deep-context

To analyze large files without exhausting token limits, progressive reading lists the file structure and reads only relevant chunks, while specialist agents run parallel architecture and dependency analysis to synthesize a coherent mental model.

The best way to map circular dependencies and dead code is using dependency analysis, which evaluates imports and structural impacts across the codebase without requiring full file reads to identify potential dead or circular dependencies.

Architectural analysis on large files over 50KB is performed by progressive reading, which lists the file structure and retrieves only relevant chunks to reduce token usage, enabling specialist agents to execute parallel pattern analysis.

Yes, codebase synthesis automatically persists architectural insights across sessions. It combines capsule context, progressive reading, and dependency results into a coherent mental model with automatic persistence for continuous layered understanding.

Yes, you can run database and code-pattern analyses in parallel. Specialist agents execute these analyses concurrently to deliver faster, fresher insights while building deep codebase understanding and mapping imports.

Staged context buildup is needed when you require architectural clarity for large codebases or unfamiliar subsystems. It uses capsule context to inject summaries and recent files, orchestrating progressive reading to guide decisions.