Release 0.4.0

Added

• DisplayResolver — sparse binding.yaml overlay that resolves surface-facing alias, description, guidance, tags, and documentation into metadata["display"]. Supports file/directory binding paths, pre-parsed data, MCP alias sanitization, and CLI alias validation.
• SyntaxVerifier — implements Verifier trait, checks .rs files parse without syntax errors via the syn crate.
• deep_resolve_refs() — now public API (was internal). Recursively resolves all $ref pointers in OpenAPI schemas, depth-limited to 16 levels.
• resolve_target() — validates and parses module_path:qualname target strings. Returns ResolvedTarget with module_path and qualname fields.

Fixed

• README: apcore dependency version updated from >= 0.13.0 to >= 0.14 (matches Cargo.toml).
• docs/API_CONTRACT.md: Scanner trait updated to async with generic App parameter; RegistryWriter new() and with_handler_factory() constructors added; AIEnhancer::new return type corrected to AIEnhancerError.

Release 0.3.1

Changed

• Rebrand: aipartnerup → aiperceivable

Release 0.3.0

Initial release. Rust port of apcore-toolkit-python v0.3.0.

Added

• ScannedModule struct — canonical representation of a scanned endpoint,
  with all 12 fields matching the Python dataclass (serde Serialize/Deserialize)
• Scanner async trait — generic over App type parameter for framework-specific
  scanning (e.g., Scanner<axum::Router>, Scanner<actix_web::App>).
  Uses #[async_trait] for async scan() method
• filter_modules() — regex-based include/exclude filtering with Result error
  handling for invalid patterns
• deduplicate_ids() — resolves duplicate module IDs by appending _2, _3, etc.
• infer_annotations_from_method() — HTTP method to ModuleAnnotations mapping
  (GET → readonly+cacheable, DELETE → destructive, PUT → idempotent)
• YAMLWriter — generates .binding.yaml files for apcore::BindingLoader with
  filename sanitization, path traversal protection, and optional verification
• RegistryWriter — registers modules directly into apcore::Registry with
  pluggable HandlerFactory for target resolution. Falls back to passthrough
  handler for schema-only registration
• HTTPProxyRegistryWriter — registers scanned modules as HTTP proxy modules
  that forward requests to a running web API. Supports path parameter substitution,
  pluggable auth headers via Arc, and JSON error extraction.
  Feature-gated behind http-proxy (reqwest dependency)
• OutputFormat enum + get_writer() factory — maps format strings to writer
  variants ("yaml", "registry", "http-proxy")
• Output verification system:
  • Verifier trait — pluggable verification protocol
  • YAMLVerifier — validates YAML syntax and required binding fields
  • JSONVerifier — validates JSON syntax
  • MagicBytesVerifier — validates file header magic bytes
  • RegistryVerifier — validates module is retrievable from registry
  • run_verifier_chain() — sequential verifier execution with catch_unwind
    panic safety (matches Python's try/except behavior)
• WriteResult / VerifyResult — structured result types for all writer and
  verifier operations (serde Serialize/Deserialize)
• WriteError — typed error via thiserror for file I/O failures
• OpenAPI utilities:
  • resolve_ref() — JSON $ref pointer resolution
  • resolve_schema() — conditional $ref resolution
  • extract_input_schema() — merges query/path parameters and request body
  • extract_output_schema() — extracts 200/201 response schema
  • deep_resolve_refs() (internal) — recursive $ref resolution for nested
    schemas (allOf/anyOf/oneOf, items, properties), depth-limited to 16
• enrich_schema_descriptions() — merges parameter descriptions into JSON Schema
  properties with optional overwrite mode
• annotations_to_value() / module_to_value() / modules_to_values() —
  serialization utilities with tracing::warn! on serialization failures
• to_markdown() — generic JSON-to-Markdown conversion with depth control,
  table heuristics, field/exclude filtering, and UTF-8 safe truncation
• AIEnhancer — SLM-based metadata enhancement using local OpenAI-compatible
  APIs (Ollama, vLLM, LM Studio) via ureq. Fills missing descriptions,
  documentation, behavioral annotations (all 11 fields), and input schemas.
  All AI-generated fields tagged with x-generated-by: slm for auditability.
  Configuration via environment variables (APCORE_AI_*) or constructor params
• Enhancer trait — pluggable interface for metadata enhancement
• AIEnhancerError — typed error enum via thiserror with Config, Connection,
  Response variants
• HandlerFactory / HandlerFn type aliases — enable framework adapters to
  provide real async handlers for registered modules

Intentionally Omitted (Python-specific)

• PythonWriter — generates Python source code; not applicable to Rust consumers
• SyntaxVerifier — validates Python AST syntax; not applicable
• flatten_pydantic_params() — Pydantic-specific; Rust has no equivalent
• resolve_target() — Python importlib dynamic import; Rust uses HandlerFactory

Dependencies

• apcore (path dep to ../apcore-rust) >= 0.13.0
• serde + serde_json + serde_yaml — serialization
• regex — pattern matching
• chrono — timestamps
• tracing — structured logging
• thiserror — domain error types
• async-trait — async trait support
• tokio — required transitively by apcore
• ureq — synchronous HTTP for AI enhancer
• reqwest (optional, http-proxy feature) — async HTTP for proxy writer

Tests

• 176 unit tests + 1 doc-test, all passing
• clippy clean (0 warnings with -D warnings)
• All quality gates: cargo fmt, cargo clippy, cargo build, cargo test