CLAUDE CODE PLUGIN
v1.3Rebuilt for the Fable 5.1 era — sees your verification loop, rewards autonomy →

See exactly how
you work with Claude.

Claude Radar reads your real project conversations and surfaces your collaboration blind spots across 9 dimensions in 3 categories — Communication, Engineering, Outcome — with Engineering looking at how far you push Claude Code's platform leverage (Skill / MCP / Subagent / CLAUDE.md / Plan). Beyond the scores, it writes a diagnosis that actually pins down your habits, plus at least 5 rewrite prompts you can paste straight into your next session. Work smoother next time — and level up for real.

Get Claude Radar How it works
Claude Radar
sample report · v1.3
Claude Radar sample report
Key Features

Walk away knowing what to change next

More than a score sheet. It tells you where you held things back, why, and exactly what to say instead — every line backed by evidence.

See how far you push Claude Code

Communication — how you direct AI. Engineering — how far you push platform leverage (Skill / MCP / Subagent / Workflow orchestration / parallel fan-out / background tasks / CLAUDE.md / hooks / Plan). Outcome — what you actually ship. 9 dimensions, 3 categories. Most people leave that leverage on the table: you're not docked for skipping advanced tools, only for using them badly — endless retries, Plans opened then abandoned.

Someone points out the problem for you

It's not a cold score sheet and done. You get a diagnosis written to your own collaboration habits: where you're strongest, what bottleneck holds you back most, and what several dimensions together reveal — every line citing real evidence from your sessions. This is the part users value most.

Suggestions you can paste back in — or install outright

Every suggestion comes with concrete wording you can paste straight into your next session, plus its expected impact. Setup suggestions go further and ship the actual file — a CLAUDE.md section, a hook config, an .mcp.json entry — ready to install. Reports are archived locally, so each new check-up shows your score deltas and which past suggestions you actually adopted. All in one single-file HTML, easy to forward to a teammate.

New in v1.3 — your verification loop counts, and autonomy gets measured

Verification is a loop, not a request. Test and build runs, reviewer subagents, browser and UI checks, hooks — Claude Radar credits the loop you built, even if you never typed "test it". And it now measures autonomy: how much of your work runs in long autonomous turns, whether you queue the next instruction instead of interrupting, how much you hand off to background agents. Manual compaction reads as deliberate tempo rather than a penalty, and the playbook skips the moves you're already making. Recalibrated by Claude Fable 5.1 for its own era.

9 Dimensions × 3 Categories

Measured, not guessed

Every score traces back to specific signals in your session history. No black box, no vibes.

Communication
3 DIMS
How clearly you direct AI through text.
Lock-On
Intent Clarity
Goals, constraints, expected outcomes
Scene Setting
Context Quality
Tech stack, file paths, project structure upfront
Steering
Course Correction
Precise corrections with reasoning, not "that's wrong"
Engineering
3 DIMS · NEW
How well you leverage Claude Code as a platform.
Toolcraft
Tool Fluency
Skill / MCP / Subagent / Plan / custom commands usage
Architecture
Setup Quality
CLAUDE.md / Memory / Agents / Settings setup
Tempo
Session Flow
Milestones, summaries, focused scope per turn
Outcome
3 DIMS · NEW
What actually gets shipped.
Efficiency
Work Density
Edits / tools / distinct files per message
Proof Check
Verification
Tests, reviews, no blind "looks good" accepts
Completion
Clean Closure
Sessions close cleanly, no abandoned retries
The Core Value

AI writes the diagnosis

Scores tell you what. Diagnosis tells you why and what to do next. Independent of scoring, evidence-grounded, no personality archetypes.

CORE DIAGNOSIS
"Strength: Your Efficiency + Completion combo is top-tier — work actually ships, sessions close cleanly. Bottleneck: Across 8 sessions you almost never asked AI to explain its plan before executing."
PATTERN READING
"High Lock-On + low Proof Check = you trust AI's execution but not its judgment. Fine for tight, well-scoped work; risky when complexity climbs."
PASTABLE PROMPT REWRITE
"Before you write any code, walk me through what could go wrong with this change and how you'd test it."
Every improvement suggestion ships with a concrete prompt you can copy into your next session, plus expected score impact.
Fairness

A 3-message bugfix is not a 50-session build

Every project is auto-classified into one of 4 profiles, each with its own dimension weights and N/A rules. Small projects no longer get unfairly penalized.

AOne-shot · ≤ 2 sessions, ≤ 15 msgs
BFeature build · 3-20 sessions, balanced edit ratio
CLong-running · 20+ sessions or > 7 days
DLearning · Q&A-heavy, few edits
Density beats volume. A 5-message project with high signal density and 3 file edits gets high confidence and full marks for Efficiency. Confidence scaling rewards signal-rich sessions, not just session count.
Get Started

Three steps to your dashboard

1

Install

Add the marketplace

/plugin marketplace add
2

Run

From your project directory

/claude-radar
3

Dashboard

Opens automatically in browser

single-file HTML

100% on your machine

Your session data never leaves your computer. Ever.

No cloud
No API key
No telemetry
No tracking
~250 KB

Install

/plugin marketplace add lemomo-ai/claude-radar
/plugin install claude-radar@claude-radar-marketplace
/claude-radar

Or clone the repo and pass --plugin-dir to claude.