Release 0.3.0

Added

• _deep_resolve_refs() — recursive $ref resolution for nested OpenAPI schemas,
  handling allOf/anyOf/oneOf, items, and properties. Depth-limited to 16
  levels to prevent infinite recursion on circular references.
• Enhancer protocol — pluggable interface for metadata enhancement, allowing
  custom enhancers beyond the built-in AIEnhancer.
• HTTPProxyRegistryWriter — registers scanned modules as HTTP proxy classes
  that forward requests to a running web API. Supports path parameter substitution,
  pluggable auth headers, and 2xx success range (with 204 returning {}).
  Requires optional httpx dependency (pip install apcore-toolkit[http-proxy]).
• get_writer("http-proxy", base_url=...) — factory support for the new
  HTTP proxy writer with **kwargs forwarding.
• Expanded __init__.py public API: exports Enhancer, HTTPProxyRegistryWriter,
  WriteError, Verifier, VerifyResult, verifier classes, serializer functions,
  resolve_ref, resolve_schema, extract_input_schema, extract_output_schema,
  and run_verifier_chain.

Fixed

• extract_output_schema() — now recursively resolves all nested $ref pointers
  (previously only handled the shallow case of array items with $ref).
• extract_input_schema() — now recursively resolves $ref inside individual
  properties after assembly.
• get_writer() return type annotation now includes HTTPProxyRegistryWriter.

Tests

• 272 tests (up from 260), all passing
• Added TestDeepResolveRefs (8 tests): top-level ref, nested properties,
  allOf/anyOf, array items, deeply nested refs, circular ref depth limit,
  immutability guarantee
• Added nested $ref tests for extract_input_schema and extract_output_schema
• Added test_http_proxy for get_writer("http-proxy") factory

---

Release 0.2.0

Added

• WriteResult dataclass — structured result type for all writer operations,
  with optional output verification support
• AIEnhancer — SLM-based metadata enhancement using local OpenAI-compatible
  APIs (Ollama, vLLM, LM Studio). Fills missing descriptions, infers behavioral
  annotations, and generates input schemas for untyped functions. All AI-generated
  fields are tagged with x-generated-by: slm for auditability.
  Annotation inference prompt and acceptance logic extended to handle all 11
  annotation fields (5 new: cacheable, cache_ttl, cache_key_fields,
  paginated, pagination_style)
• Output verification via verify=True parameter on all writers:
• YAMLWriter: validates YAML parsability and required binding fields
• PythonWriter: validates Python syntax via ast.parse()
• RegistryWriter: validates module is retrievable after registration
• CI matrix expanded to Python 3.11 + 3.12, added mypy and coverage reporting

Changed

• apcore >= 0.13.0 — Upgraded minimum dependency to support new
  ModuleAnnotations caching and pagination fields:
  cacheable, cache_ttl, cache_key_fields, paginated, pagination_style
• infer_annotations_from_method() — GET now also infers cacheable=True
• BREAKING: All writers now return list[WriteResult] instead of
  list[dict] (YAMLWriter) or list[str] (PythonWriter, RegistryWriter).
  Downstream adapters (django-apcore, flask-apcore) must update code that
  accesses writer return values as dicts or strings. Migration:
• result[0]["bindings"] → use writer._build_binding(module) for dict access
• result == ["module.id"] → [r.module_id for r in result] == ["module.id"]
• f"wrote: {item}" → f"wrote: {item.module_id}"
• Downstream packages should pin apcore-toolkit<0.2.0 until updated

Fixed

• README — added 5 concrete verifier classes (YAMLVerifier, SyntaxVerifier,
  RegistryVerifier, MagicBytesVerifier, JSONVerifier) to Core Modules table
  for documentation completeness

Tests

• 197 tests (up from 150), all passing

---

Release 0.1.0

Initial release. Extracts shared framework-agnostic logic from django-apcore
and flask-apcore into a standalone toolkit package.

Added

• ScannedModule dataclass — canonical representation of a scanned endpoint
• BaseScanner ABC with filter_modules(), deduplicate_ids(),
  infer_annotations_from_method(), and extract_docstring() utilities
• YAMLWriter — generates .binding.yaml files for apcore.BindingLoader
• PythonWriter — generates @module-decorated Python wrapper files
• RegistryWriter — registers modules directly into apcore.Registry
• to_markdown() — generic dict-to-Markdown conversion with depth control
  and table heuristics
• flatten_pydantic_params() — flattens Pydantic model parameters into
  scalar kwargs for MCP tool invocation
• resolve_target() — resolves module.path:qualname target strings
• enrich_schema_descriptions() — merges docstring parameter descriptions
  into JSON Schema properties
• annotations_to_dict() / module_to_dict() — serialization utilities
• OpenAPI utilities: resolve_ref(), resolve_schema(),
  extract_input_schema(), extract_output_schema()
• Output format factory via get_writer()
• 150 tests with 94% code coverage

Dependencies

• apcore >= 0.9.0
• pydantic >= 2.0
• PyYAML >= 6.0