context-compression

Benchmark and implement context compression strategies for long-running AI agent sessions.

Updated Nov 16, 2025

Search Tags:#agents #memory-management #structured-summaries #evaluation-framework #token-economy #context-compression

Ready to run

npx skills add https://github.com/mhintz1980/ptl-lova --skill context-compression-mhintz1980

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: context-compression
Source: https://github.com/mhintz1980/ptl-lova/tree/main/docs/agent-skills/skills/context-compression
Command: npx skills add https://github.com/mhintz1980/ptl-lova --skill context-compression-mhintz1980

SYSTEM DOCUMENTATION & REQUIREMENTS

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

What problem does it solve?

Long-running agent sessions generate millions of tokens of history, making it difficult to maintain context within fixed memory and token budgets. This skill defines disciplined context compression strategies that balance token savings with information preservation.

Core Features & Use Cases

Anchored Iterative Summarization: maintain structured, persistent summaries with explicit sections for session intent, file modifications, decisions, and next steps.
Opaque Compression: produce compact representations optimized for high compression ratios, sacrificing some interpretability.
Regenerative Full Summary: regenerate detailed structured summaries at compression boundaries to preserve signal across cycles.
Evaluation Framework: provides a probe-based evaluation workflow to assess recall, artifact tracking, continuation, and decision integrity.

Quick Start

Summarize the latest session span into anchored, structured summary focusing on intent, modified files, decisions, and next steps.

Frequently Asked Questions about context-compression

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

FAQPage Schema

How do I compress context for long-running agent sessions without losing critical details?▼

Compress context for long-running agent sessions by applying anchored iterative summarization to maintain structured, persistent summaries with explicit sections for session intent, modified files, decisions, and next steps. This approach balances token savings with essential information preservation.

What is the best way to manage token budgets in memory-constrained debugging workflows?▼

Manage token budgets in memory-constrained debugging workflows by using opaque compression to produce compact representations optimized for high compression ratios, sacrificing some interpretability to preserve signal across cycles within tight memory limits.

How does regenerative summarization preserve context across compression boundaries?▼

Regenerative summarization preserves context across compression boundaries by regenerating detailed structured summaries at compression boundaries, ensuring that critical session details, file modifications, and decisions are maintained throughout long-running agent tasks.

Can I evaluate context compression quality for AI agent code-generation sessions?▼

Evaluate context compression quality for AI agent code-generation sessions by using a probe-based evaluation framework that assesses recall, artifact tracking, continuation, and decision integrity within the compression pipeline.

Does context compression work for memory-constrained conversations with tight token budgets?▼

Context compression works for memory-constrained conversations with tight token budgets by applying disciplined compression strategies that support anchored iterative summarization, opaque compression, and regenerative summaries within a framework that tracks files, decisions, and tasks.

What are the limitations of opaque compression for structured summaries?▼

Opaque compression for structured summaries produces compact representations optimized for high compression ratios, but sacrifices some interpretability compared to anchored iterative summarization, which maintains explicit sections for session intent, file modifications, decisions, and next steps.