Context Window Manager

by Timoranjes

Maximize AI agent performance with proactive token management, intelligent compaction, and session handoff strategies.

Free

10 installsSecurity scanned

Works with the AI tools you already use

CCCGG+17 more

CCCGGVWM+14 more

See it in action

You say

My agent is starting to repeat itself and I'm at 75% context usage. Help me clean this up without losing our progress on the refactor.

Your agent does

I'll initiate a structured compaction. First, I'll update .session-state.md with our current progress. Then, I'll run: /compact "Preserve: refactor logic in /src/auth, failed test patterns, and active TODOS. Discard: tool logs and previous file versions." This will free up ~40% of your window.

About this skill

Stop Context Overflow and Agent Brain-Drain

As AI coding agents like Claude Code and Cursor tackle larger projects, they inevitably hit context limits. When an agent's context window exceeds 60-70%, reasoning quality drops, instructions are forgotten, and "hallucinations" increase. This skill provides a professional-grade framework for managing token efficiency and session longevity.

What it does

The Context Window Manager implements a phased workflow to proactively manage the lifecycle of an AI development session. It includes strategies for:

Session Anchoring: Creating lightweight project maps (CLAUDE.md/AGENTS.md) that define architecture and conventions without wasting tokens.
Intelligent Compaction: Using a decision tree to summarize conversations at critical thresholds (60%, 80%, and 90% usage) while preserving architectural decisions.
Selective File Loading: Shifting from full-file indexing to targeted line-range loading and interface-first hierarchy.
State Handoffs: Automated protocols for migrating progress to a fresh session when the current one is saturated.

Why use this skill

Prompting an AI to "summarize our chat" often results in the loss of subtle code constraints or edge cases. This skill provides specific preservation rules and command structures that ensure your agent remains sharp and accurate even during 4-hour refactoring sessions. It is the difference between an agent that completes the task and one that slowly breaks your codebase as it forgets the rules.