Architecture

CompressKit sits above the algorithm cores with a single thin buffer layer:

caller
  -> buffer layer (encode_buffer / decode_buffer, BufferTransform fn pointer)
  -> algorithm core (Huffman / Arithmetic / Range / RLE)

Why It Exists

• file-to-file CLIs and in-memory callers share the same buffer-layer path
• a single BufferTransform function-pointer alias normalizes every algorithm behind one signature
• shared size-limit checks and error codes keep the surface small

Language Host

• C++: algorithms/shared/cpp

Security Boundary

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

Verification

make test runs the shared lifecycle tests followed by the algorithm-specific suites.