Skip to content
| Marketplace
Sign in
Visual Studio Code>Visualization>Claude ReportNew to Visual Studio Code? Get it now.
Claude Report

Claude Report

Preview

haitham72

|
1 install
| (0) | Free
Claude Code and Codex usage in your editor: live plan gauges, a Metrics view, and monthly AI achievement reports per project. Reads local transcripts, no telemetry.
Installation
Launch VS Code Quick Open (Ctrl+P), paste the following command, and press enter.
Copied to clipboard
More Info

Claude Report

Claude Report

Claude Report is a VS Code extension for reviewing Claude Code and Codex activity, monitoring Claude plan usage, and writing project achievement reports. It runs in VS Code and compatible editors such as Cursor and Antigravity.

Claude Report is an independent project. It is not made, endorsed or sponsored by Anthropic or OpenAI. Claude and Claude Code are trademarks of Anthropic, PBC.

Sessions tab Metrics tab Report tab

Install and start

  1. Search for Claude Report in the Extensions view, or install it from the Visual Studio Marketplace. Cursor, Antigravity and VSCodium use Open VSX. To install a release file instead, choose Extensions → Install from VSIX…, or run code --install-extension claude-report-<version>.vsix. Local builds are in dist/.
  2. Open Claude Report in the activity bar. Sessions shows Claude Code history. Metrics reads the last 30 days of activity from local transcripts.
  3. Optionally open Connect and install the Claude Code status line hook. This requires Node.js on your PATH. The hook keeps your previous status line and supports Disconnect.
  4. In Report, choose a month or custom date range, select your projects, choose Quick or Deep, and press Create report. Cloud generation requires the signed-in claude command. Local generation requires a model in LM Studio or Ollama.

Requirements

For You need
Sessions, Metrics, model and context in the status bar VS Code 1.75 or later, or a compatible editor, and Claude Code or Codex transcripts on this computer
Plan usage gauges and trend Node.js on your PATH, the Connect status line hook, and a Claude plan that reports usage windows (Pro, Max, Team or Ultra)
Cloud reports The claude command, signed in
Local reports LM Studio or Ollama with a model loaded
PDF export Microsoft Edge, Google Chrome or Chromium

The test suite has so far only been run on Windows. The macOS and Linux code paths exist but are unverified. The optional Outlook email draft works on Windows and macOS only.

Activity and usage

Surface Data and behavior
Sessions Lists Claude Code sessions, indicates active processes, and resumes a session in its original working directory.
Status bar Shows model and context from live Claude Code readings or transcript estimates. The five-hour and weekly plan gauges require the status line hook.
Metrics → Tokens Combines supported Claude Code and Codex logs. Select a source or filter by the open project. Includes input, output and cache usage, along with supported tool, model and activity breakdowns.
Metrics → Plan usage Shows the latest five-hour and weekly percentages and reset countdowns from the Claude Code status line. Burn-rate estimates use at least 15 minutes of readings from the current quota window.
Metrics → Trend Plots five-hour plan readings for the last 24 hours and weekly readings for the last seven days. Reuses saved Claude Control readings when available. Dashed markers indicate recorded reset times, and gaps show where readings are missing.

The activity scan reads ~/.claude/projects/**/*.jsonl and Codex rollouts under ~/.codex/sessions or $CODEX_HOME/sessions. It groups transcript working directories by their nearest Git root, or by the working directory itself if it has no repository. The report picker lists projects with eligible user prompts in the selected dates, regardless of which folder is open. File edits and commits alone do not create report eligibility. Git commits are supplementary evidence attached to selected projects.

The scan covers the named transcript sources. Native chats in other assistants, browser chat sessions, direct API usage, and standalone model-server chats are outside that coverage. Plan usage history comes from the status line and the local Claude Control usage file when present. Cache efficiency is calculated from transcript token counts.

Report workflow

  • Quick scans and redacts activity, then sends a bounded evidence sample per project. The model writes a short summary and an expanded card. The expanded card can include eight examples with separate details and evidence, plus how AI helped, friction, and next steps. A short draft can receive one expansion pass when enough evidence exists. The period overview is generated separately from the project drafts.
  • Deep adds an independent review for each project at the selected depth. Low includes the user's prompts. Medium also includes relevant filenames and root project documentation. High also includes selected changed lines. Cloud mode runs up to three project reviews at a time on Haiku, then Sonnet synthesizes their findings in the final report. A run reviews at one depth only and stays inside a request budget shown before it starts. The preflight plans around the reviewer context, batches oversized evidence by session, and reuses cached reviews with matching evidence.
  • Advice is a separate section of the report, apart from the achievements. It shows one strength, one thing to watch and one habit to try, each with where it was seen, based on how your prompts were phrased. It gives no scores, and under ten prompts it is labelled a first look. Its Learn more links come from a fixed list of official documentation pages, never from the model.
  • Progress shows List and Flow views, elapsed time, completion, and any failures. Open the finished report when ready. The HTML has a section nav, a ranked project list, a token chart that scales to the period or selected project, and System, Light and Dark themes. Examples are labelled Requested, Recorded or Verified.
  • Manage lets you open a report, edit or remove projects, export, enhance all or selected projects with another model, compare revisions, review runtime timings, and manage saved copies. Enhance normally reuses the saved evidence sample. Create a fresh report for the same dates to use the current scanner and available history.
  • AI time estimates are optional. Edit an example to record time yourself. A user-supplied value takes precedence where a task matches.

In Cloud, Quick defaults to Haiku at medium effort for project cards and Sonnet at high effort for the overview. Deep uses Haiku for project reviews and Sonnet for its final synthesis. In Local, the selected model handles both parts on this computer. See Reports and Local models.

An example monthly report with totals, an overview, a daily token chart and a ranked project list

A project row expanded to show how AI was used, examples with evidence and time, what did not work and next steps

The Advice section: a strength, a thing to watch and a habit to try

All screenshots use invented projects and numbers.

Saved reports and exported files

Saved runs live in ~/.claude-report/reports/<YYYY-MM>/runs/<timestamp>/. Each run includes the HTML and JSON plus optional private evidence and metadata used for enhancement and runtime logs. A revision creates a separate run.

An exported copy goes to Downloads/Reports/<Month_Year>/Report_MM-DD-YYYY_HH-mm-ss_name.html by default. A custom export folder becomes the parent of the month folders. Use Manage → Manage exported copies → Show in folder to locate a tracked copy. PDF export uses Edge or Chrome.

Privacy

The extension scans local files and has no telemetry. Cloud generation sends selected, redacted evidence through the configured Claude Code sign-in. Local generation sends it to the local model server. The extension filters environment-file prompts and redacts common secret patterns. Review exports before sharing. The optional status line hook updates Claude Code settings after Connect and restores the previous setting on Disconnect. See the privacy reference.

Development

The extension uses Node.js, worker threads and VS Code webviews. It has no runtime package dependencies or bundler. The extension host is extension.js. The sidebar and the self-contained report are in template/. Data collection, generation, rendering and storage are in src/. The model instructions are in skills/.

For report design work, start with template/report.html and use design/build-preview.js with DEMO=1 to preview synthetic data. The layout should keep the period overview, activity charts, project examples, and prompting review easy to scan separately. The same HTML renders in the editor, the browser, and PDF export.

Change Start here
Sessions, metrics, project identity src/sessions.js, src/scan-worker.js, src/scan.js, src/codex-activity.js
Status line and plan history bin/hook.js, src/live.js, template/panel/metrics.js
Report generation and models src/generate.js, src/overview.js, src/claude.js, src/local-report.js, skills/
Deep evidence, scheduling and cache src/deep-git.js, src/deep-review.js, src/deep-stages.js
Saved reports, editing and exports src/report-store.js, src/report-management.js, src/report-editor.js, src/export.js
Report UI and progress template/report.html, src/render.js, template/panel/shell.html
node --test
npm.cmd run package

Use npm run package on shells that allow the npm script. The build writes dist/claude-report-<version>.vsix.

Architecture · Troubleshooting · Contributing · Releasing · Changelog

The Metrics layout is adapted from Claude Control by Lucas do Nordeste. Claude Report is released under the MIT license.

  • Contact us
  • Jobs
  • Privacy
  • Manage cookies
  • Terms of use
  • Trademarks
  • Your Privacy Choices
  • Consumer Health Privacy
© 2026 Microsoft