Session Memory & Summarization

Session Memory & Summarization Skill

Provides automatic conversation continuity across sessions by loading recent memory at session start and generating AI summaries during compaction.

What It Does

• Memory Loading: Automatically injects the latest memory file into new sessions as a system message
• AI Summarization: Generates concise summaries when approaching token limits, written to daily memory files
• Seamless Experience: No manual intervention required — just natural conversation flow

Hooks

session:start

Runs when a new session begins. Loads the most recent .md file from the memory/ directory and adds it to the session context as a system message. Content is truncated to ~4000 characters to stay within context limits.

session:compact:before

Runs before automatic compaction when token usage exceeds thresholds. If conditions are met (20+ messages OR 60% of token limit used), generates an AI summary using agent.generateSummary() and prepends it to today's memory file with a timestamp.

Installation

clawhub install session-context

Or manually:

cd ~/.openclaw/workspace/skills
git clone https://github.com/thomasmarcel/openclaw-skill-session-context.git session-context
openclaw skills enable session-context

Requirements

• OpenClaw ≥ 0.29.0
• Workspace with memory/ directory (created automatically)
• Access to agent's LLM for summarization

Configuration

Customize thresholds in hooks/session/compact:before/handler.js:

return (
  msgCount >= 20 || // minimum messages
  tokenCount > maxTokens * 0.6 // trigger at 60% of limit
);

Adjust truncation in hooks/session/start/handler.js:

const truncated = content.substring(0, 4000); // change as needed

Memory Structure

memory/
  2025-04-03.md  # daily files, newest entries at top
  2025-04-04.md

Each file contains timestamped summaries throughout the day. No global MEMORY.md is required, though you can modify the loader to use it if preferred.

How It Works

1. During a conversation: As token usage grows, OpenClaw monitors session size.
2. Before compaction: The session:compact:before hook checks thresholds. If met, it calls the agent's LLM to create a summary, then writes it to memory/YYYY-MM-DD.md.
3. Compaction proceeds: Older messages are pruned, keeping the summary and recent context.
4. Next session: The session:start hook loads the newest memory file, giving immediate continuity.

License

MIT