ct-toolkit 0.3.1__tar.gz → 0.3.2__tar.gz

{ct_toolkit-0.3.1 → ct_toolkit-0.3.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ct-toolkit
-Version: 0.3.1
+Version: 0.3.2
 Summary: Computational Theseus Toolkit — Identity Continuity Guardrails for Agentic Systems
 Author: Hakan Damar
 License: Apache-2.0
@@ -51,6 +51,7 @@ CT Toolkit is an open-source security layer designed to preserve the **identity
 ---
 
 ## 📖 Official Documentation
+
 For full API reference, architecture details, examples, and integration guides, visit our documentation site:
 👉 [**https://hakandamar.github.io/ct-toolkit/**](https://hakandamar.github.io/ct-toolkit/)
 
@@ -63,6 +64,7 @@ For full API reference, architecture details, examples, and integration guides,
 In complex agentic workflows, LLMs tend to "drift" from their original instructions. CT Toolkit provides the mathematical and cryptographic guardrails to ensure your agents remain aligned with their core constitution, even across deep hierarchies.
 
 - **Constitutional Kernels**: Axiomatic identity anchors.
+- **Autonomous Self-Correction**: Active L2->L1 feedback loop that retries and corrects divergent responses before they reach the user.
 - **Divergence Engine**: Multi-tiered drift analysis (L1/L2/L3).
 - **Hierarchical Propagation**: Mother-to-child constraint inheritance.
 - **Provenance Log**: Immutable HMAC-signed interaction history.
@@ -92,15 +94,16 @@ print(f"Divergence Score: {response.divergence_score}")
 
 ## 🚦 Project Health & Status
 
-| Metric | Status |
-| :--- | :--- |
-| **Tests** | ✅ 218/218 passing (93% coverage) |
-| **Last Phase** | ✅ Phase 4: OS Support, Deep Agents & Alerts (Complete) |
-| **Current Goal** | 🔶 Phase 5: Vault and Security Infrastructure |
+| Metric           | Status                                                  |
+| :--------------- | :------------------------------------------------------ |
+| **Tests**        | ✅ 231/232 passing (90% coverage)                       |
+| **Last Phase**   | ✅ Phase 4: OS Support, Deep Agents & Alerts (Complete) |
+| **Current Goal** | 🔶 Phase 5: Vault & Security (In Progress)              |
 
 For a detailed breakdown of the 8-phase roadmap, see [**PROJECT_STATUS.md**](docs/PROJECT_STATUS.md).
 
 ### Framework & Model Support
+
 Seamlessly integrate with your favorite frameworks and local models:
 
 - **Local Models**: Support for LM Studio, Ollama, and local Qwen/Llama endpoints.
@@ -111,9 +114,11 @@ Seamlessly integrate with your favorite frameworks and local models:
 ---
 
 ## Theoretical Foundation
+
 Translating the framework proposed in [**The Computational Theseus (2025)**](https://hakandamar.com/the-computational-theseus-engineering-identity-continuity-as-a-guardrail-against-sequential-963918c1720d) into engineering practice.
 
 ---
 
 ## License
+
 Apache License 2.0.

{ct_toolkit-0.3.1 → ct_toolkit-0.3.2}/README.md

@@ -12,6 +12,7 @@ CT Toolkit is an open-source security layer designed to preserve the **identity
 ---
 
 ## 📖 Official Documentation
+
 For full API reference, architecture details, examples, and integration guides, visit our documentation site:
 👉 [**https://hakandamar.github.io/ct-toolkit/**](https://hakandamar.github.io/ct-toolkit/)
 
@@ -24,6 +25,7 @@ For full API reference, architecture details, examples, and integration guides,
 In complex agentic workflows, LLMs tend to "drift" from their original instructions. CT Toolkit provides the mathematical and cryptographic guardrails to ensure your agents remain aligned with their core constitution, even across deep hierarchies.
 
 - **Constitutional Kernels**: Axiomatic identity anchors.
+- **Autonomous Self-Correction**: Active L2->L1 feedback loop that retries and corrects divergent responses before they reach the user.
 - **Divergence Engine**: Multi-tiered drift analysis (L1/L2/L3).
 - **Hierarchical Propagation**: Mother-to-child constraint inheritance.
 - **Provenance Log**: Immutable HMAC-signed interaction history.
@@ -53,15 +55,16 @@ print(f"Divergence Score: {response.divergence_score}")
 
 ## 🚦 Project Health & Status
 
-| Metric | Status |
-| :--- | :--- |
-| **Tests** | ✅ 218/218 passing (93% coverage) |
-| **Last Phase** | ✅ Phase 4: OS Support, Deep Agents & Alerts (Complete) |
-| **Current Goal** | 🔶 Phase 5: Vault and Security Infrastructure |
+| Metric           | Status                                                  |
+| :--------------- | :------------------------------------------------------ |
+| **Tests**        | ✅ 231/232 passing (90% coverage)                       |
+| **Last Phase**   | ✅ Phase 4: OS Support, Deep Agents & Alerts (Complete) |
+| **Current Goal** | 🔶 Phase 5: Vault & Security (In Progress)              |
 
 For a detailed breakdown of the 8-phase roadmap, see [**PROJECT_STATUS.md**](docs/PROJECT_STATUS.md).
 
 ### Framework & Model Support
+
 Seamlessly integrate with your favorite frameworks and local models:
 
 - **Local Models**: Support for LM Studio, Ollama, and local Qwen/Llama endpoints.
@@ -72,9 +75,11 @@ Seamlessly integrate with your favorite frameworks and local models:
 ---
 
 ## Theoretical Foundation
+
 Translating the framework proposed in [**The Computational Theseus (2025)**](https://hakandamar.com/the-computational-theseus-engineering-identity-continuity-as-a-guardrail-against-sequential-963918c1720d) into engineering practice.
 
 ---
 
 ## License
+
 Apache License 2.0.

{ct_toolkit-0.3.1 → ct_toolkit-0.3.2}/ct_toolkit/core/wrapper.py

@@ -59,6 +59,10 @@ class WrapperConfig:
     embedding_model: str = "text-embedding-3-small"
     enterprise_mode: bool = False    # Run all tiers all the time
     parent_kernel: ConstitutionalKernel | None = None  # Propagated from mother agent
+    
+    # -- Auto-Correction Loop --
+    auto_correction: bool = False
+    max_correction_retries: int = 1
 
     # -- Dynamic Stability-Plasticity Scheduling --
     elasticity_max_thresholds: tuple[float, float, float] | None = None  # (max_l1, max_l2, max_l3)
@@ -385,34 +389,52 @@ class TheseusWrapper:
                 f"Please provide a client or set the appropriate environment variables."
             )
 
-        start_time = time.monotonic()
-
-        try:
-            raw_response = self._call_provider(messages, model=model, **kwargs)
-        except Exception as e:
-            logger.error(f"Provider API error: {e}")
-            raise
-
-        elapsed = time.monotonic() - start_time
-        content = self._extract_content(raw_response)
-        model_used = self._extract_model(raw_response, model)
-        self._last_model = model_used
-
-        # Compute current interaction experience
-        interaction_count = 0
-        if self._config.log_requests:
-            interaction_count = self._provenance_log.get_interaction_count(
-                template=self._config.template,
-                kernel_name=self._kernel.name,
-                model=model_used
+        retries = 0
+        max_retries = self._config.max_correction_retries if self._config.auto_correction else 0
+        
+        while retries <= max_retries:
+            start_time = time.monotonic()
+    
+            try:
+                raw_response = self._call_provider(messages, model=model, **kwargs)
+            except Exception as e:
+                logger.error(f"Provider API error: {e}")
+                raise
+    
+            elapsed = time.monotonic() - start_time
+            content = self._extract_content(raw_response)
+            model_used = self._extract_model(raw_response, model)
+            self._last_model = model_used
+    
+            # Compute current interaction experience
+            interaction_count = 0
+            if self._config.log_requests:
+                interaction_count = self._provenance_log.get_interaction_count(
+                    template=self._config.template,
+                    kernel_name=self._kernel.name,
+                    model=model_used
+                )
+    
+            # Divergence Engine (L1 -> L2 -> L3)
+            skip_l3 = self._config.auto_correction and retries < max_retries
+            div_result = self._run_divergence_engine(
+                message=message,
+                response=content,
+                interaction_count=interaction_count,
+                skip_l3=skip_l3,
             )
-
-        # Divergence Engine (L1 -> L2 -> L3)
-        div_result = self._run_divergence_engine(
-            message=message,
-            response=content,
-            interaction_count=interaction_count,
-        )
+            
+            from ct_toolkit.divergence.l2_judge import JudgeVerdict
+            if getattr(div_result, 'l2_verdict', None) == JudgeVerdict.MISALIGNED and retries < max_retries:
+                logger.warning(f"Auto-correction triggered. Reason: {div_result.l2_reason}")
+                messages.extend([
+                    {"role": "assistant", "content": content},
+                    {"role": "user", "content": f"Identity Drift Detected. Your previous response violated the core identity instructions. Reason: {div_result.l2_reason}. Please revise your response to strictly adhere to the system prompt."}
+                ])
+                retries += 1
+                continue
+                
+            break
 
         # Provenance Log record
         provenance_id = None
@@ -582,10 +604,10 @@ class TheseusWrapper:
             return raw.model
         return fallback or "unknown"
 
-    def _run_divergence_engine(self, message: str, response: str, interaction_count: int = 0) -> Any:
+    def _run_divergence_engine(self, message: str, response: str, interaction_count: int = 0, skip_l3: bool = False) -> Any:
         """Runs the staged divergence engine (L1 -> L2 -> L3)."""
         try:
-            return self._divergence_engine.analyze(message, response, interaction_count)
+            return self._divergence_engine.analyze(message, response, interaction_count, skip_l3=skip_l3)
         except Exception as e:
             logger.error(f"Divergence Engine execution failed: {e}")
             from ct_toolkit.divergence.engine import DivergenceResult, DivergenceTier

{ct_toolkit-0.3.1 → ct_toolkit-0.3.2}/ct_toolkit/divergence/engine.py

@@ -154,6 +154,7 @@ class DivergenceEngine:
         request_text: str,
         response_text: str,
         interaction_count: int = 0,
+        skip_l3: bool = False,
     ) -> DivergenceResult:
         """
         Performs full divergence analysis.
@@ -166,8 +167,8 @@ class DivergenceEngine:
             )
 
         if self._enterprise:
-            return self._enterprise_analyze(request_text, response_text)
-        return self._standard_analyze(request_text, response_text)
+            return self._enterprise_analyze(request_text, response_text, skip_l3=skip_l3)
+        return self._standard_analyze(request_text, response_text, skip_l3=skip_l3)
 
     def get_drift_report(self, window_size: int = 50, model: str | None = None) -> DriftReport | None:
         """
@@ -189,6 +190,7 @@ class DivergenceEngine:
         self,
         request_text: str,
         response_text: str,
+        skip_l3: bool = False,
     ) -> DivergenceResult:
         """Progressive analysis: L1 → L2 if needed → L3 if needed."""
 
@@ -236,6 +238,18 @@ class DivergenceEngine:
             )
 
         # L3 — L2 problematic or L1 too high
+        if skip_l3:
+            return DivergenceResult(
+                tier=DivergenceTier.L2_JUDGE,
+                l1_score=l1_score,
+                l2_verdict=l2_result.verdict,
+                l2_confidence=l2_result.confidence,
+                l2_reason=l2_result.reason,
+                action_required=True,
+                cascade_blocked=True,
+                summary="L3 skipped due to auto-correction retry logic",
+            )
+
         if not self._l3_available:
             return DivergenceResult(
                 tier=DivergenceTier.L2_JUDGE,
@@ -269,6 +283,7 @@ class DivergenceEngine:
         self,
         request_text: str,
         response_text: str,
+        skip_l3: bool = False,
     ) -> DivergenceResult:
         """Enterprise mode: L1 + L2 + L3 all run, decision based on total score."""
         l1_score = self._run_l1(response_text)
@@ -278,7 +293,7 @@ class DivergenceEngine:
             l2_result = self._run_l2(request_text, response_text)
 
         l3_report = None
-        if self._l3_available:
+        if self._l3_available and not skip_l3:
             l3_report = self._run_l3(max_probes=5)  # Fast subset in Enterprise
 
         # Calculate total risk score

{ct_toolkit-0.3.1 → ct_toolkit-0.3.2}/ct_toolkit.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ct-toolkit
-Version: 0.3.1
+Version: 0.3.2
 Summary: Computational Theseus Toolkit — Identity Continuity Guardrails for Agentic Systems
 Author: Hakan Damar
 License: Apache-2.0
@@ -51,6 +51,7 @@ CT Toolkit is an open-source security layer designed to preserve the **identity
 ---
 
 ## 📖 Official Documentation
+
 For full API reference, architecture details, examples, and integration guides, visit our documentation site:
 👉 [**https://hakandamar.github.io/ct-toolkit/**](https://hakandamar.github.io/ct-toolkit/)
 
@@ -63,6 +64,7 @@ For full API reference, architecture details, examples, and integration guides,
 In complex agentic workflows, LLMs tend to "drift" from their original instructions. CT Toolkit provides the mathematical and cryptographic guardrails to ensure your agents remain aligned with their core constitution, even across deep hierarchies.
 
 - **Constitutional Kernels**: Axiomatic identity anchors.
+- **Autonomous Self-Correction**: Active L2->L1 feedback loop that retries and corrects divergent responses before they reach the user.
 - **Divergence Engine**: Multi-tiered drift analysis (L1/L2/L3).
 - **Hierarchical Propagation**: Mother-to-child constraint inheritance.
 - **Provenance Log**: Immutable HMAC-signed interaction history.
@@ -92,15 +94,16 @@ print(f"Divergence Score: {response.divergence_score}")
 
 ## 🚦 Project Health & Status
 
-| Metric | Status |
-| :--- | :--- |
-| **Tests** | ✅ 218/218 passing (93% coverage) |
-| **Last Phase** | ✅ Phase 4: OS Support, Deep Agents & Alerts (Complete) |
-| **Current Goal** | 🔶 Phase 5: Vault and Security Infrastructure |
+| Metric           | Status                                                  |
+| :--------------- | :------------------------------------------------------ |
+| **Tests**        | ✅ 231/232 passing (90% coverage)                       |
+| **Last Phase**   | ✅ Phase 4: OS Support, Deep Agents & Alerts (Complete) |
+| **Current Goal** | 🔶 Phase 5: Vault & Security (In Progress)              |
 
 For a detailed breakdown of the 8-phase roadmap, see [**PROJECT_STATUS.md**](docs/PROJECT_STATUS.md).
 
 ### Framework & Model Support
+
 Seamlessly integrate with your favorite frameworks and local models:
 
 - **Local Models**: Support for LM Studio, Ollama, and local Qwen/Llama endpoints.
@@ -111,9 +114,11 @@ Seamlessly integrate with your favorite frameworks and local models:
 ---
 
 ## Theoretical Foundation
+
 Translating the framework proposed in [**The Computational Theseus (2025)**](https://hakandamar.com/the-computational-theseus-engineering-identity-continuity-as-a-guardrail-against-sequential-963918c1720d) into engineering practice.
 
 ---
 
 ## License
+
 Apache License 2.0.

{ct_toolkit-0.3.1 → ct_toolkit-0.3.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "ct-toolkit"
-version = "0.3.1"
+version = "0.3.2"
 description = "Computational Theseus Toolkit — Identity Continuity Guardrails for Agentic Systems"
 readme = "README.md"
 requires-python = ">=3.11"