@artemiskit/adapter-anthropic 0.1.10 → 0.1.12

package/CHANGELOG.md

@@ -1,5 +1,142 @@
 # @artemiskit/adapter-anthropic
 
+## 0.1.12
+
+### Patch Changes
+
+- ## v0.3.0 - SDK, Guardian Mode & OWASP Compliance
+
+  This major release delivers the full programmatic SDK, runtime protection with Guardian Mode, OWASP LLM Top 10 2025 attack vectors, and agentic framework adapters.
+
+  ### Programmatic SDK (`@artemiskit/sdk`)
+
+  The new SDK package provides a complete programmatic API for LLM evaluation:
+
+  - **ArtemisKit class** with `run()`, `redteam()`, and `stress()` methods
+  - **Jest integration** with custom matchers (`toPassAllCases`, `toHaveSuccessRate`, etc.)
+  - **Vitest integration** with identical matchers
+  - **Event handling** for real-time progress updates
+  - **13 custom matchers** for run, red team, and stress test assertions
+
+  ```typescript
+  import { ArtemisKit } from "@artemiskit/sdk";
+  import { jestMatchers } from "@artemiskit/sdk/jest";
+
+  expect.extend(jestMatchers);
+
+  const kit = new ArtemisKit({ provider: "openai", model: "gpt-4o" });
+  const results = await kit.run({ scenario: "./tests.yaml" });
+  expect(results).toPassAllCases();
+  ```
+
+  ### Guardian Mode (Runtime Protection)
+
+  New Guardian Mode provides runtime protection for AI/LLM applications:
+
+  - **Three operating modes**: `testing`, `guardian`, `hybrid`
+  - **Prompt injection detection** and blocking
+  - **PII detection & redaction** (email, SSN, phone, API keys)
+  - **Action validation** for agent tool/function calls
+  - **Intent classification** with risk assessment
+  - **Circuit breaker** for automatic blocking on repeated violations
+  - **Rate limiting** and **cost limiting**
+  - **Custom policies** via TypeScript or YAML
+
+  ```typescript
+  import { createGuardian } from "@artemiskit/sdk/guardian";
+
+  const guardian = createGuardian({ mode: "guardian", blockOnFailure: true });
+  const protectedClient = guardian.protect(myLLMClient);
+  ```
+
+  ### OWASP LLM Top 10 2025 Attack Vectors
+
+  New red team mutations aligned with OWASP LLM Top 10 2025:
+
+  | Mutation             | OWASP | Description                    |
+  | -------------------- | ----- | ------------------------------ |
+  | `bad-likert-judge`   | LLM01 | Exploit evaluation capability  |
+  | `crescendo`          | LLM01 | Multi-turn gradual escalation  |
+  | `deceptive-delight`  | LLM01 | Positive framing bypass        |
+  | `system-extraction`  | LLM07 | System prompt leakage          |
+  | `output-injection`   | LLM05 | XSS, SQLi in output            |
+  | `excessive-agency`   | LLM06 | Unauthorized action claims     |
+  | `hallucination-trap` | LLM09 | Confident fabrication triggers |
+
+  ```bash
+  akit redteam scenario.yaml --owasp LLM01,LLM05
+  akit redteam scenario.yaml --owasp-full
+  ```
+
+  ### Agentic Framework Adapters
+
+  New adapters for testing agentic AI systems:
+
+  **LangChain Adapter** (`@artemiskit/adapter-langchain`)
+
+  - Test chains, agents, and runnables
+  - Capture intermediate steps and tool usage
+  - Support for LCEL, ReAct agents, RAG chains
+
+  **DeepAgents Adapter** (`@artemiskit/adapter-deepagents`)
+
+  - Test multi-agent systems and workflows
+  - Capture agent traces and inter-agent messages
+  - Support for sequential, parallel, and hierarchical workflows
+
+  ```typescript
+  import { createLangChainAdapter } from "@artemiskit/adapter-langchain";
+  import { createDeepAgentsAdapter } from "@artemiskit/adapter-deepagents";
+
+  const adapter = createLangChainAdapter(myChain, {
+    captureIntermediateSteps: true,
+  });
+  const result = await adapter.generate({ prompt: "Test query" });
+  ```
+
+  ### Supabase Storage Enhancements
+
+  Enhanced cloud storage capabilities:
+
+  - **Analytics tables** for metrics tracking
+  - **Case results table** for granular analysis
+  - **Baseline management** for regression detection
+  - **Trend analysis** queries
+
+  ### Bug Fixes
+
+  - **adapter-openai**: Use `max_completion_tokens` for newer OpenAI models (o1, o3, gpt-4.5)
+  - **redteam**: Resolve TypeScript and flaky test issues in OWASP mutations
+  - **adapters**: Fix TypeScript build errors for agentic adapters
+  - **core**: Add `langchain` and `deepagents` to ProviderType union
+
+  ### Examples
+
+  New comprehensive examples organized by feature:
+
+  - `examples/guardian/` - Guardian Mode examples (testing, guardian, hybrid modes)
+  - `examples/sdk/` - SDK usage examples (Jest, Vitest, events)
+  - `examples/adapters/` - Agentic adapter examples
+  - `examples/owasp/` - OWASP LLM Top 10 test scenarios
+
+  ### Documentation
+
+  - Complete SDK documentation with API reference
+  - Guardian Mode guide with all three modes explained
+  - Agentic adapters documentation (LangChain, DeepAgents)
+  - Test matchers reference for Jest/Vitest
+  - OWASP LLM Top 10 testing scenarios
+
+- Updated dependencies
+  - @artemiskit/core@0.3.0
+
+## 0.1.11
+
+### Patch Changes
+
+- Updated dependencies [16604a6]
+  - @artemiskit/core@0.2.4
+
 ## 0.1.10
 
 ### Patch Changes

package/dist/index.js

@@ -1961,12 +1961,12 @@ class PromptCachingBetaMessageStream {
     }
     __classPrivateFieldGet2(this, _PromptCachingBetaMessageStream_instances, "m", _PromptCachingBetaMessageStream_endRequest).call(this);
   }
-  [(_PromptCachingBetaMessageStream_currentMessageSnapshot = new WeakMap, _PromptCachingBetaMessageStream_connectedPromise = new WeakMap, _PromptCachingBetaMessageStream_resolveConnectedPromise = new WeakMap, _PromptCachingBetaMessageStream_rejectConnectedPromise = new WeakMap, _PromptCachingBetaMessageStream_endPromise = new WeakMap, _PromptCachingBetaMessageStream_resolveEndPromise = new WeakMap, _PromptCachingBetaMessageStream_rejectEndPromise = new WeakMap, _PromptCachingBetaMessageStream_listeners = new WeakMap, _PromptCachingBetaMessageStream_ended = new WeakMap, _PromptCachingBetaMessageStream_errored = new WeakMap, _PromptCachingBetaMessageStream_aborted = new WeakMap, _PromptCachingBetaMessageStream_catchingPromiseCreated = new WeakMap, _PromptCachingBetaMessageStream_handleError = new WeakMap, _PromptCachingBetaMessageStream_instances = new WeakSet, _PromptCachingBetaMessageStream_getFinalMessage = function _PromptCachingBetaMessageStream_getFinalMessage() {
+  [(_PromptCachingBetaMessageStream_currentMessageSnapshot = new WeakMap, _PromptCachingBetaMessageStream_connectedPromise = new WeakMap, _PromptCachingBetaMessageStream_resolveConnectedPromise = new WeakMap, _PromptCachingBetaMessageStream_rejectConnectedPromise = new WeakMap, _PromptCachingBetaMessageStream_endPromise = new WeakMap, _PromptCachingBetaMessageStream_resolveEndPromise = new WeakMap, _PromptCachingBetaMessageStream_rejectEndPromise = new WeakMap, _PromptCachingBetaMessageStream_listeners = new WeakMap, _PromptCachingBetaMessageStream_ended = new WeakMap, _PromptCachingBetaMessageStream_errored = new WeakMap, _PromptCachingBetaMessageStream_aborted = new WeakMap, _PromptCachingBetaMessageStream_catchingPromiseCreated = new WeakMap, _PromptCachingBetaMessageStream_handleError = new WeakMap, _PromptCachingBetaMessageStream_instances = new WeakSet, _PromptCachingBetaMessageStream_getFinalMessage = function _PromptCachingBetaMessageStream_getFinalMessage2() {
     if (this.receivedMessages.length === 0) {
       throw new AnthropicError("stream ended without producing a PromptCachingBetaMessage with role=assistant");
     }
     return this.receivedMessages.at(-1);
-  }, _PromptCachingBetaMessageStream_getFinalText = function _PromptCachingBetaMessageStream_getFinalText() {
+  }, _PromptCachingBetaMessageStream_getFinalText = function _PromptCachingBetaMessageStream_getFinalText2() {
     if (this.receivedMessages.length === 0) {
       throw new AnthropicError("stream ended without producing a PromptCachingBetaMessage with role=assistant");
     }
@@ -1975,11 +1975,11 @@ class PromptCachingBetaMessageStream {
       throw new AnthropicError("stream ended without producing a content block with type=text");
     }
     return textBlocks.join(" ");
-  }, _PromptCachingBetaMessageStream_beginRequest = function _PromptCachingBetaMessageStream_beginRequest() {
+  }, _PromptCachingBetaMessageStream_beginRequest = function _PromptCachingBetaMessageStream_beginRequest2() {
     if (this.ended)
       return;
     __classPrivateFieldSet2(this, _PromptCachingBetaMessageStream_currentMessageSnapshot, undefined, "f");
-  }, _PromptCachingBetaMessageStream_addStreamEvent = function _PromptCachingBetaMessageStream_addStreamEvent(event) {
+  }, _PromptCachingBetaMessageStream_addStreamEvent = function _PromptCachingBetaMessageStream_addStreamEvent2(event) {
     if (this.ended)
       return;
     const messageSnapshot = __classPrivateFieldGet2(this, _PromptCachingBetaMessageStream_instances, "m", _PromptCachingBetaMessageStream_accumulateMessage).call(this, event);
@@ -2013,7 +2013,7 @@ class PromptCachingBetaMessageStream {
       case "message_delta":
         break;
     }
-  }, _PromptCachingBetaMessageStream_endRequest = function _PromptCachingBetaMessageStream_endRequest() {
+  }, _PromptCachingBetaMessageStream_endRequest = function _PromptCachingBetaMessageStream_endRequest2() {
     if (this.ended) {
       throw new AnthropicError(`stream has ended, this shouldn't happen`);
     }
@@ -2023,7 +2023,7 @@ class PromptCachingBetaMessageStream {
     }
     __classPrivateFieldSet2(this, _PromptCachingBetaMessageStream_currentMessageSnapshot, undefined, "f");
     return snapshot;
-  }, _PromptCachingBetaMessageStream_accumulateMessage = function _PromptCachingBetaMessageStream_accumulateMessage(event) {
+  }, _PromptCachingBetaMessageStream_accumulateMessage = function _PromptCachingBetaMessageStream_accumulateMessage2(event) {
     let snapshot = __classPrivateFieldGet2(this, _PromptCachingBetaMessageStream_currentMessageSnapshot, "f");
     if (event.type === "message_start") {
       if (snapshot) {
@@ -2427,12 +2427,12 @@ class MessageStream {
     }
     __classPrivateFieldGet3(this, _MessageStream_instances, "m", _MessageStream_endRequest).call(this);
   }
-  [(_MessageStream_currentMessageSnapshot = new WeakMap, _MessageStream_connectedPromise = new WeakMap, _MessageStream_resolveConnectedPromise = new WeakMap, _MessageStream_rejectConnectedPromise = new WeakMap, _MessageStream_endPromise = new WeakMap, _MessageStream_resolveEndPromise = new WeakMap, _MessageStream_rejectEndPromise = new WeakMap, _MessageStream_listeners = new WeakMap, _MessageStream_ended = new WeakMap, _MessageStream_errored = new WeakMap, _MessageStream_aborted = new WeakMap, _MessageStream_catchingPromiseCreated = new WeakMap, _MessageStream_handleError = new WeakMap, _MessageStream_instances = new WeakSet, _MessageStream_getFinalMessage = function _MessageStream_getFinalMessage() {
+  [(_MessageStream_currentMessageSnapshot = new WeakMap, _MessageStream_connectedPromise = new WeakMap, _MessageStream_resolveConnectedPromise = new WeakMap, _MessageStream_rejectConnectedPromise = new WeakMap, _MessageStream_endPromise = new WeakMap, _MessageStream_resolveEndPromise = new WeakMap, _MessageStream_rejectEndPromise = new WeakMap, _MessageStream_listeners = new WeakMap, _MessageStream_ended = new WeakMap, _MessageStream_errored = new WeakMap, _MessageStream_aborted = new WeakMap, _MessageStream_catchingPromiseCreated = new WeakMap, _MessageStream_handleError = new WeakMap, _MessageStream_instances = new WeakSet, _MessageStream_getFinalMessage = function _MessageStream_getFinalMessage2() {
     if (this.receivedMessages.length === 0) {
       throw new AnthropicError("stream ended without producing a Message with role=assistant");
     }
     return this.receivedMessages.at(-1);
-  }, _MessageStream_getFinalText = function _MessageStream_getFinalText() {
+  }, _MessageStream_getFinalText = function _MessageStream_getFinalText2() {
     if (this.receivedMessages.length === 0) {
       throw new AnthropicError("stream ended without producing a Message with role=assistant");
     }
@@ -2441,11 +2441,11 @@ class MessageStream {
       throw new AnthropicError("stream ended without producing a content block with type=text");
     }
     return textBlocks.join(" ");
-  }, _MessageStream_beginRequest = function _MessageStream_beginRequest() {
+  }, _MessageStream_beginRequest = function _MessageStream_beginRequest2() {
     if (this.ended)
       return;
     __classPrivateFieldSet3(this, _MessageStream_currentMessageSnapshot, undefined, "f");
-  }, _MessageStream_addStreamEvent = function _MessageStream_addStreamEvent(event) {
+  }, _MessageStream_addStreamEvent = function _MessageStream_addStreamEvent2(event) {
     if (this.ended)
       return;
     const messageSnapshot = __classPrivateFieldGet3(this, _MessageStream_instances, "m", _MessageStream_accumulateMessage).call(this, event);
@@ -2479,7 +2479,7 @@ class MessageStream {
       case "message_delta":
         break;
     }
-  }, _MessageStream_endRequest = function _MessageStream_endRequest() {
+  }, _MessageStream_endRequest = function _MessageStream_endRequest2() {
     if (this.ended) {
       throw new AnthropicError(`stream has ended, this shouldn't happen`);
     }
@@ -2489,7 +2489,7 @@ class MessageStream {
     }
     __classPrivateFieldSet3(this, _MessageStream_currentMessageSnapshot, undefined, "f");
     return snapshot;
-  }, _MessageStream_accumulateMessage = function _MessageStream_accumulateMessage(event) {
+  }, _MessageStream_accumulateMessage = function _MessageStream_accumulateMessage2(event) {
     let snapshot = __classPrivateFieldGet3(this, _MessageStream_currentMessageSnapshot, "f");
     if (event.type === "message_start") {
       if (snapshot) {