olyx-guardrails 1.0.0 → 1.1.1

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: '069f552d05ca4f0ee61bf1496b27da5961477bb75eb422424b1cd19b31e1ac61'
-  data.tar.gz: 9189f069fc540a3ecab429ebec194cb9c870917966837fb4c7fff428d705ed78
+  metadata.gz: 789eb5e71a301345c3fd0d28e59ffee0b0d9cf67ee9686f0912e5a743ccd093b
+  data.tar.gz: d89fc57b450761fe6e530fa116e39413eeb17edeb9c73023fbd8247ace98a727
 SHA512:
-  metadata.gz: 264d94b6a842c268cd30430ecf1cb6f2d49557d12266f331e594160e7100d8539b1c35351e4dbab58dd0525c97760c738a4f139270aefcb8d2f251835bd04e5c
-  data.tar.gz: 75bd129939354ed2333d622cc215cc4a34b0967e39fcb482345719ddd312f9637ceb6a13beddfd27cbfa242f6057bd82178590e8ab98e63e10d79a461fe99b6d
+  metadata.gz: 17390bf8cbd4961a4e4a669a5ca0b8f80afcf2663fceaea85e5fe31050e789252b5952ef8063ef80d8ffaa5b78f7e69d4a4c96dba8ce1724fcfe090ccb5f3e50
+  data.tar.gz: d3e28269e0760c9c09c83f25f8499b74391ee1d7ebf5d3d9ab1a4fe9236d56c2f90b63daa6c603068f6a40fa2398577d08a2f8631d1c05c37e01a478d9ec3a2a

data/CHANGELOG.md

@@ -3,6 +3,40 @@
 All notable changes to olyx-guardrails are documented here.
 Format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
+## [1.1.1] - 2026-07-24
+
+### Changed
+
+- The published gem now contains only runtime code, generator templates, and
+  its README, changelog, and license. Guides and runnable examples remain
+  available from the canonical repository.
+- Removed two internal, one-use indirection layers from Rails policy-path
+  normalization and policy redaction. Public behavior is unchanged.
+- The package manifest is now tested as an exact consumer-facing contract, and
+  path-like Rails policy configuration has direct boundary coverage.
+
+## [1.1.0] - 2026-07-23
+
+### Changed
+
+- Rails metadata now validates consistently as a Hash at every facade and
+  adapter boundary.
+- Blocking exceptions use boundary-neutral wording, Active Job selectors
+  validate when declared, and inherited job declarations remain effective.
+- Notifier misuse raises `ArgumentError`; `nil` now means only that a valid
+  decision has zero risk.
+- The Rails generator provides copyable policy customization instructions and
+  tests the customized YAML path.
+- Contributor setup, complete local quality validation, and maintainer release
+  verification have canonical commands and documentation.
+- Closely coupled proxy objects were folded into their owning runtime,
+  configuration, result-building, and notification components.
+- Provider risk scores now require a finite Numeric value; numeric strings are
+  ignored instead of being coerced across the untrusted provider boundary.
+- Contributor and Rails appraisal locks use the current `net-imap` patch.
+- CI refreshes the Ruby advisory database and blocks known-vulnerable locked
+  dependencies without making the offline local quality gate network-dependent.
+
 ## [1.0.0] - 2026-07-21
 
 Initial public release.
@@ -13,7 +47,7 @@ Initial public release.
   prompt-injection, secret, and restricted-policy findings without
   transforming the input. Returns `allowed`, per-category detection flags,
   `risk_score`, and a `checks` array with one entry per category. The
-  `length` check runs first; content and AI checks are skipped (not just
+  `length` check runs first; content and LLM checks are skipped (not just
   failed) when it does, so an oversized payload never pays their cost.
 - `Olyx::Guardrails.redact` — transforms text by removing every
   regex-detected PII, secret, and restricted-policy match, without making an
@@ -33,9 +67,9 @@ Initial public release.
   monitoring-only restricted-content rules, `substring`/`whole_word`/`regexp`
   term matching, case-aware regex support, configurable safe replacements,
   and `Policy.from_h` for YAML/JSON-style Hash loading.
-- `ai_failure_mode:` controls how an analyzer failure is handled: `:allow`
-  (default, keeps the deterministic result), `:block` (adds a failed `ai`
-  check and rejects), or `:raise` (`AiAnalyzerError`).
+- `llm_failure_mode:` controls how a provider failure is handled: `:allow`
+  (default, keeps the deterministic result), `:block` (adds a failed `llm`
+  check and rejects), or `:raise` (`LlmProviderError`).
 - Invalid configuration — duplicate rule names, empty-matching or invalid
   patterns, malformed message arrays, invalid custom regexes — raises
   `ArgumentError` at construction time rather than silently degrading into a
@@ -61,31 +95,29 @@ Initial public release.
   span. Custom patterns extend detection and are classified as secrets; use
   `PolicyRule` for named business restrictions instead.
 
-### AI analyzer hook
+### LLM provider hook
 
-- Optional `ai_analyzer:` callable receives `(text, context)` and returns a
+- Optional `llm_provider:` callable receives `(text, context)` and returns a
   Hash or a schema-model object implementing `deep_to_h`/`to_h`. Findings
   can add a violation but cannot clear a deterministic one. Boolean fields
   must be actual `true`/`false` values; non-finite risk scores are ignored
   rather than crashing. Exceptions and malformed responses are bounded under
-  `result[:ai_analysis][:error]` and handled per `ai_failure_mode`.
-- Optional `Integrations::OpenAIAnalyzer` connector for the official OpenAI
-  Ruby SDK: sends a strict `OpenAI::BaseModel` schema through the Responses
-  API, consumes the parsed schema-model result, defaults to `store: false`,
-  and has no model allowlist — it forwards any String/Symbol model
-  identifier and relies on Structured Outputs support as the compatibility
-  check. Refusals, API failures, and missing parsed output degrade into the
-  same `result[:ai_analysis][:error]` path. The `openai` gem is optional and
-  loaded lazily.
+  `result[:llm_analysis][:error]` and handled per `llm_failure_mode`.
+- Provider transport is application-owned. The gem has no vendor SDK adapters,
+  endpoint assumptions, model allowlists, or model-specific parameters.
+  A framework-free local HTTP example demonstrates an open-source-first
+  classifier boundary using only Ruby standard-library HTTP and JSON support.
 
 ### Notifications
 
 - Vendor-neutral `Notifier` dispatches one sanitized, deeply frozen,
   versioned event to named callable handlers. Handlers run synchronously and
   independently — one failure doesn't stop the others, and errors are
-  returned per handler rather than raised. Input previews, AI reasons,
+  returned per handler rather than raised. Input previews, LLM reasons,
   metadata keys/values, and handler errors are bounded and redacted with the
   policy's restricted-content rules and the built-in PII/secret detectors.
+- Notification events expose optional provider reasoning as `llm_reason`;
+  the pre-release `ai_reason` field is not retained.
 
 ### Rails integration
 
@@ -103,15 +135,23 @@ Initial public release.
 - `Rails::ActiveJobHandler` enqueues sanitized notification events through
   any Active Job backend, resolving a reload-safe String/Symbol job
   constant at delivery time.
+- The initial compatibility window covers Rails 8.0 and 8.1. Only listed Rails
+  series are tested and supported; subsequent gem releases may remove a series
+  after its upstream security support ends.
 
 ### Quality and security posture
 
-- No runtime dependencies for the core checks; Rails and the OpenAI SDK are
-  both optional and loaded only when used.
-- CI enforces RuboCop, a Flog structural-complexity gate (max 10 per
-  ordinary method, 60 per class/module, with documented DSL exemptions
-  only), and a RubyCritic maintainability gate, alongside the Appraisal
-  matrix across Rails 7.2, 8.0, and 8.1.
+- The core has one lightweight runtime dependency, Ruby's `base64` bundled gem.
+  Rails remains optional and is loaded only when used.
+- CI enforces RuboCop — including calibrated structural-complexity cops
+  (`Metrics/AbcSize`, `CyclomaticComplexity`, `PerceivedComplexity`,
+  `ClassLength`) in place of a separate complexity tool — and a RubyCritic
+  maintainability gate, alongside the Appraisal matrix across Rails 8.0
+  and 8.1.
+- Native RDoc covers every supported public class, module, constant, attribute,
+  and method. CI blocks undocumented additions to the explicit public API
+  manifest while leaving implementation-only constants outside the
+  compatibility contract.
 - Least-privilege, SHA-pinned CI actions; CodeQL and OpenSSF Scorecard
   scanning; a private vulnerability-reporting process (see
-  [SECURITY.md](SECURITY.md)).
+  [security policy](https://github.com/Olyx-labs/olyx-guardrails/security/policy)).