douglac

Compact Chat — AI skill for Claude Code

AI community

Structured handoff documents to transfer context between AI chat sessions.

How to install Compact Chat

This entry records only its repository, not the path inside it, so there is no exact command to give. Open douglac/compact-chat and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Compact Chat does

Structured handoff documents to transfer context between AI chat sessions. Cursor / Claude Code / Codex skill.

Alternatives in AI

Cursor Talk To Figma MCP — TalkToFigma: MCP integration between AI Agent (Cursor, Claude Code, Codex) and Figma, allowing Agentic AI to c 7k ★
Debate — AI Debate Hub - Structured four-way debates between Claude, Sonnet, Gemini, and Codex 2.8k ★
Project Butler — Project memory system for AI coding assistants (Claude Code, Cursor, Codex): session logs, project wiki, rules 372 ★

README

compact-chat

**Every token your AI re-reads is money you're burning.**

Long Cursor and Claude Code sessions accumulate massive context. The model re-processes every previous message on each turn — and you pay for every token. A 60-message session doesn't cost 60x the first message. It costs **1 + 2 + 3 + ... + 60 = 1,830x**. That's quadratic cost growth for linear work.

At some point the context window fills up, the model starts hallucinating, and you open a new chat — only to spend 15 minutes (and more tokens) re-explaining everything.

`compact-chat` compresses an entire session into a structured handoff document. Paste it into a fresh chat and **resume with full context at a fraction of the token cost**.

The Problem

Message #1:    50 tokens    →  you pay 50
Message #10:   50 tokens    →  you pay 500  (model re-reads all 10)
Message #30:   50 tokens    →  you pay 1,500
Message #60:   50 tokens    →  you pay 3,000

Total session: ~91,500 tokens consumed for 3,000 tokens of actual content

Eventually the context fills, quality drops, and you start over — paying again to re-establish what the model already knew.

The Fix

You:    "compact"
Agent:  *distills 60 messages into ~800 tokens of structured context*
You:    *opens new chat, pastes the compact*
Agent:  *resumes instantly — fresh context window, full knowledge, minimal tokens*

**One compact replaces thousands of tokens of stale conversation.** Your next session starts lean and accurate instead of bloated and confused.

Why This Matters

Metric	Without compact-chat	With compact-chat
Context after 60 messages	~90k tokens (full history)	~800 tokens (structured summary)
Resume cost in new chat	~5k tokens re-explaining	~800 tokens (paste compact)
Model accuracy	Degrades as context grows	Fresh — peak accuracy
Architecture decisions	Buried in conversation noise	Explicit table with rati