Architecture

CompressKit now uses a two-layer in-memory integration model above the existing algorithm cores:

caller
  -> buffer layer (encode_buffer / decode_buffer)
  -> streaming layer (process / flush / finish / reset)
  -> existing algorithm core

Why It Exists

• file-to-file CLIs now share the same buffer-layer path as in-memory callers
• lifecycle semantics are consistent across C++17, Go, and Rust
• shared error handling makes later frame-format and conformance work easier to build on

Language Hosts

• C++: algorithms/shared/cpp
• Go: algorithms/shared/go
• Rust: algorithms/shared/rust

Security Boundary

• input accepted by the streaming layer is capped at 4 GiB
• decoded output produced by the streaming layer is capped at 1 GiB

Verification

make test includes shared streaming lifecycle tests for all three languages before the algorithm-specific suites.