lex-llm 0.4.18 → 0.5.0

data/spec/legion/extensions/llm/conformance/fixtures/canonical_stop_reason_matrix.json

@@ -0,0 +1,36 @@
+{
+  "description": "Stop reason mapping matrix — every provider value maps to a canonical enum",
+  "canonical_stop_reasons": ["end_turn", "tool_use", "max_tokens", "stop_sequence", "content_filter", "error"],
+  "provider_mappings": {
+    "anthropic": {
+      "end_turn": "end_turn",
+      "tool_use": "tool_use",
+      "max_tokens": "max_tokens",
+      "stop_sequence": "stop_sequence",
+      "content_filter": "content_filter"
+    },
+    "openai": {
+      "stop": "end_turn",
+      "tool_calls": "tool_use",
+      "length": "max_tokens",
+      "content_filter": "content_filter"
+    },
+    "openai_responses": {
+      "completed": "end_turn",
+      "incomplete": "max_tokens",
+      "cancelled": "stop_sequence",
+      "failed": "error"
+    },
+    "vllm": {
+      "stop": "end_turn",
+      "tool_use": "tool_use",
+      "length": "max_tokens"
+    },
+    "bedrock_converse": {
+      "end_turn": "end_turn",
+      "tool_use": "tool_use",
+      "max_tokens": "max_tokens",
+      "guardrail_intervened": "content_filter"
+    }
+  }
+}

data/spec/legion/extensions/llm/conformance/fixtures/canonical_streaming_accumulated_response.json

@@ -0,0 +1,20 @@
+{
+  "description": "Expected accumulated Canonical::Response after processing canonical_streaming_tool_call_chunks.json — verify accumulate(chunks) == parse(full)",
+  "text": "Let me check the weather for you.",
+  "thinking": null,
+  "tool_calls": [
+    {
+      "id": "call_def456",
+      "name": "get_weather",
+      "arguments": {"location": "San Francisco, CA", "unit": "fahrenheit"},
+      "source": "client",
+      "status": "pending"
+    }
+  ],
+  "usage": {
+    "input_tokens": 45,
+    "output_tokens": 28
+  },
+  "stop_reason": "tool_use",
+  "model": "test-model-1"
+}

data/spec/legion/extensions/llm/conformance/fixtures/canonical_streaming_error_chunks.json

@@ -0,0 +1,26 @@
+{
+  "description": "Streaming error mid-chunk sequence — provider errors during stream per G5/G6",
+  "request_id": "req_stream_error_001",
+  "chunks": [
+    {
+      "request_id": "req_stream_error_001",
+      "index": 0,
+      "type": "text_delta",
+      "delta": "The weather"
+    },
+    {
+      "request_id": "req_stream_error_001",
+      "index": 1,
+      "type": "error",
+      "stop_reason": "error",
+      "metadata": {
+        "error": {
+          "type": "overloaded",
+          "message": "Upstream provider timed out",
+          "code": 504
+        },
+        "partial_text": "The weather"
+      }
+    }
+  ]
+}

data/spec/legion/extensions/llm/conformance/fixtures/canonical_streaming_text_chunks.json

@@ -0,0 +1,33 @@
+{
+  "description": "Streaming text-only chunk sequence — sanitized from real E2E traffic",
+  "request_id": "req_stream_text_001",
+  "chunks": [
+    {
+      "request_id": "req_stream_text_001",
+      "index": 0,
+      "type": "text_delta",
+      "delta": "Hello"
+    },
+    {
+      "request_id": "req_stream_text_001",
+      "index": 1,
+      "type": "text_delta",
+      "delta": ", world!"
+    },
+    {
+      "request_id": "req_stream_text_001",
+      "index": 2,
+      "type": "text_delta",
+      "delta": " How can I help you today?"
+    },
+    {
+      "request_id": "req_stream_text_001",
+      "type": "done",
+      "stop_reason": "end_turn",
+      "usage": {
+        "input_tokens": 12,
+        "output_tokens": 10
+      }
+    }
+  ]
+}

data/spec/legion/extensions/llm/conformance/fixtures/canonical_streaming_thinking_chunks.json

@@ -0,0 +1,42 @@
+{
+  "description": "Streaming thinking + text chunk sequence — sanitized from real E2E traffic",
+  "request_id": "req_stream_thinking_001",
+  "chunks": [
+    {
+      "request_id": "req_stream_thinking_001",
+      "index": 0,
+      "type": "thinking_delta",
+      "delta": "Let me think about this step by step.",
+      "signature": null
+    },
+    {
+      "request_id": "req_stream_thinking_001",
+      "index": 1,
+      "type": "thinking_delta",
+      "delta": "The key concepts are superposition and entanglement.",
+      "signature": "eyJ0eXAiOiJKV1QiLCJhbGciOiJFUzI1NiJ9"
+    },
+    {
+      "request_id": "req_stream_thinking_001",
+      "index": 2,
+      "type": "text_delta",
+      "delta": "Quantum computing uses qubits that can be in multiple states at once."
+    },
+    {
+      "request_id": "req_stream_thinking_001",
+      "index": 3,
+      "type": "text_delta",
+      "delta": " This enables parallel computation on a massive scale."
+    },
+    {
+      "request_id": "req_stream_thinking_001",
+      "type": "done",
+      "stop_reason": "end_turn",
+      "usage": {
+        "input_tokens": 15,
+        "output_tokens": 45,
+        "thinking_tokens": 80
+      }
+    }
+  ]
+}

data/spec/legion/extensions/llm/conformance/fixtures/canonical_streaming_tool_call_chunks.json

@@ -0,0 +1,41 @@
+{
+  "description": "Streaming tool-call chunk sequence — multi-chunk stateful tool-loop per A7. Each tool_call_delta carries the complete current state (not raw incremental fragments); the assembler merges deltas via buffer_policy.",
+  "request_id": "req_stream_tool_001",
+  "chunks": [
+    {
+      "request_id": "req_stream_tool_001",
+      "index": 0,
+      "type": "text_delta",
+      "delta": "Let me check the weather for you."
+    },
+    {
+      "request_id": "req_stream_tool_001",
+      "index": 1,
+      "type": "tool_call_delta",
+      "tool_call": {
+        "id": "call_def456",
+        "name": "get_weather",
+        "arguments": {"location": "San Francisco, CA"}
+      }
+    },
+    {
+      "request_id": "req_stream_tool_001",
+      "index": 2,
+      "type": "tool_call_delta",
+      "tool_call": {
+        "id": "call_def456",
+        "name": "get_weather",
+        "arguments": {"location": "San Francisco, CA", "unit": "fahrenheit"}
+      }
+    },
+    {
+      "request_id": "req_stream_tool_001",
+      "type": "done",
+      "stop_reason": "tool_use",
+      "usage": {
+        "input_tokens": 45,
+        "output_tokens": 28
+      }
+    }
+  ]
+}

data/spec/legion/extensions/llm/conformance/fixtures/canonical_system_prompt_request.json

@@ -0,0 +1,14 @@
+{
+  "id": "req_system_prompt_001",
+  "system": "You are a helpful assistant that always responds in haiku form.",
+  "messages": [
+    {
+      "role": "user",
+      "content": [{ "type": "text", "text": "Tell me about Ruby." }]
+    }
+  ],
+  "params": {
+    "max_tokens": 512
+  },
+  "stream": false
+}

data/spec/legion/extensions/llm/conformance/fixtures/canonical_thinking_request.json

@@ -0,0 +1,18 @@
+{
+  "id": "req_thinking_001",
+  "messages": [
+    {
+      "role": "user",
+      "content": [{ "type": "text", "text": "Explain quantum computing in simple terms." }]
+    }
+  ],
+  "thinking": {
+    "effort": "high",
+    "budget": 4096
+  },
+  "params": {
+    "max_tokens": 2048,
+    "max_thinking_tokens": 4096
+  },
+  "stream": false
+}

data/spec/legion/extensions/llm/conformance/fixtures/canonical_thinking_response.json

@@ -0,0 +1,17 @@
+{
+  "text": "Quantum computing uses quantum bits (qubits) that can represent both 0 and 1 simultaneously, unlike classical bits.",
+  "thinking": {
+    "content": "I need to explain quantum computing simply. Start with the basic difference between classical and quantum bits, then mention superposition and entanglement as key concepts.",
+    "signature": "eyJ0eXAiOiJKV1QiLCJhbGciOiJFUzI1NiJ9"
+  },
+  "tool_calls": [],
+  "usage": {
+    "input_tokens": 15,
+    "output_tokens": 45,
+    "thinking_tokens": 120
+  },
+  "stop_reason": "end_turn",
+  "model": "test-model-thinking-1",
+  "routing": {},
+  "metadata": {}
+}

data/spec/legion/extensions/llm/conformance/fixtures/canonical_tool_results_continuation_request.json

@@ -0,0 +1,75 @@
+{
+  "description": "Full multi-turn history: mixed client/server tool calls + thinking block per G4",
+  "id": "req_tool_continuation_001",
+  "system": "You are a helpful assistant.",
+  "messages": [
+    {
+      "role": "user",
+      "content": [{"type": "text", "text": "What's the weather in San Francisco, and summarize today's news about it?"}]
+    },
+    {
+      "role": "assistant",
+      "content": [
+        {"type": "text", "text": ""},
+        {"type": "tool_use", "id": "call_weather_01", "name": "get_weather", "input": {"location": "San Francisco, CA", "unit": "fahrenheit"}}
+      ],
+      "tool_calls": [
+        {
+          "id": "call_weather_01",
+          "exchange_id": "exch_turn1_001",
+          "name": "get_weather",
+          "arguments": {"location": "San Francisco, CA", "unit": "fahrenheit"},
+          "source": "client",
+          "status": "success",
+          "result": {"temperature": 68, "unit": "fahrenheit", "conditions": "partly cloudy"}
+        },
+        {
+          "id": "call_summarize_01",
+          "exchange_id": "exch_turn1_002",
+          "name": "summarize_article",
+          "arguments": {"topic": "San Francisco weather today"},
+          "source": "registry",
+          "status": "success",
+          "result": {"headline": "Partly cloudy skies, mild temperatures continue in Bay Area", "summary": "San Francisco enjoys its typical spring weather..."}
+        }
+      ]
+    },
+    {
+      "role": "tool",
+      "tool_call_id": "call_weather_01",
+      "content": [{"type": "tool_result", "tool_use_id": "call_weather_01", "text": "{\"temperature\":68,\"unit\":\"fahrenheit\",\"conditions\":\"partly cloudy\"}"}]
+    },
+    {
+      "role": "tool",
+      "tool_call_id": "call_summarize_01",
+      "content": [{"type": "tool_result", "tool_use_id": "call_summarize_01", "text": "{\"headline\":\"SF weather report\",\"summary\":\"Partly cloudy, 68F\"}"}]
+    }
+  ],
+  "tools": {
+    "get_weather": {
+      "name": "get_weather",
+      "description": "Get current weather for a location",
+      "parameters": {
+        "type": "object",
+        "properties": {
+          "location": {"type": "string"},
+          "unit": {"type": "string", "enum": ["celsius", "fahrenheit"]}
+        },
+        "required": ["location"]
+      }
+    },
+    "summarize_article": {
+      "name": "summarize_article",
+      "description": "Summarize a news article by topic",
+      "parameters": {
+        "type": "object",
+        "properties": {
+          "topic": {"type": "string"}
+        },
+        "required": ["topic"]
+      }
+    }
+  },
+  "params": {"max_tokens": 1024},
+  "stream": false
+}

data/spec/legion/extensions/llm/conformance/fixtures/canonical_tool_use_response.json

@@ -0,0 +1,25 @@
+{
+  "text": "",
+  "thinking": null,
+  "tool_calls": [
+    {
+      "id": "call_abc123",
+      "exchange_id": "exch_001",
+      "name": "get_weather",
+      "arguments": {
+        "location": "San Francisco, CA",
+        "unit": "fahrenheit"
+      },
+      "source": "client",
+      "status": "pending"
+    }
+  ],
+  "usage": {
+    "input_tokens": 45,
+    "output_tokens": 28
+  },
+  "stop_reason": "tool_use",
+  "model": "test-model-1",
+  "routing": {},
+  "metadata": {}
+}

data/spec/legion/extensions/llm/conformance/fixtures/canonical_tools_request.json

@@ -0,0 +1,34 @@
+{
+  "id": "req_tools_001",
+  "messages": [
+    {
+      "role": "user",
+      "content": [{ "type": "text", "text": "What's the weather in San Francisco?" }]
+    }
+  ],
+  "tools": {
+    "get_weather": {
+      "name": "get_weather",
+      "description": "Get the current weather in a given location",
+      "parameters": {
+        "type": "object",
+        "properties": {
+          "location": {
+            "type": "string",
+            "description": "The city and state, e.g. San Francisco, CA"
+          },
+          "unit": {
+            "type": "string",
+            "enum": ["celsius", "fahrenheit"],
+            "description": "The unit of temperature"
+          }
+        },
+        "required": ["location"]
+      }
+    }
+  },
+  "params": {
+    "max_tokens": 1024
+  },
+  "stream": false
+}