Context Manager

Overview

This agent is designed to combat the 'silent tax' of token wastage inherent in long AI coding sessions. Instead of dumping entire project contents into the prompt, Context Manager enforces disciplined context handling, ensuring that only the most relevant information reaches the LLM at any given time.

Capabilities

• Progressive Loading: It guides users to load context incrementally—starting with project structure (ls/glob), then specific function signatures, and finally loading full files only when necessary for editing.
• Context Hygiene Commands: Implements commands like /compact (to run when usage hits 60-70%), /clear, and /compact to actively clean up stale or irrelevant context between unrelated tasks.
• Path Referencing: Encourages referencing files using explicit paths (@file_path) rather than pasting raw content into the chat, maintaining cleaner prompts.

Example Use Cases

1. Starting a New Feature: Instead of pasting 5 related files, use the agent to first list the directory structure, then ask it to analyze only the UserService.ts file and its dependencies.
2. Switching Tasks: After debugging a networking module, if you switch to working on UI components, run /clear or start a new session block to prevent stale network context from confusing the model.
3. Large Refactoring: For massive refactors, use progressive loading: analyze signatures first, then load only the files marked for modification.