Copilot Insights

Understand how you work with AI. Build the skills to do it better.
A local-first dashboard and Copilot CLI extension that analyzes your AI coding sessions,
connects collaboration patterns to token usage, and helps you improve over time.

Note: This is an independent, community-built project. It is not affiliated with, endorsed by, or sponsored by GitHub or Microsoft. "GitHub Copilot" is a trademark of GitHub, Inc.

Scope: Copilot Insights is built around GitHub Copilot CLI session history. VS Code / VS Code Insiders Copilot session analysis is available as an optional data source that you can enable from Settings when you want the dashboard to read supported local workspace storage.

Quick Start • Dashboard • CLI Tools • How It Works

Why?

Every time you say "no, not that" or "go back to the previous approach," that's signal. It means there's a gap between what you asked for and what the agent did. Copilot Insights surfaces those moments so you can learn from them.

📊 See your patterns — Which corrections do you make most often?
🎯 Skill Building — 7-pillar WTI framework with dev plan, retros, and progress tracking
💡 Get coaching — Personalized dev plans, daily check-ins, and retros
📈 Watch your trends — Pillar scores over 7/30/90 days or all time
🔍 Replay sessions — Annotated turn-by-turn session replay
🧪 Practice prompting — Sandbox for instant feedback + rewrite challenges with coaching nudges
💻 VS Code Sessions — Optionally analyze Copilot sessions from VS Code / VS Code Insiders when enabled in Settings
📡 Live monitoring — Real-time feed of corrections as they happen
💰 Local token/session usage — See input, output, cached tokens, local prompts/requests, estimated credits, and optimization tips
🌗 Dark & light mode — System-aware theme with manual toggle and localStorage persistence
♿ Accessible — Full WAI-ARIA keyboard navigation, semantic HTML, focus-visible styles

Inspired by this investigation into using Copilot session data as telemetry for agent improvement.

What It Detects

Category	Examples
🔄 Iterative Refinement	"no, that's wrong", "not what I asked", "this is wrong"
↩️ Course Change	"actually, do X instead", "scratch that", "I changed my mind"
💬 Clarification Needed	"still broken", "I already said", "why did you do that"
🔁 Reinforced Instruction	"like I said", "one more time", "again"
🔀 Direction Change	"undo that", "go back", "revert", "change it back"

It also detects repeated file edits — when the same file is edited 3+ times in a session, which often indicates unclear requirements.

Quick Start

Prerequisites

Node.js 20+
Copilot CLI installed and used at least once (creates ~/.copilot/session-store.db)
OS: macOS, Linux, or Windows

Fastest Path

Install or upgrade to the latest release, then launch the dashboard and optionally link the Copilot CLI extension. If the dashboard is empty, run copilot once first so Copilot CLI creates ~/.copilot/session-store.db.

Windows

irm https://raw.githubusercontent.com/jackbatzner/copilot-insights/main/scripts/install.ps1 | iex
copilot-insights
copilot-insights link

macOS / Linux

curl -fsSL https://raw.githubusercontent.com/jackbatzner/copilot-insights/main/scripts/install.sh | bash
copilot-insights
copilot-insights link

Run the same installer command again any time you want to upgrade to the newest release. Then open http://localhost:3002, and reload Copilot CLI extensions with /extensions reload if you linked the extension.

1. Install

Option A: Recommended installer scripts

Windows

irm https://raw.githubusercontent.com/jackbatzner/copilot-insights/main/scripts/install.ps1 | iex

macOS / Linux

curl -fsSL https://raw.githubusercontent.com/jackbatzner/copilot-insights/main/scripts/install.sh | bash

Both scripts download the latest GitHub Release tarball, verify its SHA-256 checksum, and install it globally with npm. Rerun the same command later to upgrade.

$env:COPILOT_INSIGHTS_VERSION = "v0.2.4"
irm https://raw.githubusercontent.com/jackbatzner/copilot-insights/main/scripts/install.ps1 | iex
Remove-Item Env:\COPILOT_INSIGHTS_VERSION

curl -fsSL https://raw.githubusercontent.com/jackbatzner/copilot-insights/main/scripts/install.sh | bash -s -- v0.2.4

Use -Force or --force after downloading the script locally if you want to reinstall the currently installed version, and -DryRun or --dry-run to resolve the release without changing your machine.

Option B: GitHub shorthand fallback

npm i -g jackbatzner/copilot-insights
copilot-insights
# → http://localhost:3002

This still works for many environments, but the installer scripts are more reliable on Windows because they install from the packaged GitHub Release tarball instead of npm's git dependency path.

Option C: From source

git clone https://github.com/jackbatzner/copilot-insights.git
cd copilot-insights
npm install

npm install is the normal source-checkout path. It installs the nested server/ and ui/ dependencies and builds the UI via the root postinstall script. Use npm run setup if you want the same bootstrap flow without relying on root install scripts.

2. Launch the Dashboard

copilot-insights   # if installed from GitHub
npm start          # if installed from source
# → http://localhost:3002

Open http://localhost:3002 to see your dashboard. Run copilot-insights --help any time to see the available commands and flags.

3. Use as a Copilot CLI Extension (optional)

This repo follows the official Copilot CLI extension layout: .github/extensions/copilot-insights/extension.mjs.

To get insights directly inside Copilot CLI chat from a global or source install, link the packaged extension directory into your Copilot user extensions folder:

The Copilot SDK is supplied by Copilot CLI when the extension runs, so a normal npm i -g jackbatzner/copilot-insights install does not need a separate @github/copilot-sdk package.

copilot-insights link    # if installed globally
node bin/cli.mjs link    # if running from source

To remove the extension later:

copilot-insights unlink    # if installed globally
node bin/cli.mjs unlink    # if running from source

Then reload extensions (/extensions reload) or restart Copilot CLI. The extension registers 8 tools that the agent can invoke:

> How am I doing with my prompting?          → insights_summary
> Scan my recent sessions                    → insights_analyze
> What are my most common correction patterns? → insights_patterns
> Compare sessions abc123 and def456         → insights_compare
> Coach this prompt before I send it         → insights_coach
> Launch the insights dashboard              → insights_dashboard
> Stop the dashboard                         → insights_stop

Dashboard

First-time users are greeted with a Welcome walkthrough that explains what the tool does before diving in:

The main dashboard uses progressive disclosure — new users see the key insight, quick stats, and clear next steps. Deeper data is available in collapsible sections. Navigation is organized into Core (Overview, Skill Building, Practice Lab, Sessions) and Advanced (Token Usage, Analytics, Live, Instructions, Settings, VS Code) groups:

📸 More screenshots

Sessions list — sortable, filterable by repo, hide noisy sessions from analysis

Skill Building overview — 7-part WTI workspace with pillar cards, quick wins, retro, and dev plan entry points

Skill Building: Intent — drill into one scored pillar with evidence, guidance, and targeted improvements

Analytics — work style, prompt length, session depth

Skill Building: Dev Plan — track improvement goals, habits, and follow-up actions in one place

Practice Lab — sandbox for instant prompt feedback + rewrite challenges with coaching panel

Instructions — convention gap analysis with copy-paste snippets

Settings — enable optional VS Code loading without changing the default CLI-first workflow

Session Detail — turn-by-turn replay with annotations

Live Monitor — real-time feed with pattern badges and coaching alerts

VS Code Sessions — opt in from Settings, then analyze Copilot sessions from VS Code with turn-by-turn detail

Token Usage — local session DB/session-state usage with input/output/cached tokens, prompts, requests, estimated local credits, and usage trends

Token Optimization — personalized token-saving tips, model recommendations, and savings plan

Token Models — per-model breakdown with local token totals, cache hit rate, estimated local credits, and usage distribution

Pages

Welcome — 3-step onboarding for first-time users; always accessible via /welcome
Overview — Tier hero card, collapsible stats/trends/skill growth/work style sections, prompting insights; "Since last visit" summary for returning users
Skill Building — 7-pillar WTI framework (Intent, Work Design, Quality Control, Evaluation, plus Retro, Dev Plan, Overview) with pillar cards, tabbed deep-dives, trend detection with sample-size guard, Chronicle tips, and dev plan goal tracking
Sessions — Sortable table of all sessions with redirections, filterable by repo; hide noisy sessions from analysis
Session Detail — Turn-by-turn timeline showing exactly where corrections happened; hide/unhide from detail view
Analytics — Hourly productivity, prompt length, repo health, tool usage
Practice Lab — Sandbox for instant prompt feedback (score, pattern detection, coaching nudges) + rewrite challenges using your real sessions or curated examples
Instructions — Custom instruction effectiveness analysis with ready-to-paste markdown snippets
Settings — Enable optional data sources like VS Code loading while keeping the default workflow focused on Copilot CLI sessions
VS Code Sessions — Opt-in analysis for Copilot sessions from VS Code and VS Code Insiders with turn-by-turn detail, retry on error, and expandable session cards
Live Monitor — Real-time session feed with pattern badges, coaching alerts, pause/resume
Token Usage — 5-tab local usage dashboard (Overview, Models, Cost Insights, Budget, Optimization) with progressive token/session loading, input/output/cached token breakdowns, prompts, inferred requests, estimated local credits, model breakdowns, personalized token-saving insights, budget projections, and live pricing from GitHub docs

All pages include a timeframe selector (7d / 30d / 90d / All time) that syncs across pages — change it once and every page updates. Your selection is persisted to localStorage.

The dashboard is mobile responsive: on small screens a hamburger menu replaces the sidebar, tables scroll horizontally, and the layout adapts to narrow viewports. API caching (60-second TTL on GET requests with LRU eviction at 200 entries) keeps navigation snappy, and the refresh button clears the cache for fresh data. Hidden sessions are persisted to disk (~/.copilot/copilot-insights-hidden.json) so they survive server restarts. API resilience includes AbortController with 30s timeout, Promise.allSettled for parallel requests, and retry buttons on error states.

Practice Lab

The Practice Lab is an interactive sandbox for improving your prompting skills:

🧪 Sandbox Mode — Type any prompt and click Analyze to get instant feedback: a 0-100 score, pattern detection (vague language, missing context, etc.), quality checks, and actionable rewrite suggestions. Coaching nudges help you improve — "try mentioning the specific file" or "what should the result look like?"

🏆 Rewrite Challenge — Choose between your own low-scoring prompts ("My Bad Prompts") or a curated library of 80+ bad prompts covering best practices from GitHub, Anthropic, Google, and OpenAI. Filter by topic (vague prompts, missing constraints, no examples, etc.), rewrite the prompt, and see your score improve. Before you rewrite, a coaching panel shows what's wrong: detected problems, missing quality signals, and before/after rewrite examples — so you know exactly what to fix. Personalized recommendations highlight which categories need the most work based on your real session patterns.

Scoring is based on heuristics derived from published prompting guides — see docs/prompting-resources.md for the full list of sources.

Live Monitor

The Live Monitor polls your session database every 5 seconds for new turns and displays them in a live feed. Each turn is annotated with detected redirection patterns shown as colored badges. When a high-severity pattern (weight ≥ 3) is detected, an inline coaching card appears with contextual tips.

Pause/resume — Stop and restart polling with one click
Status indicator — Green pulsing dot when active, yellow when paused
Coaching alerts — Actionable advice for iterative refinements, clarification needs, direction changes, etc.

CLI Tools

Tool	Description
`insights_analyze`	Scan recent sessions, ranked by correction severity
`insights_session`	Deep-dive a specific session with turn-by-turn timeline
`insights_patterns`	Most common correction patterns with real examples
`insights_summary`	Quick snapshot: tier badge, pillar scores, coaching tip
`insights_compare`	Compare two sessions side-by-side
`insights_coach`	Prompt coaching: immediate feedback, periodic review, and progress tracking
`insights_dashboard`	Launch the web dashboard from the CLI
`insights_stop`	Stop the dashboard server

How It Works

The extension reads from ~/.copilot/session-store.db (read-only), the SQLite database where Copilot CLI stores session history. It scans user messages against 30+ regex patterns, categorizes matches, and scores the results. For token usage analysis, it also reads local JSONL event files from ~/.copilot/session-state/ and local usage events when available, then falls back to text estimates when exact counts are unavailable. It does not import GitHub organization usage metrics or billing reports; local estimated credits are coaching estimates based on current Copilot model pricing. VS Code sessions are read from state.vscdb files in VS Code's workspace storage.

~/.copilot/session-store.db (read-only)
  → Read user messages from turns table
  → Match against 30+ correction patterns
  → Categorize and aggregate
  → Chronicle builder: clean messages, classify intent

~/.copilot/session-state/*/events.jsonl (read-only)
  → Read local model and output token data
  → Combine with local DB events for input/output/cached token breakdowns when available
  → Estimate local credits using official GitHub Copilot pricing
  → Generate personalized token/session optimization tips

VS Code workspace storage (read-only)
  → Discover VS Code / VS Code Insiders sessions
  → Analyze clarity, delegation, judgment, efficiency

  → Serve via Express API → React dashboard

Privacy: All data stays local. The tool only reads your existing session database — it never writes to it, and nothing is sent to external services.

Development

# UI dev mode (hot reload on :5174, proxies API to :3002)
cd ui && npm run dev

# Server dev mode (auto-restart on changes)
cd server && npm run dev

See CONTRIBUTING.md for contribution guidelines.

Releasing

Recommended: merge what you want into main, then run the Manual release GitHub Actions workflow from the main branch. Pick patch, minor, major, or explicit, and the workflow will bump versions, update the changelog, commit, tag, and push. The existing tag-based release workflow then validates the tag, packages the tarball, and creates the GitHub Release.

Local fallback:

npm run release patch         # 0.1.0 → 0.1.1
npm run release minor         # 0.1.0 → 0.2.0
npm run release major         # 0.1.0 → 1.0.0
npm run release 0.2.0-beta.1  # explicit version
npm run release patch --dry-run  # preview without changes

The local script follows the same flow: it bumps versions, updates the changelog, commits, tags, and pushes. GitHub Actions then reruns install/build/lint/test validation, packages the tarball, and creates the GitHub Release automatically.

Architecture

graph LR
    subgraph "Data Source"
        DB[(~/.copilot/session-store.db)]
        JSONL[(~/.copilot/session-state/*.jsonl)]
        VSCDB[(VS Code state.vscdb)]
    end

    subgraph "Analysis Engine"
        DB -->|read-only| Analyzer[analyzer.mjs]
        JSONL -->|token data| Tokens[tokens.mjs]
        VSCDB -->|read-only| VSCode[vscode-sessions.mjs]
        Analyzer --> Patterns[30+ regex patterns]
        Analyzer --> Pillars[Clarity · Efficiency · Delegation]
        Analyzer --> Tiers[Tier scoring]
        Tokens --> TokenAnalytics[token-analytics.mjs]
        Patterns --> Practice[practice.mjs]
        DB -->|polling| LiveFeed[/api/live/feed]
        LiveFeed --> Patterns
        DB --> Chronicle[chronicle.mjs]
        Chronicle --> IntentClassifier[intent-classifier.mjs]
        Chronicle --> DevPlan[dev-plan.mjs]
    end

    subgraph "Delivery"
        Analyzer --> API[Express API :3002]
        Practice --> API
        LiveFeed --> API
        TokenAnalytics --> API
        VSCode --> API
        Chronicle --> API
        DevPlan --> API
        API --> UI[React Dashboard]
        Analyzer --> CLI[Copilot CLI Extension]
        CLI --> Tools[8 insights_* tools]
    end

copilot-insights/
├── .github/
│   └── extensions/
│       └── copilot-insights/
│           └── extension.mjs  # Official discovered Copilot CLI entry point
├── src/
│   ├── db.mjs             # SQLite read-only access + batch query support
│   ├── patterns.mjs       # 30+ regex patterns, 5 categories
│   ├── analyzer.mjs       # Core analysis engine
│   ├── chronicle.mjs      # Session chronicle builder with message cleaning
│   ├── intent-classifier.mjs  # WTI-aligned intent classification
│   ├── dev-plan.mjs       # Personalized coaching & scoring
│   ├── vscode-sessions.mjs    # VS Code session discovery & analysis
│   ├── practice.mjs       # Prompt analysis for Practice Lab (no DB dependency)
│   ├── challenge-library.mjs  # 83 curated bad prompts with tags and hints
│   ├── tiers.mjs          # Tier badge system (shared UI + CLI)
│   ├── suggestions.mjs    # Prompt rewrite engine
│   ├── delegation.mjs     # Delegation analysis
│   ├── judgment.mjs       # Judgment analysis
│   ├── efficiency.mjs     # Efficiency scoring
│   ├── clarity.mjs        # Clarity scoring
│   ├── sprawl.mjs         # Tool/file sprawl detection
│   ├── instructions.mjs   # Convention gap analysis
│   ├── instruction-failures.mjs  # Instruction failure detection
│   ├── session-insights.mjs  # Per-session improvement suggestions
│   ├── work-style.mjs     # Work style analysis
│   ├── trends.mjs         # Time-series trend detection
│   ├── tokens.mjs         # Token usage estimation, JSONL reader, model pricing
│   ├── token-analytics.mjs # Cost insights, model recommendations, savings plan
│   └── formatter.mjs      # Markdown formatting (CLI output)
├── server/
│   └── index.mjs          # Express API + static UI + atomic writes + write mutex
├── ui/src/
│   ├── pages/
│   │   ├── SkillBuilding/     # Decomposed skill building module
│   │   │   ├── index.jsx      # Orchestrator (data loading, tabs, retry)
│   │   │   ├── DevPlanContext.jsx  # React context for dev plan state
│   │   │   ├── shared.jsx     # Shared components (AddToDevPlanButton, MiniStat, etc.)
│   │   │   ├── OverviewTab.jsx, IntentTab.jsx, WorkDesignTab.jsx
│   │   │   ├── QualityControlTab.jsx, EvaluationTab.jsx
│   │   │   ├── RetroTab.jsx, DevPlanTab.jsx
│   │   │   └── (7 tab components total)
│   │   ├── Overview.jsx, Welcome.jsx, Sessions.jsx, SessionDetail.jsx
│   │   ├── Analytics.jsx, Practice.jsx, Instructions.jsx
│   │   ├── VSCodeSessions.jsx, LiveMonitor.jsx, TokenUsage.jsx
│   │   └── (12 pages total)
│   ├── components/        # TabBar, CollapsibleSection, EmptyState, MetricHelp, etc.
│   └── api.js             # API client with safeFetch, LRU cache, AbortController
├── docs/
│   ├── prompting-resources.md  # Official guides, academic papers, scoring reference
│   └── screenshots/       # Auto-captured screenshots and demo GIF
├── scripts/               # Mock data seeder + screenshot/GIF capture
└── .github/workflows/     # CI + Release (GitHub Releases)

License

MIT

Disclaimer

This project is not affiliated with, endorsed by, or sponsored by GitHub or Microsoft. "GitHub Copilot" and "Copilot" are trademarks of GitHub, Inc. This project uses the name "Copilot Insights" solely to describe its function as a tool that works with locally available GitHub Copilot session data.

Troubleshooting

"No sessions found" / empty dashboard

You need at least one Copilot CLI session. Run copilot in any repo to create one.
Confirm that ~/.copilot/session-store.db exists on your machine.

"Port 3002 already in use"

Kill the existing process, or launch on a different port: copilot-insights --port 3003

"Cannot find module 'better-sqlite3'"

From a source checkout, run npm install in the project root to install all dependencies.
If you need to bypass root install scripts, run npm run setup instead.

"Copilot session DB is missing required tables"

Your ~/.copilot/session-store.db may be from an incompatible Copilot CLI version.
Update Copilot CLI and try again, or open an issue.

Dashboard loads but shows errors

Ensure the server is running (npm start) before opening the dashboard.
Check the terminal for server-side error messages.

Future Ideas

Gamification — Scoring, XP, levels, achievements, and goals to encourage improvement
Team leaderboards — Anonymous comparison across teams
Custom instruction generation — Auto-generate .github/copilot-instructions.md from common patterns
OpenTelemetry tracing — Opt-in distributed tracing via @github/copilot-sdk for debugging

Name		Name	Last commit message	Last commit date
Latest commit History 162 Commits
.github		.github
bin		bin
docs		docs
scripts		scripts
server		server
src		src
test		test
ui		ui
.editorconfig		.editorconfig
.gitattributes		.gitattributes
.gitignore		.gitignore
.nvmrc		.nvmrc
CHANGELOG.md		CHANGELOG.md
CODE_OF_CONDUCT.md		CODE_OF_CONDUCT.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
SECURITY.md		SECURITY.md
SUPPORT.md		SUPPORT.md
eslint.config.mjs		eslint.config.mjs
logo.svg		logo.svg
package-lock.json		package-lock.json
package.json		package.json

Navigation Menu

Repository files navigation

Copilot Insights

Why?

What It Detects

Quick Start

Prerequisites

Fastest Path

1. Install

2. Launch the Dashboard

3. Use as a Copilot CLI Extension (optional)

Dashboard

Pages

Practice Lab

Live Monitor

CLI Tools

How It Works

Development

Releasing

Architecture

License

Disclaimer

Troubleshooting

Future Ideas

About

Releases

Packages

Used by

Contributors

Languages

Folders and files

Latest commit

History

Repository files navigation

Copilot Insights

Why?

What It Detects

Quick Start

Prerequisites

Fastest Path

1. Install

2. Launch the Dashboard

3. Use as a Copilot CLI Extension (optional)

Dashboard

Pages

Practice Lab

Live Monitor

CLI Tools

How It Works

Development

Releasing

Architecture

License

Disclaimer

Troubleshooting

Future Ideas

About

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

Watchers

Forks

Releases

Packages

Used by

Contributors

Languages