rubik 0.1.1 → 0.1.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 13f2d00bb2c72a92d48741df2cb3c3e84fc970a9ce3e4e844722afa58ef59599
-  data.tar.gz: f3314db467118bbd56aa79625f954f5c9e76611c56b5eb5e467d442fa440aa63
+  metadata.gz: a0ec56cafd854c2e2999c68cdfdf787322ca4d7d2a2866fac4b4e8ce9f1efac7
+  data.tar.gz: ec584d8ac5ee2039423d71968cce852afb65fcb072cbdbb87325ef94693e6a5d
 SHA512:
-  metadata.gz: a5c29f5af7119d6b9189e37d39390c6dd2463caa1fd870e49022b92a0ed72481b227616000a84a2e201f26c7a65009f7fcb05ac6591db6ca1b9ecae3f81bc6d4
-  data.tar.gz: 3ab1188c4607f7dbcbe4011111723f530555052497154eaea6c8d47b09dc927bfcfa43b50b88869f2dc8bc1e402c7589a569d59ad584d29b991bfff765292d08
+  metadata.gz: 3a535d5c0f45c74363387e7df47da8c87b1296be897fdd09adc17e4adfcc576b48555066cb0b3618d27f1eeeba052c53fe1ccb7a670598a43cad57bb8a4cc942
+  data.tar.gz: d75bb114c57fb2221f17f9d89a1a0fe1811b9e36ba52289337030b822210ef5373118e7d8438c4000da9dfca65fb4dadf3b5f85bdb8ad117ee50d56469c1a782

data/ATTRIBUTION.md

@@ -54,6 +54,15 @@ SOFTWARE.
   - **Spatial Composability:** Reactive Coeffect Dependency DAG ($\Sigma \vdash C : \tau$) implemented in [`Rubik::Runtime::ComponentFiber`](lib/rubik/runtime/fiber.rb) and [`Rubik::Runtime::Manager`](lib/rubik/runtime/manager.rb).
   - Zero-downtime hot reloading, atomic rollback on activation failure, and leak-free lifecycle management.
 
+### SKILL.state (Structured Execution State)
+- **Paper:** *"SKILL.state: Scalable Long-Horizon Agent Skills"* ([arXiv:2608.26263](https://arxiv.org/abs/2608.26263))
+- **Authors:** Sanket Badhe, Priyanka Tiwari, Jonghyun Chung
+- **Rubik Implementation:**
+  - The structured goal ledger ([`orchestrator/goal_ledger/`](lib/rubik/orchestrator/goal_ledger/), 2026-10-10): the `!goal` loop replaces append-only conversation memory with an explicit, mutable, rev-guarded execution state Σ (objective, tasks, findings, decisions, open questions, environment notes) — the model sees the current state, not an ever-growing history.
+  - Validated JSON-patch intake (RFC 6902 subset) with the `:syntax` / `:schema` / `:conflict` rejection taxonomy and bounded-retry model feedback (`GoalLedger::Patch` / `Schema` / `Validator`).
+  - The paper's world-drift recovery, coded as tri-state FS re-verification: declarative `verify` assertions are re-checked at step boundaries; a failed assertion flips the task to `stale` with an internal rev bump, so the `:conflict` the model then hits IS the drift signal (`GoalLedger::Verifier` / `VerifySpec` / `DriftNote`).
+  - Observation projection: banked tool results compact into per-row `<absorbed_into_ledger>` tombstones while the last K turns stay raw (`GoalLedger::ObservationProjection`).
+
 ---
 
 ## 🌳 3. AST Parsing & Multi-Language Ecosystem
@@ -112,7 +121,7 @@ SOFTWARE.
 | **`strings-ansi`** | Piotr Murach | MIT | ANSI-aware string truncation and alignment for REPL formatters. |
 | **`reline`** | Hiroshi SHIBATA & Ruby Team | Ruby License / BSD-2-Clause | Line-editing, history navigation, autocompletion hooks, and headless terminal adaptation. |
 | **`prism`** | Kevin Newton & Ruby Core Team | MIT | High-performance Ruby parsing and AST extraction ([`Rubik::AST::RubyExtractor`](lib/rubik/ast/ruby_extractor.rb)). |
-| **`minitest`** | Ryan Davis & seattlerb | MIT | Fast, modular test suite powering Rubik's 1,690+ unit, integration, and contract tests. |
+| **`minitest`** | Ryan Davis & seattlerb | MIT | Fast, modular test suite powering Rubik's 3,400+ unit, integration, and contract tests. |
 | **`rubocop`** | Bozhidar Batsov & contributors | MIT | Static code analysis, style enforcement, and architectural hygiene gates. |
 | **`benchmark`, `logger`, `pstore`, `reline`, `fiddle`** | Ruby core / rubygems teams | Ruby License / BSD-2-Clause | Ruby stdlib gems pinned explicitly in `rubik.gemspec` as they are unbundled from the default set in Ruby 3.4/3.5. |
 

data/CHANGELOG.md

@@ -10,8 +10,37 @@ The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
 Nothing yet.
 
+## [0.1.2] - 2026-10-11
+
+The Anthropic correctness release: the provider now speaks the modern
+extended-thinking contract, so pinning any current Claude model works
+instead of failing with HTTP 400.
+
+### Fixed
+
+- **Anthropic provider speaks the modern thinking contract:** the
+  default model pin moves from the stale `claude-3-7-sonnet-20250219`
+  to `claude-sonnet-5-5`, and the extended-thinking payload follows the
+  model line ? 4.6+ and the 5.x/Fable models get the adaptive form
+  (`thinking: {type: "adaptive"}` with an `output_config` effort),
+  3-7-sonnet through the 4.5 line keep the `budget_tokens` math, and
+  pre-thinking lines get nothing attached. 0.1.1 sent the legacy form
+  to every Anthropic model, which Claude 4.7+ rejects with HTTP 400.
+
+### Changed
+
+- Extended thinking auto-enables on every thinking-capable Anthropic
+  model line (previously only claude-3-7-sonnet); `thinking: false`
+  still disables it everywhere and `thinking_budget_ratio` shapes the
+  legacy form only. Requests that carry thinking still force
+  `temperature` to 1.0.
+- README quick start documents the native Anthropic provider
+  (`ANTHROPIC_API_KEY` + `!model anthropic/<model>`) next to the
+  OpenAI-compatible example.
+
 ## [0.1.1] - 2026-10-11
 
+
 The Linux surface release: CI now proves Linux on Ruby 3.4 and 4.0
 (`linux-test` is a required check), and the host seams that surface
 exposed are fixed. Ships the structured goal ledger, the default-on

data/README.md

@@ -11,7 +11,7 @@ English | [Türkçe](README.tr.md)
 [![Protocol](https://img.shields.io/badge/protocol-AGENTS.md-blue.svg)](AGENTS.md)
 [![Commands Guide](https://img.shields.io/badge/docs-COMMANDS.md-yellow.svg)](COMMANDS.md)
 [![Config Guide](https://img.shields.io/badge/docs-CONFIG.md-green.svg)](CONFIG.md)
-[![Test Suite](https://img.shields.io/badge/tests-3202%20runs%20%7C%20100%25%20passed-brightgreen.svg)]()
+[![Test Suite](https://img.shields.io/badge/tests-3403%20runs%20%7C%20100%25%20passed-brightgreen.svg)]()
 
 ---
 
@@ -31,6 +31,12 @@ Requires **Ruby 3.4+**:
 gem install rubik
 ```
 
+The installer puts the `rubik` launcher into RubyGems' bin directory (already on your `PATH`), so you can start it from any project directory:
+
+```bash
+rubik
+```
+
 or with Bundler:
 
 ```ruby
@@ -73,6 +79,9 @@ ln -sf $(pwd)/bin/rubik ~/.local/bin/rubik
 rubik
 ```
 
+> [!NOTE]
+> **The gem and a clone can coexist.** On POSIX the symlink always runs your clone: `bin/rubik` resolves `lib/` from its own real path, so a `git pull` is live on the very next launch — even while the `rubik` gem stays installed. The gem-installed `rubik` keeps serving the last released version until you run `gem update rubik`. On Windows, whichever `bin` folder comes first in `PATH` wins — `where.exe rubik` shows the winner.
+
 #### 🪟 Windows (PowerShell / Command Prompt)
 
 Rubik ships with built-in `.cmd`, `.bat`, and `.ps1` wrappers for Windows:
@@ -105,6 +114,12 @@ rubik > !provider add deepseek openai_compatible https://api.deepseek.com/v1/cha
 rubik > !model deepseek/deepseek-coder
 ```
 
+Anthropic talks over its **native Messages API** (not an OpenAI shim): the `anthropic` provider is registered out of the box — set `ANTHROPIC_API_KEY` and pick a model:
+
+```bash
+rubik > !model anthropic/claude-sonnet-5-5
+```
+
 ### First Session Example
 
 ```text

data/lib/rubik/llm/anthropic.rb

@@ -9,7 +9,32 @@ module Rubik
     class AnthropicProvider < BaseProvider
       attr_reader :name, :model, :formatter, :headers, :api_key, :api_key_env
 
-      def initialize(name: nil, model: "claude-3-7-sonnet-20250219", formatter: AnthropicFormatter.new, headers: {},
+      # Extended-thinking support by model line (Anthropic docs, 2026-10):
+      #   :adaptive — 4.6+ and the 5.x/Fable lines: `thinking: {type: "adaptive"}`
+      #     with an `output_config` effort; the legacy budget_tokens form is
+      #     REJECTED with a 400 from 4.7 onward.
+      #   :legacy   — claude-3-7-sonnet through the 4.5 line: the manual
+      #     `thinking: {type: "enabled", budget_tokens: N}` form is the only
+      #     accepted mode.
+      #   nil       — 3.5-and-earlier lines accept no thinking at all, and
+      #     unrecognized names (custom aliases/proxies) get no auto thinking.
+      def self.thinking_mode_for(model_name)
+        s = model_name.to_s
+        return :legacy if s.include?("claude-3-7-sonnet")
+        return nil if s.match?(/\Aclaude-[23]/) # number-first 2.x/3.x lines (3-7 handled above)
+
+        # Single-digit captures: Anthropic versions are one digit per segment,
+        # so a snapshot date suffix (claude-sonnet-4-20250514) can never be
+        # swallowed as the minor version.
+        version = s.match(/\Aclaude-(?:sonnet|opus|haiku|fable)-(\d)(?:-(\d))?/)
+        return nil unless version
+
+        major = version[1].to_i
+        minor = version[2].to_i
+        major > 4 || (major == 4 && minor >= 6) ? :adaptive : :legacy
+      end
+
+      def initialize(name: nil, model: "claude-sonnet-5-5", formatter: AnthropicFormatter.new, headers: {},
                      api_key: nil, api_key_env: nil)
         super()
         @name = name.to_s.downcase
@@ -170,14 +195,25 @@ module Rubik
         thinking_opt = opts[:thinking]
         thinking_disabled = [false, :disabled].include?(thinking_opt)
         ratio = (opts[:thinking_budget_ratio] || 0.5).to_f
+        # Thinking is on for every model line that supports it unless
+        # explicitly disabled; thinking_budget_ratio only shapes the legacy
+        # budget form — adaptive models reason as they see fit.
+        mode = self.class.thinking_mode_for(target_model)
 
         if thinking_opt.is_a?(Hash)
           payload[:thinking] = thinking_opt
-        elsif !thinking_disabled && (thinking_opt == true || target_model.to_s.include?("claude-3-7-sonnet"))
-          if max_tokens > 2048
-            budget = [(max_tokens * ratio).to_i, 1024].max
-            budget = [budget, max_tokens - 1024].min
-            payload[:thinking] = { type: "enabled", budget_tokens: budget } if budget >= 1024
+        elsif !thinking_disabled && (thinking_opt == true || mode)
+          if mode == :legacy
+            if max_tokens > 2048
+              budget = [(max_tokens * ratio).to_i, 1024].max
+              budget = [budget, max_tokens - 1024].min
+              payload[:thinking] = { type: "enabled", budget_tokens: budget } if budget >= 1024
+            end
+          else
+            # Adaptive models reject the legacy form outright, and for
+            # unrecognized names adaptive is the modern-generation default.
+            payload[:thinking] = { type: "adaptive" }
+            payload[:output_config] = { effort: "high" }
           end
         end
 

data/lib/rubik/llm/registry.rb

@@ -38,7 +38,7 @@ module Rubik
         @providers = {
           "ollama" => OllamaProvider.new(model: "qwen2.5-coder:7b", headers: global_headers),
           "openai" => OpenAIProvider.new(model: "gpt-4o", headers: global_headers),
-          "anthropic" => AnthropicProvider.new(model: "claude-3-7-sonnet-20250219", headers: global_headers)
+          "anthropic" => AnthropicProvider.new(model: "claude-sonnet-5-5", headers: global_headers)
         }
         @models_catalog = {}
         load_from_config(config) if config
@@ -82,7 +82,7 @@ module Rubik
         end
         if @providers["anthropic"].instance_of?(AnthropicProvider)
           @providers["anthropic"] = AnthropicProvider.new(
-            model: @providers["anthropic"]&.model || "claude-3-7-sonnet-20250219",
+            model: @providers["anthropic"]&.model || "claude-sonnet-5-5",
             headers: global_headers
           )
         end
@@ -132,7 +132,7 @@ module Rubik
                             when "anthropic"
                               AnthropicProvider.new(
                                 name: p_name,
-                                model: default_model || "claude-3-7-sonnet-20250219",
+                                model: default_model || "claude-sonnet-5-5",
                                 # Honor the same api_key / api_key_env that
                                 # openai-compatible specs already forward;
                                 # otherwise Anthropic auth fails silently.
@@ -257,7 +257,7 @@ module Rubik
         when "anthropic"
           anthropic_headers = (custom.respond_to?(:headers) ? custom.headers : nil) || global_headers
           AnthropicProvider.new(
-            model: model_name || "claude-3-7-sonnet-20250219",
+            model: model_name || "claude-sonnet-5-5",
             headers: anthropic_headers,
             api_key: custom.respond_to?(:api_key) ? custom.api_key : nil,
             api_key_env: custom.respond_to?(:api_key_env) ? custom.api_key_env : nil

data/lib/rubik/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Rubik
-  VERSION = "0.1.1"
+  VERSION = "0.1.2"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: rubik
 version: !ruby/object:Gem::Version
-  version: 0.1.1
+  version: 0.1.2
 platform: ruby
 authors:
 - Özgür Ulukır