Module semantic_chunks 

Semantic Chunking with Attention Bridges.

Groups content into semantic chunks (function bodies, import blocks, type definitions) rather than treating lines independently. Orders chunks for optimal LLM attention flow:

1. Most relevant chunk FIRST (high-attention position)
2. Its immediate dependencies (imports, types it uses) adjacent
3. Supporting context in the middle
4. Tail anchor: brief reference back to the primary chunk (attention bridge)

This exploits how transformer attention actually works: local coherence + global anchors beats scattered high-importance lines.

Structs

SemanticChunk

Enums

ChunkKind

Functions

detect_chunks

Detect semantic boundaries in content and group lines into chunks.

order_for_attention

Score chunks by task relevance and reorder for optimal attention flow.

render_with_bridges

Render chunks back to text with attention bridges.