Simplify Claude

Claude Code finishes a session and hands you a wall of jargon. This turns it into a short, plain-English summary with one click: what you asked for, what changed, what to check, what's unfinished.

Works in VS Code and Cursor. Uses OpenAI (cheapest model by default) or Anthropic.

How you use it

After a Claude Code session, click Simplify session in the status bar (bottom-right). A side panel opens with the plain-English summary. That's it. It never types into your chat and never competes with the claude prompt in your terminal.

You can also right-click any selected text and choose Simplify selected text to plain-English just that.

How it works

Claude Code saves every session as a transcript in ~/.claude/projects/. The extension reads the latest transcript for your current workspace, sends the conversation (your asks + Claude's messages + a one-line note per tool call) to your chosen AI, and shows the result. Because it reads the whole session, the summary has real context; it is not a reword of the last message.

First-time setup

On install you'll get a prompt to set up. Or run the command Simplify Claude: Choose provider and set API key anytime. Pick OpenAI (recommended, cheapest) or Anthropic, then paste your key.

• OpenAI keys start with sk-, from platform.openai.com/api-keys
• Anthropic keys start with sk-ant-, from console.anthropic.com

Settings

• simplifyClaude.provider — openai (default) or anthropic.
• simplifyClaude.openaiModel — default gpt-4.1-nano (cheapest). gpt-5.6-luna for richer output.
• simplifyClaude.anthropicModel — default claude-haiku-4-5.
• simplifyClaude.maxTranscriptChars — how much of the session is sent (default 60,000).
• simplifyClaude.extraInstructions — e.g. "explain in Roman Urdu".
• simplifyClaude.showStatusBarButton — hide the button if you prefer the Command Palette.

Security

• Keys live in your OS keychain via VS Code SecretStorage. Never in settings.json, never in plain text, never logged.
• Transcript text goes only to your chosen provider's official API. No middle server, no telemetry.
• File contents Claude read during the session are stripped before sending. Only your messages, Claude's replies, and one-line tool notes leave your machine.
• Zero runtime dependencies, so no npm supply-chain risk.
• The results panel escapes all content, so nothing in a transcript can inject code into the view.

Note

This reads Claude Code transcripts. Typing the command name into a running claude chat does nothing useful; use the status bar button or Command Palette instead.