Skillget

tokenscope

skill
Security Audit
Pass
Health Pass
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Community trust — 10 GitHub stars
Code Pass
  • Code scan — Scanned 9 files during light audit, no dangerous patterns found
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

The first open-source token profiler for AI coding agents. Analyzes session logs, detects repeated content and bloated tool results, and shows you exactly where your context window budget goes with concrete fixes and estimated savings.

README.md

npm version

tokenscope

Chrome DevTools for your context window. See exactly where your AI tokens go — and stop wasting them.

tokenscope treemap report

Your agent session just burned 500K tokens. Where did they go? tokenscope profiles your session logs and tells you: which files got re-sent six times, which tool result dumped 20K tokens of JSON the model never used, and what it's costing you.

npx tokenscope-ai ~/.claude/projects/my-project

tokenscope terminal output

Why

  • 100% local. No API key, no account, no telemetry. Your logs never leave your machine.
  • Actionable. Every finding comes with a concrete fix and an estimated saving — like a linter, not a dashboard.
  • Fast. Profiles a session in under a second.

Install

npx tokenscope-ai <session.jsonl>        # zero-install
npm install -g tokenscope-ai            # or install globally

Usage

tokenscope-ai ~/.claude/projects/my-project        # newest session in a directory
tokenscope-ai session.jsonl                        # a specific session file
tokenscope-ai session.jsonl --html report.html    # + shareable HTML treemap report

The HTML report is a single self-contained file — open it in any browser, share it with your team, attach it to a PR.

Supported sources

Source Status
Claude Code session logs (~/.claude/projects/) ✅ v0.1
Universal proxy mode (any tool, any provider) 🔜 v0.2
OpenAI / raw API JSONL dumps 🔜 v0.3
Gemini CLI, aider, community adapters 🔜 adapter spec

Token counts use a local tokenizer (o200k) and are estimates (~±5%) — profiling is about proportions and deltas, not billing precision.

Findings rules

Code Detects
W001 Repeated content — the same chunks sent multiple times
W002 Bloated tool results — oversized outputs dominating the context

More rules (cache-miss analysis, conversation decay, dead-weight system prompt sections) are on the roadmap. Have an idea for a rule? Open an issue.

Contributing

Adapters are ~100 lines: parse your tool's log format into the unified session model (src/parser/claudeCode.js is the reference). PRs welcome.

License

MIT

Reviews (0)

No results found