IDLs + client generation (Codama / Shank / Kinobi)

Goal

Never hand-maintain multiple program clients by manually re-implementing serializers. Prefer an IDL-driven, code-generated workflow.

Codama (preferred)

• Use Codama as the "single program description format" to generate:
• TypeScript clients (including Kit-friendly output)
• Rust clients (when available/needed)
• documentation artifacts

Anchor → Codama

If the program is Anchor: 1) Produce Anchor IDL from the build 2) Convert Anchor IDL to Codama nodes (nodes-from-anchor) 3) Render a Kit-native TypeScript client (codama renderers)

Native Rust → Shank → Codama

If the program is native: 1) Use Shank macros to extract a Shank IDL from annotated Rust 2) Convert Shank IDL to Codama 3) Generate clients via Codama renderers

Repository structure recommendation

• programs/<name>/ (program source)
• idl/<name>.json (Anchor/Shank IDL)
• codama/<name>.json (Codama IDL)
• clients/ts/<name>/ (generated TS client)
• clients/rust/<name>/ (generated Rust client)

Generation guardrails

• Codegen outputs should be checked into git if:
• you need deterministic builds
• you want users to consume the client without running codegen
• Otherwise, keep codegen in CI and publish artifacts.

"Do not do this"

• Do not write IDLs by hand unless you have no alternative.
• Do not hand-write Borsh layouts for programs you own; use the IDL/codegen pipeline.