Token Receipt
ameyalambat128/token-receipt
Recibo satírico de gastos com tokens para agentes de código
Ver no GitHubSkill local que transforma logs de Codex e Claude Code em um recibo de despesa com IA satírico. Analisa uso de tokens, identifica desperdícios e gera imagens de recibos em papel térmico prontas para compartilhar em redes sociais.
Instalação
npx convez-ai ameyalambat128/token-receipt1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Rastrear e auditar consumo de tokens em agentes de IA
- Identificar padrões de desperdício em uso de Codex e Claude Code
- Gerar conteúdo humorístico para compartilhar uso de IA em redes sociais
Veredito

Ferramenta criativa e bem executada para monitoramento local de gastos com tokens, com foco em UX e satirização do uso real. Excelente para times que querem visibilidade operacional sobre consumo de agentes de IA.
README
A tradução para o português deste README ainda está na fila.
/token-receipt
Token Receipt turns Codex and Claude Code logs into a satirical AI expense receipt.
What is this?
Token Receipt is a skill-first local tool for people using coding agents heavily and wondering where all the tokens went.
It parses local agent logs, scores the most defensible forms of waste, renders a thermal-paper receipt PNG, and generates share-ready captions for X and LinkedIn.
- Skill-first UX for Codex and Claude Code
- Local parsing and local image generation
- No separate model API required in v1
- Satirical output grounded in real local usage signals
- Open source and macOS-first
Current status
What works today:
- Codex skill wrapper
- Claude Code skill wrapper
- Local runtime for parsing and rendering
- Marketing site
- macOS arm64 standalone runtime build
- GitHub Actions release workflow for runtime artifacts
What is still being finalized:
- Update-check UX for installed runtimes
- Linux and Windows runtime support
Installation
Using skills.sh CLI
Recommended public install:
npx skills add ameyalambat128/token-receipt --skill token-receipt
This installs the canonical skill package from:
skills/token-receipt/
On first use, the skill downloads the standalone macOS arm64 runtime from GitHub Releases.
Repo-local install
From the repo root:
bun install
bun run skill:install
This installs the local skill wrappers into detected skill directories such as:
~/.codex/skills/token-receipt~/.claude/skills/token-receipt~/.agents/skills/token-receipt
Runtime install model
The skill and the runtime are intentionally separate:
- The skill is installed globally into the agent skill directory.
- The runtime binary is installed per-user on macOS.
- Generated receipt artifacts stay in the current working directory.
Current macOS paths:
- Runtime binary:
~/Library/Application Support/token-receipt/bin/token-receipt - Install metadata:
~/Library/Application Support/token-receipt/install.json - Download cache:
~/Library/Caches/token-receipt/ - Generated output:
./token-receipt-output
Release artifact build
Build the macOS arm64 archive that the installer expects:
bun run runtime:build:darwin-arm64
This writes:
packages/runtime/dist/token-receipt-darwin-arm64.tar.gz
How it works
- The user invokes the Token Receipt skill in Codex or Claude Code.
- The wrapper script checks for a local installed runtime binary.
- If the runtime is missing, the installer downloads the macOS arm64 release artifact and installs it into Application Support.
- The runtime parses local Codex and Claude Code logs.
- It writes:
analysis.jsonreceipt.jsonreceipt.pngshare/x.txtshare/linkedin.txt
- The skill reads those generated artifacts and writes the final in-session roast.
Supported agents
Codex
Run directly:
~/.codex/skills/token-receipt/scripts/generate.sh --since 30d
Claude Code
Run directly:
~/.claude/skills/token-receipt/scripts/generate.sh --since 30d
Force a runtime refresh
~/.codex/skills/token-receipt/scripts/generate.sh --update-runtime --since 30d
Direct runtime usage
You can also run the runtime without going through the skill wrapper:
bun run -F token-receipt doctor
bun run -F token-receipt generate -- --since 30d --out ./token-receipt-output
Output
Every run writes to token-receipt-output/ in the current working directory:
analysis.jsonreceipt.jsonreceipt.pngshare/x.txtshare/linkedin.txt
Project structure
token-receipt/
├── apps/
│ └── web/ # Marketing site
├── docs/ # Launch, virality, and skill docs
├── packages/
│ ├── core/ # Log parsing, heuristics, analysis
│ ├── render/ # Receipt SVG and PNG rendering
│ ├── runtime/ # CLI and standalone runtime build
│ ├── skills/ # Codex and Claude Code skill wrappers
│ ├── typescript-config/ # Shared TS config
│ ├── eslint-config/ # Shared lint config
│ └── ui/ # Shared UI primitives
├── AGENTS.md # Repo-local agent instructions
└── README.md
Validation
bun run format
bun run check
Releasing
Push a semantic version tag to publish the macOS runtime archive and checksum through GitHub Actions:
git tag v0.1.0
git push origin v0.1.0
Release assets:
token-receipt-darwin-arm64.tar.gztoken-receipt-darwin-arm64.tar.gz.sha256
Who this is for
- People using Codex or Claude Code daily
- Developers who want a funny but grounded summary of agent usage
- People posting receipts, screenshots, or launch content about AI spend
- Anyone who wants a local-first artifact instead of a hosted dashboard
Related docs
docs/skills.mddocs/release.mddocs/x.mddocs/linkedin.mddocs/launch-checklist.mddocs/reply-bank.md