claw-compactor

Compresses workspace memory and transcripts to reduce token costs.

2.2k|207|Updated Feb 10, 2026

Search Tags:#markdown #ai #memory #pipeline #compression #tokens #codebook

Ready to run

npx skills add https://github.com/aeromomo/claw-compactor --skill claw-compactor

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: claw-compactor
Source: https://github.com/aeromomo/claw-compactor/tree/main
Command: npx skills add https://github.com/aeromomo/claw-compactor --skill claw-compactor

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires tiktoken, and includes scripts (resource) and references (resource) components.

What problem does it solve?

Claw Compactor reduces token usage by compressing memory and transcripts using five layered techniques, enabling faster AI workflows.

Core Features & Use Cases

5-layer compression pipeline including rule-based preprocessing, dictionary encoding, observation-based summaries, and tiered context loading
Lossless compression for memory content; selective lossy compression for transcripts while preserving decisions and actions
Automatic generation of MEMORY.md tiered summaries (L0/L1/L2) and codebooks for efficient context loading
On-demand references and scripts for auditing, troubleshooting, and extending the pipeline

Quick Start

Point this AI at a workspace and run the full pipeline to compress memory and transcripts.

Frequently Asked Questions about claw-compactor

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I compress markdown memory files to reduce token usage in AI workflows?▼

To compress markdown memory files and reduce token usage, this skill applies a 5-layer pipeline that generates tiered MEMORY.md summaries and a dictionary codebook for efficient context loading. It enforces deterministic, auditable transformations.

What is the best way to compress session transcripts while preserving decisions and actions?▼

The best way to compress session transcripts is using selective lossy compression that preserves decisions and actions. This skill reduces transcript tokens by applying dictionary encoding and observation-based summaries alongside memory compression.

Does claw-compactor require tiktoken to run the memory compression pipeline?▼

Yes, the memory compression pipeline requires the tiktoken dependency to process markdown memory files and transcripts. It uses this to generate tiered L0/L1/L2 summaries and a codebook for efficient context loading.

Can I audit and reverse the compression steps applied to my workspace memory?▼

You can audit and reverse compression steps because the pipeline enforces deterministic, auditable transformations with reversible compression. It provides on-demand scripts and references for troubleshooting, auditing, and extending the pipeline.

How do I generate tiered MEMORY.md summaries and a codebook for AI context loading?▼

You generate tiered MEMORY.md summaries and a dictionary codebook by pointing the AI at a workspace and running the full compression pipeline. It automatically creates L0/L1/L2 summaries for efficient context loading.