Token Receipt: The Screenshot That Shows Your AI Coding Agent's True Cost

Let's be honest: when you're deep in a coding session with Claude Code, Codex, or Cursor, token counts feel abstract. You're not watching dollars drain from an account in real-time. But at the end of the month, the invoice arrives—and it's full of surprises.

Token Receipt flips this experience on its head. Instead of waiting for a bill, you get one mid-sprint. It's a receipt, styled like a thermal-paper printout, that itemizes what your coding agent has actually been doing in your local sessions.

Not Just About the Numbers

Here's what makes Token Receipt genuinely interesting: it's not trying to be another observability dashboard. The developer behind it understood something important—developers share screenshots. We drop error messages into Slack, post terminal output in GitHub issues, and screenshot unusual behaviors for our teams.

A receipt you can actually screenshot and share hits differently than a graph in a dashboard. It tells a story: "Look at how many times I made this agent re-read the same file" or "Turns out my context window is basically an emotional support animal."

What Actually Ends Up on the Receipt

The tool parses local session logs from whatever agent you're running and extracts patterns. V1 focuses on signals that are defensible from the logs themselves—things like repeated file reads, planning-before-touching patterns, MCP tool tourism (yes, that's a real category), and those delightful subagent middle management situations where your agent spawns helpers to coordinate helpers.

Cache-heavy sessions and low-output expensive runs also make the cut. If you've ever watched an agent methodically explore your codebase, accomplish a three-line fix, and then generate what feels like a small novel of output—those sessions get their own line item.

The Local-First Philosophy

One thing worth highlighting: Token Receipt doesn't require yet another API key. The skill-first path uses your existing Codex or Claude Code session to do the final formatting. The runtime stays local, reads your session logs, and generates structured facts. No telemetry, no uploads of your prompts or code in v1.

This approach matters for developer trust. When you're using a tool that analyzes your coding patterns, you probably don't want that data flying somewhere else. Token Receipt respects that boundary by design.

Why Kiro CLI Gets Special Treatment

Kiro CLI support is interesting because it uses local credit usage instead of token-derived estimates. The reasoning is practical: not every agent exposes the same counters. So instead of guessing, Token Receipt reads what's actually available and works with those signals. If you're using Kiro as part of your daily workflow, your receipt will reflect those sessions alongside Codex and Claude Code.

Cursor support is marked experimental for similar reasons—tool activity and workspace metadata are readable, but the spend accounting isn't as precise as the other agents. The trade-off is documented clearly: Cursor receipts show behavior but not exact costs.

The Real Value Proposition

At its core, Token Receipt is a self-awareness tool dressed up as a fun novelty. Those "bad habits" it exposes—repeated reads, context bloat, over-planning—aren't bugs. They're patterns that might indicate your workflow could be optimized, or that your prompts need refinement, or that you need a different approach to working with your agent.

The screenshot format makes it easy to share these insights with teammates. When everyone's laughing about how many times the agent re-read the config file, you're also implicitly having a conversation about how to work more efficiently.

Getting Started

If you want to see what your agent sessions actually look like itemized, installation is straightforward through the skills system:

npx skills add ameyalambat128/token-receipt --skill token-receipt

Then just call /token-receipt in your next session. Or get specific: /token-receipt Generate a receipt for my last 30 days of usage if you want the full historical roast.

The project is MIT licensed and open source, so if you want to extend it for other agents or customize the receipt formatting, the repository is available for contribution.

Bottom Line

Token Receipt won't save you money directly—it's not cutting your API costs. But it might change how you think about working with AI coding agents. Once you see your habits itemized, it's hard to go back to blindly trusting the context window. Sometimes a thermal-paper receipt is exactly the accountability you didn't know you needed.