Skip to content

SunrisesIllNeverSee/sigrank-vscode

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

SigRank Token Efficiency

Measure your AI coding operator token efficiency (Yield/Υ) inline in VS Code.

What it does

SigRank reads token counts only (never prompt content) from AI coding tools used inside VS Code, computes your token cascade efficiency, and displays it in the status bar + a dashboard panel. You can submit your session to the SigRank leaderboard with one click.

The metric: Yield (Υ) = (cache_read × output) / input² — rewards cache reuse + output per unit of input, not gross consumption. Two operators can spend the same and get wildly different leverage.

Supported tools

Tool Token counts read from Status
Claude Code ~/.claude/projects/*/*.jsonl ✅ Supported
Roo Code ~/.roo/usage-tracking.json ✅ Supported
Cline ~/.cline/data/tasks/ ⏳ Blocked (cache tokens not persisted to disk)
Continue ~/.continue/ ⏳ Blocked (VS Code extension doesn't write prompt.log)
GitHub Copilot Chat workspace storage debug logs ⏳ Investigation pending (cache_create missing)

Privacy

This extension reads token COUNTS only — never prompt content, code, or transcripts.

  • Token counts (input, output, cache_read, cache_create) are read from local log files.
  • No prompt text, code, or conversation content is read, stored, or transmitted.
  • Submission to the leaderboard is opt-in (manual button click) unless sigrank.autoSubmit is explicitly enabled.
  • ed25519 device keys are stored locally at ~/.sigrank/keypair.json. The private key never leaves your machine.
  • The only data sent to the server is: token counts, your codename, your device ID, and the ed25519 signature.

Getting started

  1. Install the extension from the VS Code Marketplace.
  2. Enroll your device: run SigRank: Enroll Device from the command palette. You'll need a sign-in code from signalaf.com.
  3. Start coding with Claude Code (in the integrated terminal) or Roo Code. The status bar will show your live Υ once token data is available.
  4. Submit to the leaderboard: run SigRank: Submit Session to Leaderboard or click the submit button in the dashboard.

Commands

Command Description
SigRank: Show Dashboard Open the dashboard panel with full metrics + submit button
SigRank: Submit Session to Leaderboard Submit your current token counts to the SigRank board
SigRank: Enroll Device Link this device to your operator account (one-time)

Settings

Setting Default Description
sigrank.autoSubmit false Automatically submit token counts at session end (off by default — privacy-first)
sigrank.tools.claudeCode true Read Claude Code session logs
sigrank.tools.rooCode true Read Roo Code usage tracking
sigrank.statusBar.showLeverage true Show leverage alongside Υ in the status bar
sigrank.apiBaseUrl https://signalaf.com SigRank API base URL (change for self-hosted instances)

How it works

VS Code Extension
  ├─ reads local AI-tool logs (Claude Code JSONL, Roo Code JSON) → token counts only
  ├─ computes Υ = (cache_read × output) / input² → same cascade engine as the web app
  ├─ status bar shows live score → zero context-switch
  ├─ webview dashboard shows full breakdown + leaderboard link
  └─ submit → ed25519-signed POST to /api/v1/snapshots → same endpoint as the Python agent

The extension is a thin UI + log-reading shell over the existing SigRank architecture. The cascade engine, the ingest endpoint, and the leaderboard all already exist — the extension just reads local logs + renders the score inline + submits.

Built on

  • ccusage — the token-usage reader that SigRank builds on (the extension reads the same Claude Code logs ccusage reads)
  • @noble/ed25519 — pure-JS ed25519 signing (no native modules)

Links

License

See LICENSE.

About

Measure your AI coding operator token efficiency (Yield/Υ) inline in VS Code. Token counts only — never prompt content.

Topics

Resources

License

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors