@ailib-official/ai-protocol 0.8.4

package/v2-alpha/providers/anthropic.yaml

@@ -0,0 +1,134 @@
+# AI-Protocol V2-Alpha: Anthropic Provider Manifest
+# Migrated from v1/providers/anthropic.yaml with V2 structured capabilities
+$schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v2/provider.json"
+
+# === Ring 1: Core Skeleton ===
+id: anthropic
+protocol_version: "2.0"
+name: "Anthropic"
+version: "v1"
+status: stable
+category: ai_provider
+official_url: "https://docs.anthropic.com"
+support_contact: "https://support.anthropic.com"
+
+endpoint:
+  base_url: "https://api.anthropic.com/v1"
+  chat: "/messages"
+  auth:
+    type: "custom_header"
+    header: "x-api-key"
+    token_env: "ANTHROPIC_API_KEY"
+    extra_headers:
+      - name: "anthropic-version"
+        value: "2023-06-01"
+
+error_classification:
+  by_http_status:
+    "400": "invalid_request"
+    "401": "authentication"
+    "403": "permission_denied"
+    "404": "not_found"
+    "413": "request_too_large"
+    "429": "rate_limited"
+    "500": "server_error"
+    "529": "overloaded"
+  by_error_code:
+    "invalid_request_error": "invalid_request"
+    "authentication_error": "authentication"
+    "permission_error": "permission_denied"
+    "not_found_error": "not_found"
+    "request_too_large": "request_too_large"
+    "rate_limit_error": "rate_limited"
+    "api_error": "server_error"
+    "overloaded_error": "overloaded"
+
+# === Ring 2: Capability Mapping ===
+capabilities:
+  required:
+    - text
+    - streaming
+  optional:
+    - vision
+    - tools
+    - parallel_tools
+    - agentic
+    - reasoning
+    - batch
+  feature_flags:
+    structured_output: false
+    parallel_tool_calls: true
+    extended_thinking: true
+    streaming_usage: true
+    system_messages: true
+
+parameters:
+  temperature: { type: float, range: [0.0, 2.0], default: 1.0 }
+  # provider_specific_constraints:
+  #   max_value_override: 1.0  # Anthropic实际限制为[0.0, 1.0]
+  max_tokens: { type: integer, min: 1, max: 8192, required: true }
+  top_p: { type: float, range: [0.0, 1.0] }
+  top_k: { type: integer, min: 0 }
+  stream: { type: boolean }
+
+streaming:
+  decoder:
+    format: "anthropic_sse"
+    strategy: "anthropic_event_stream"
+  event_map:
+    - match: "$.type == 'content_block_delta' && $.delta.type == 'text_delta'"
+      emit: "PartialContentDelta"
+      extract:
+        content: "$.delta.text"
+    - match: "$.type == 'content_block_delta' && $.delta.type == 'input_json_delta'"
+      emit: "PartialToolCall"
+      extract:
+        partial_json: "$.delta.partial_json"
+    - match: "$.type == 'content_block_delta' && $.delta.type == 'thinking_delta'"
+      emit: "ThinkingDelta"
+      extract:
+        thinking: "$.delta.thinking"
+    - match: "$.type == 'message_delta'"
+      emit: "StreamEnd"
+      extract:
+        finish_reason: "$.delta.stop_reason"
+        usage: "$.usage"
+  accumulator:
+    stateful_tool_parsing: true
+    key_path: "$.delta.partial_json"
+    flush_on: "$.type == 'content_block_stop'"
+
+# === Ring 3: Advanced Extensions ===
+api_families: ["messages"]
+default_api_family: "messages"
+
+endpoints:
+  chat:
+    path: "/messages"
+    method: "POST"
+    adapter: "anthropic"
+
+rate_limit_headers:
+  # Provider-specific headers (保留用于原始响应解析)
+  # Normalized fields (rate_limit_*) are auto-filled by runtime parser.
+  retry_after: "retry-after"
+  requests_limit: "anthropic-ratelimit-requests-limit"
+  requests_remaining: "anthropic-ratelimit-requests-remaining"
+  tokens_limit: "anthropic-ratelimit-tokens-limit"
+  tokens_remaining: "anthropic-ratelimit-tokens-remaining"
+
+retry_policy:
+  strategy: "exponential_backoff"
+  max_retries: 2
+  min_delay_ms: 1000
+  max_delay_ms: 60000
+  jitter: "full"
+  retry_on_http_status: [408, 409, 429, 500, 529]
+
+termination:
+  source_field: "stop_reason"
+  mapping:
+    end_turn: "end_turn"
+    max_tokens: "max_tokens"
+    stop_sequence: "stop_sequence"
+    tool_use: "tool_use"

package/v2-alpha/providers/gemini.yaml

@@ -0,0 +1,144 @@
+# AI-Protocol V2-Alpha: Google Gemini Provider Manifest
+# Migrated from v1/providers/gemini.yaml with V2 structured capabilities
+$schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v2/provider.json"
+
+# === Ring 1: Core Skeleton ===
+id: gemini
+protocol_version: "2.0"
+name: "Google Gemini"
+version: "v1beta"
+status: stable
+category: ai_provider
+official_url: "https://ai.google.dev/docs"
+support_contact: "https://support.google.com/gemini"
+
+endpoint:
+  base_url: "https://generativelanguage.googleapis.com/v1beta"
+  chat: "/models/{model}:generateContent"
+  auth:
+    type: "query_param"
+    param_name: "key"
+    token_env: "GEMINI_API_KEY"
+
+error_classification:
+  by_http_status:
+    "400": "invalid_request"
+    "401": "authentication"
+    "403": "permission_denied"
+    "404": "not_found"
+    "429": "rate_limited"
+    "500": "server_error"
+    "503": "overloaded"
+  by_error_code:
+    "INVALID_ARGUMENT": "invalid_request"
+    "UNAUTHENTICATED": "authentication"
+    "PERMISSION_DENIED": "permission_denied"
+    "NOT_FOUND": "not_found"
+    "RESOURCE_EXHAUSTED": "rate_limited"
+    "INTERNAL": "server_error"
+    "UNAVAILABLE": "overloaded"
+    "DEADLINE_EXCEEDED": "timeout"
+
+# === Ring 2: Capability Mapping ===
+capabilities:
+  required:
+    - text
+    - streaming
+  optional:
+    - vision
+    - audio
+    - video
+    - tools
+    - parallel_tools
+    - reasoning
+    - embeddings
+    - structured_output
+  feature_flags:
+    structured_output: true
+    parallel_tool_calls: true
+    extended_thinking: true
+    streaming_usage: false
+    system_messages: true
+
+parameters:
+  temperature: { type: float, range: [0.0, 2.0], default: 1.0 }
+  maxOutputTokens: { type: integer, min: 1, max: 65536, alias: "max_tokens", deprecated: true, required: true }
+  topP: { type: float, range: [0.0, 1.0], alias: "top_p", deprecated: true }
+  topK: { type: integer, min: 0, alias: "top_k", deprecated: true }
+
+streaming:
+  decoder:
+    format: "gemini_sse"
+    strategy: "gemini_generate"
+  event_map:
+    - match: "$.candidates[0].content.parts[0].text"
+      emit: "PartialContentDelta"
+      extract:
+        content: "$.candidates[0].content.parts[0].text"
+    - match: "$.candidates[0].content.parts[0].functionCall"
+      emit: "PartialToolCall"
+      extract:
+        tool_call: "$.candidates[0].content.parts[0].functionCall"
+    - match: "$.candidates[0].finishReason"
+      emit: "StreamEnd"
+      extract:
+        finish_reason: "$.candidates[0].finishReason"
+  candidate:
+    candidate_id_path: "$.candidates[*].index"
+    fan_out: true
+
+# === Ring 3: Advanced Extensions ===
+api_families: ["generate_content"]
+default_api_family: "generate_content"
+
+endpoints:
+  chat:
+    path: "/models/{model}:generateContent"
+    method: "POST"
+    adapter: "gemini"
+  stream_chat:
+    path: "/models/{model}:streamGenerateContent"
+    method: "POST"
+    adapter: "gemini"
+  embeddings:
+    path: "/models/{model}:embedContent"
+    method: "POST"
+    adapter: "gemini"
+
+services:
+  list_models:
+    path: "/models"
+    method: "GET"
+    response_binding: "models"
+  get_model:
+    path: "/models/{model}"
+    method: "GET"
+
+rate_limit_headers:
+  # Provider-specific headers (Gemini 没有rate limit头部)
+  # Normalized fields (运行时标准访问)
+  # rate_limit_requests_limit: (配置为null，Gemini不支持)
+  # rate_limit_requests_remaining: (配置为null，Gemini不支持)
+  # rate_limit_tokens_limit: (配置为null，Gemini不支持)
+  # rate_limit_tokens_remaining: (配置为null，Gemini不支持)
+  # rate_limit_retry_after: (配置为null，Gemini不支持)
+  # Empty object (Gemini does not support rate limit headers)
+  {}
+retry_policy:
+  strategy: "exponential_backoff"
+  max_retries: 3
+  min_delay_ms: 1000
+  max_delay_ms: 30000
+  jitter: "full"
+  retry_on_http_status: [429, 500, 503]
+
+termination:
+  source_field: "finishReason"
+  mapping:
+    STOP: "end_turn"
+    MAX_TOKENS: "max_tokens"
+    SAFETY: "refusal"
+    RECITATION: "refusal"
+    OTHER: "other"
+    BLOCKLIST: "refusal"
+    PROHIBITED_CONTENT: "refusal"

package/v2-alpha/providers/openai.yaml

@@ -0,0 +1,154 @@
+# AI-Protocol V2-Alpha: OpenAI Provider Manifest
+# Migrated from v1/providers/openai.yaml with V2 structured capabilities
+$schema: "https://raw.githubusercontent.com/ailib-official/ai-protocol/main/schemas/v2/provider.json"
+
+# === Ring 1: Core Skeleton ===
+id: openai
+protocol_version: "2.0"
+name: "OpenAI"
+version: "v1"
+status: stable
+category: ai_provider
+official_url: "https://platform.openai.com/docs"
+support_contact: "https://help.openai.com"
+
+endpoint:
+  base_url: "https://api.openai.com/v1"
+  chat: "/chat/completions"
+  embeddings: "/embeddings"
+  auth:
+    type: "bearer"
+    header: "Authorization"
+    prefix: "Bearer"
+    token_env: "OPENAI_API_KEY"
+
+error_classification:
+  by_http_status:
+    "400": "invalid_request"
+    "401": "authentication"
+    "403": "permission_denied"
+    "404": "not_found"
+    "429": "rate_limited"
+    "500": "server_error"
+    "503": "overloaded"
+  by_error_code:
+    "invalid_api_key": "authentication"
+    "model_not_found": "not_found"
+    "context_length_exceeded": "request_too_large"
+    "rate_limit_exceeded": "rate_limited"
+    "insufficient_quota": "quota_exhausted"
+    "server_error": "server_error"
+
+# === Ring 2: Capability Mapping ===
+capabilities:
+  required:
+    - text
+    - streaming
+  optional:
+    - vision
+    - tools
+    - parallel_tools
+    - embeddings
+    - structured_output
+    - batch
+    - audio
+    - image_generation
+  feature_flags:
+    structured_output: true
+    parallel_tool_calls: true
+    streaming_usage: true
+    system_messages: true
+    image_generation: true
+
+parameters:
+  temperature: { type: float, range: [0.0, 2.0], default: 1.0 }
+  max_tokens: { type: integer, min: 1, max: 128000, required: true }
+  top_p: { type: float, range: [0.0, 1.0] }
+  frequency_penalty: { type: float, range: [-2.0, 2.0] }
+  presence_penalty: { type: float, range: [-2.0, 2.0] }
+  stream: { type: boolean }
+  n: { type: integer, min: 1 }
+  response_format: { type: object }
+
+streaming:
+  decoder:
+    format: "sse"
+    strategy: "openai_chat"
+    done_signal: "[DONE]"
+    prefix: "data: "
+  event_map:
+    - match: "$.choices[0].delta.content"
+      emit: "PartialContentDelta"
+      extract:
+        content: "$.choices[0].delta.content"
+    - match: "$.choices[0].delta.tool_calls"
+      emit: "PartialToolCall"
+      extract:
+        tool_calls: "$.choices[0].delta.tool_calls"
+    - match: "$.choices[0].finish_reason"
+      emit: "StreamEnd"
+      extract:
+        finish_reason: "$.choices[0].finish_reason"
+    - match: "$.usage"
+      emit: "UsageReport"
+      extract:
+        usage: "$.usage"
+  candidate:
+    candidate_id_path: "$.choices[*].index"
+    fan_out: true
+  accumulator:
+    stateful_tool_parsing: true
+    key_path: "$.choices[0].delta.tool_calls[0].function.arguments"
+    flush_on: "$.choices[0].finish_reason == 'tool_calls'"
+
+# === Ring 3: Advanced Extensions ===
+api_families: ["chat_completions", "completions", "embeddings", "assistants", "realtime"]
+default_api_family: "chat_completions"
+
+endpoints:
+  chat:
+    path: "/chat/completions"
+    method: "POST"
+    adapter: "openai"
+  embeddings:
+    path: "/embeddings"
+    method: "POST"
+    adapter: "openai"
+
+services:
+  list_models:
+    path: "/models"
+    method: "GET"
+    response_binding: "data"
+
+rate_limit_headers:
+  # Provider-specific headers (保留用于原始响应解析)
+  requests_limit: "x-ratelimit-limit-requests"
+  requests_remaining: "x-ratelimit-remaining-requests"
+  tokens_limit: "x-ratelimit-limit-tokens"
+  tokens_remaining: "x-ratelimit-remaining-tokens"
+  retry_after: null
+  # Normalized fields (运行时标准访问)
+  # rate_limit_requests_limit: (自动解析并填充)
+  # rate_limit_requests_remaining: (自动解析并填充)
+  # rate_limit_requests_reset: (自动解析并填充)
+  # rate_limit_tokens_limit: (自动解析并填充)
+  # rate_limit_tokens_remaining: (自动解析并填充)
+  # rate_limit_tokens_reset: (自动解析并填充)
+  # rate_limit_retry_after: (自动解析并填充)
+retry_policy:
+  strategy: "exponential_backoff"
+  max_retries: 3
+  min_delay_ms: 1000
+  max_delay_ms: 30000
+  jitter: "full"
+  retry_on_http_status: [429, 500, 502, 503]
+
+termination:
+  source_field: "finish_reason"
+  mapping:
+    stop: "end_turn"
+    length: "max_tokens"
+    content_filter: "refusal"
+    tool_calls: "tool_use"
+    function_call: "tool_use"