v0.6.0 — Intelligent Codegen & Auto-Generating Docs

Highlights

Intelligent Codegen Pipeline — The code generation engine has been rebuilt from the ground up. A type-driven inference engine replaces hard-coded lookup tables, and a new Code IR (intermediate representation) is now the sole emission path, replacing the old string-concatenation approach. This makes the codegen more maintainable, extensible, and correct.

Auto-Generating Documentation — Documentation is now a fully automated artifact of the build pipeline. README, API reference, cookbook recipes, architecture guides, and migration docs are all generated from source via just docs.

What's New

Codegen

• Type-driven inference engine: Aliases, field policies, extras, and parent references are now derived from ADK type metadata instead of hard-coded tables
• Code IR emission: All .py, .pyi, and test scaffold generation goes through a structured IR with emit_python() and emit_stub(), producing cleaner and more consistent output
• Seed resilience: delegate and all list-append extras now survive just seed regeneration via seed.manual.toml and enhanced infer_extras()
• Ruff-clean output: Generated code passes ruff check cleanly with isort-compatible import grouping and PEP 8 spacing

Documentation Infrastructure

• Dynamic README: README.md is auto-generated from README.template.md with live Mermaid architecture diagrams and verified quick-start code
• Semantic API reference: Methods are grouped by category (Core Configuration, Callbacks, Control Flow) with inline code examples extracted from docstrings
• Visual cookbooks: Each cookbook recipe executes in a sandbox to render its own architecture as a Mermaid DAG
• Concept harvesting: Architectural theory is compiled into architecture-and-concepts.md from raw specs
• Zero-drift pipeline: All generators wired into just docs — docs can never go stale

Features (since v0.5.2)

• Agent.context() for wiring context engineering transforms
• S.capture() to bridge conversation history to session state
• C module with core context engineering primitives
• Event visibility inference and VisibilityPlugin
• Contract checker with cross-channel coherence analysis
• Structured output DX with @ operator and .output_schema()
• .isolate() for specialist agents
• .delegate() for coordinator pattern

CI

• Codegen pipeline in CI now matches justfile pattern: ruff check --fix || true → ruff format → ruff check

Breaking Changes

None. All public APIs are backwards-compatible.

Stats

• 1263 tests passing
• 132 builders generated
• ~537 methods (aliases + callbacks + forwarded)
• 9 output modules
• 54 cookbook recipes with auto-generated Mermaid diagrams