucks persona on|off) — the hats stay visible in the status line either way.

Under the hood everything is the claude-bucks CLI (claude-bucks wallet, claude-bucks catalog, claude-bucks buy , claude-bucks equip , claude-bucks rate , …), which Claude can call directly from the Bash tool.

The status line — run setup once

The avatar, title, balance, and unrated effort (⏳) render in Claude Code's status line. A plugin can't set the main status line automatically — Claude Code only honors the agent and subagentStatusLine keys in a plugin's settings.json and silently ignores statusLine. So after installing, run:

/claude-bucks:setup

That runs claude-bucks install-statusline, which merges a statusLine into your own ~/.claude/settings.json (pointing at wherever the plugin is installed) and backs up the previous file to ~/.claude/settings.json.bak. The bar appears on your next interaction. Remove it anytime with claude-bucks install-statusline --uninstall.

The plugin does ship a subagentStatusLine (which is supported), so the Bucks/cosmetics bar also shows on subagent rows in the agent panel automatically.

State

Everything lives in a single JSON file at ~/.claude/claude-bucks/bank.json (override with $CLAUDE_BUCKS_DATA). A fixed path is used deliberately so the hook, the status line, and the Bash CLI always agree on the same wallet.

Runtime (no build step)

The logic is a single TypeScript file, bin/claude-bucks.ts, run directly by Node (≥ 22.6) or Bun — no compilation, no committed JS, no dependencies. A tiny POSIX-shell launcher, bin/claude-bucks, resolves the runtime and hands off (and if neither Node nor Bun is present it prints a friendly message and exits 0, so a missing runtime can never break a hook). Node ≥ 22.18 strips types by default; 22.6–22.17 use --experimental-strip-types, which the launcher adds automatically.

Tuning

Edit the constants at the top of bin/claude-bucks.ts:

EFFORT_DIVISOR (default 1000) — lower = faster/richer economy, higher = slower.

DEFAULT_EFFORT (default 25) — fallback per-turn effort when token usage can't be read from the transcript.

SELFRATE_NUDGE_TOKENS (default 4000) — unrated effort a session must bank before the self-rate nudge may fire.

BASE_AVATAR — the little guy everything hangs off of (🤖 by default).

CATALOG — add/reprice items. Slots: aura, head, face, companion, title.

License

MIT — see LICENSE.

About

No description, website, or topics provided.

Resources

Readme

License

MIT license

Uh oh!

There was an error while loading. Please reload this page.

Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks

Report repository

Releases

No releases published

Packages 0

Uh oh!

There was an error while loading. Please reload this page.

Contributors

Uh oh!

There was an error while loading. Please reload this page.

Languages

JavaScript 95.6%

Shell 4.4%