100% Offline
0 tokens consumed
No signup required
Free forever
Every Claude model
Free VS Code Extension · Claude Code
Claude Code, finally visible.

ClaudeSense reads your local session logs and tells you how Claude is really performing — mood signals, health scores, prompt coaching, and cost tracking. Free. Offline. Nothing leaves your machine.

Session mood · turn by turn · live
😊
🙂
😐
😕
😩
😫
😐
🙂
😊
Session health B · 78
Sound familiar?

Flying blind with
Claude Code?

Claude Code is powerful — but it gives you no feedback on how well you're using it. ClaudeSense fills that gap.

💸

No idea where the money goes

The Anthropic dashboard shows a total. But which session cost $40? Which single prompt burned half your budget? Which model was overkill for what you asked? There's no way to know.

🔁

Repeating the same mistakes

Editing files before reading them. Vague prompts that need three retries. Marathon sessions where quality slowly tanks. Nobody tells you — so the habits stick.

📏

Context fills up. Quality quietly drops.

You don't know you're at 85% context until Claude starts forgetting your instructions from two hours ago. By then, you've wasted time and money on output you'll have to redo.

Session Intelligence

Know the moment
a session turns.

Every response gets a mood score. Watch frustration build, catch drift before it costs you a restart.

  • Per-turn emoji mood — 😊 clean · 😐 meh · 😩 frustrated · 😫 disaster, scored every turn
  • Session signals — the dominant pattern of the session: high corrections, uncached bursts, recovery arcs
  • Live updates — mood tracks in real time while Claude Code is running, not just after
Session Mood B · 78
😊
🙂
😐
😕
😩
😫
🙂
😊
🔁
High Corrections
38% of turns re-issued — Claude drifting.
📦
Uncached Bursts
4 turns with zero cache reuse.
Recovery
Session pulled back to 🙂 by turn 8.
Tool Usage · This Session
Most-used tools
Read
34×
Edit
24×
Bash
16×
Write
1.4×
reads per edit
83%
informed edits
4
blind edits
⚠ 4 files edited without a prior read — Claude may have guessed at their contents.
Tool Intelligence

See exactly what Claude
does, turn by turn.

Tool call breakdown, read-before-edit ratio, and blind edit detection. Understand Claude's reasoning — not just its output.

  • Blind edit flag — catches every file Claude modified without reading first. The single clearest signal of guesswork.
  • Tool usage bars — see which tools dominated and how that compares to well-structured sessions
  • Read / Write ratio — one number that tells you whether Claude was working with knowledge or improvising
Prompt Coaching

Build better prompting habits,
automatically.

ClaudeSense scores every prompt you send against 22 rules and 6 framework detectors — then shows you the pattern that's holding your sessions back. No extra work. It happens in the background, from your existing logs.

Your Prompter Persona · This Session
🎯
The Sniper

Tight, focused prompts. You know exactly what you want and you lead with the verb. Claude responds cleanly.

Avg prompt score 84 / 100 · 9 prompts analysed
RISEN ×3 CoT ×2 XML Tags ×4 COSTAR APE Few-Shot
Top Fired Rules · This Week
A003Vague Task
18×
A001Missing Format
14×
V001Politeness Padding
11×
S001Multi-task
A002No Length Cap
Weekly Pattern
📐
The Frameworker

9 sessions used structured frameworks this week. Your correction rate is 12% below average.

At a Glance

One dashboard. Every metric.

A-F health grades per session, a 90-day activity heatmap, weekly spend totals, and per-turn cost breakdown — all inside VS Code, no browser tab required.

A 91
Excellent session
High cache · clean prompts · no blind edits
Cache Efficiency25/25
Prompt Focus22/25
Code Awareness25/25
Context Health19/25
Prompt Craft20/25
90-Day Activity Heatmap Jan → Apr 2025
127
sessions · 90 days
$41.20
total spend est.
B+
avg health grade
4.8K
turns logged
Prompt Lab

Score your prompt
before you send it.

A built-in prompt composer in the activity bar. Type or paste a prompt, get a score in under 500ms. 22 rules, 6 framework detectors, 22 ready-to-use templates. All local, zero tokens.

Open the Lab  ⌘ Shift L
Getting Started

Three steps.
30 seconds.

01 — Install

Search "ClaudeSense" in VS Code extensions

One click from the VS Code Marketplace. No account, no configuration, no setup. Works the moment it's installed.

02 — Open

Click the CS mark in the activity bar

Your sessions appear instantly — graded, sorted, browsable. The status bar shows live context percentage while Claude Code runs.

03 — Improve

Pick one coaching nudge per session

ClaudeSense points to the exact habit pulling your health score down. Fix one thing. Watch the score move. Repeat.

Privacy

Zero tokens consumed.
Zero data sent.

ClaudeSense reads only your local ~/.claude/projects/ files — the same logs Claude Code already writes to your disk. No network calls. No telemetry. No tokens consumed by the extension itself. The only tokens you ever spend are the ones you send to Claude Code yourself.

⚡ Offline · 0 tokens
No account · No signup · No telemetry
FAQ

Questions worth answering.

ClaudeSense is free. The only costs are the ones Anthropic already bills you for using Claude Code itself — the extension doesn't add any.
No. Zero network calls in the extension. It reads local files, renders a dashboard, writes a pricing config to your home directory. That's the whole I/O surface.
Yes — it reads sessions regardless of how you're billed. But the dollar estimates use public API rates, so they won't match your actual Pro/Max subscription bill. Treat the cost numbers as directional, not a bill.
A built-in prompt composer in the activity bar sidebar. Paste or type a prompt, get a score in under 500ms. 22 scoring rules, 6 framework detectors (COSTAR / RISEN / APE / CoT / Few-Shot / XML Tags), 22 templates. Everything runs locally — no API calls, no tokens consumed.
Yes. Edit ~/.claude/claudesense-pricing.json — rates are per million tokens, and the file regenerates with sensible defaults if you delete it. Useful if you're on Bedrock or Vertex with different pricing.
No — Claude Code only. ClaudeSense parses the specific JSONL format that Claude Code writes to ~/.claude/projects/. Other tools use different log formats or none at all.
Who built this
IS
Ishdeep S Sahni
Senior Staff PM · AI Workflow & Skills Coach

I built ClaudeSense because I kept losing track of what was actually happening inside my Claude Code sessions — not the output, but the process. How much context am I burning? Am I prompting well or just retrying? Is Claude reading files before editing them? The logs had all the answers. ClaudeSense makes them readable.

Ready?

Claude Code,
finally visible.

Install in 10 seconds. No signup. Free forever.

Install ClaudeSense · VS Code Marketplace
Zero tokens consumed 100% local · offline No account required Free forever