@graphorin/provider 0.5.0 → 0.6.0

package/dist/internal/http.js.map

@@ -1 +1 @@
-{"version":3,"file":"http.js","names":["out: Record<string, unknown>","parsed: unknown","buffer: string[]","resp: Response"],"sources":["../../src/internal/http.ts"],"sourcesContent":["/**\n * Internal HTTP helpers shared across the local-LLM adapters.\n *\n * @internal\n */\n\nimport type { ProviderEvent } from '@graphorin/core';\n\nimport { ProviderHttpError } from '../errors/errors.js';\n\n/**\n * Convert a graphorin `Message` to the OpenAI-compatible chat-completion\n * shape. The shape is the lingua-franca of the bundled local adapters\n * (`llamaCppServerAdapter` and `openAICompatibleAdapter`); the\n * native-Ollama path uses its own conversion.\n *\n * @internal\n */\nexport function toOpenAIChatMessages(\n  messages: ReadonlyArray<{\n    readonly role: 'system' | 'user' | 'assistant' | 'tool';\n    readonly content: string | ReadonlyArray<unknown>;\n    readonly toolCalls?: ReadonlyArray<{\n      readonly toolCallId: string;\n      readonly toolName: string;\n      readonly args: unknown;\n    }>;\n    readonly toolCallId?: string;\n  }>,\n): ReadonlyArray<Record<string, unknown>> {\n  return messages.map((msg) => {\n    const out: Record<string, unknown> = {\n      role: msg.role,\n      content: typeof msg.content === 'string' ? msg.content : flattenContent(msg.content),\n    };\n    if (msg.toolCalls !== undefined && msg.toolCalls.length > 0) {\n      out.tool_calls = msg.toolCalls.map((tc) => ({\n        id: tc.toolCallId,\n        type: 'function',\n        function: {\n          name: tc.toolName,\n          arguments: typeof tc.args === 'string' ? tc.args : JSON.stringify(tc.args),\n        },\n      }));\n    }\n    if (msg.role === 'tool' && msg.toolCallId !== undefined) {\n      out.tool_call_id = msg.toolCallId;\n    }\n    return out;\n  });\n}\n\n/**\n * Convert a graphorin `Message` to Ollama's **native** `/api/chat` shape\n * (PS-13). Unlike the OpenAI form, Ollama's Go server expects `tool_calls`\n * with object `arguments` (a `map[string]any`, never a JSON string) and no\n * `id` / `type` fields — sending those breaks its unmarshaller and any\n * multi-turn replay of assistant tool calls.\n *\n * @internal\n */\nexport function toOllamaChatMessages(\n  messages: ReadonlyArray<{\n    readonly role: 'system' | 'user' | 'assistant' | 'tool';\n    readonly content: string | ReadonlyArray<unknown>;\n    readonly toolCalls?: ReadonlyArray<{\n      readonly toolCallId: string;\n      readonly toolName: string;\n      readonly args: unknown;\n    }>;\n    readonly toolCallId?: string;\n  }>,\n): ReadonlyArray<Record<string, unknown>> {\n  return messages.map((msg) => {\n    const out: Record<string, unknown> = {\n      role: msg.role,\n      content: typeof msg.content === 'string' ? msg.content : flattenContent(msg.content),\n    };\n    if (msg.toolCalls !== undefined && msg.toolCalls.length > 0) {\n      out.tool_calls = msg.toolCalls.map((tc) => ({\n        function: {\n          name: tc.toolName,\n          arguments: toArgsObject(tc.args),\n        },\n      }));\n    }\n    return out;\n  });\n}\n\n/**\n * Coerce a tool-call `args` value into the object map Ollama expects. Strings\n * (e.g. an OpenAI-style JSON blob produced upstream) are parsed leniently;\n * anything that isn't a JSON object becomes `{}`.\n */\nfunction toArgsObject(args: unknown): Record<string, unknown> {\n  if (typeof args === 'string') {\n    try {\n      const parsed: unknown = JSON.parse(args);\n      return typeof parsed === 'object' && parsed !== null\n        ? (parsed as Record<string, unknown>)\n        : {};\n    } catch {\n      return {};\n    }\n  }\n  return typeof args === 'object' && args !== null ? (args as Record<string, unknown>) : {};\n}\n\nfunction flattenContent(parts: ReadonlyArray<unknown>): string {\n  const buffer: string[] = [];\n  for (const part of parts) {\n    if (typeof part === 'string') {\n      buffer.push(part);\n      continue;\n    }\n    if (typeof part === 'object' && part !== null) {\n      const obj = part as { type?: string; text?: string };\n      if (obj.type === 'text' && typeof obj.text === 'string') {\n        buffer.push(obj.text);\n      }\n    }\n  }\n  return buffer.join('');\n}\n\n/**\n * Wrap a `fetch` call with HTTP error mapping. The helper does not\n * assume any particular streaming format — callers receive the raw\n * `Response` and dispatch on its body.\n *\n * @internal\n */\n/**\n * Default per-request timeout for the baseUrl adapters (PS-24). Scoped\n * to time-to-response (headers): the timer is cleared the moment the\n * server answers, so long streaming bodies are never killed — only a\n * hung server that never responds. Generous because a cold local\n * llama-server can take tens of seconds to load a model.\n *\n * @stable\n */\nexport const DEFAULT_REQUEST_TIMEOUT_MS = 120_000;\n\nexport async function callJsonHttp(args: {\n  readonly providerName: string;\n  readonly url: string;\n  readonly headers: Record<string, string>;\n  readonly body: unknown;\n  readonly signal?: AbortSignal;\n  readonly fetchImpl?: typeof fetch;\n  /**\n   * Time-to-response budget (PS-24). Default\n   * {@link DEFAULT_REQUEST_TIMEOUT_MS}; `0` disables.\n   */\n  readonly timeoutMs?: number;\n}): Promise<Response> {\n  const fetchImpl = args.fetchImpl ?? globalThis.fetch.bind(globalThis);\n  const timeoutMs = args.timeoutMs ?? DEFAULT_REQUEST_TIMEOUT_MS;\n  const timeoutCtl = timeoutMs > 0 ? new AbortController() : undefined;\n  const timer =\n    timeoutCtl !== undefined ? setTimeout(() => timeoutCtl.abort(), timeoutMs) : undefined;\n  const signal =\n    args.signal !== undefined && timeoutCtl !== undefined\n      ? AbortSignal.any([args.signal, timeoutCtl.signal])\n      : (args.signal ?? timeoutCtl?.signal);\n  let resp: Response;\n  try {\n    resp = await fetchImpl(args.url, {\n      method: 'POST',\n      headers: args.headers,\n      body: JSON.stringify(args.body),\n      ...(signal !== undefined ? { signal } : {}),\n    });\n  } catch (cause) {\n    if (timeoutCtl?.signal.aborted === true && args.signal?.aborted !== true) {\n      throw new ProviderHttpError({\n        providerName: args.providerName,\n        status: 0,\n        message: `request timed out after ${timeoutMs}ms reaching ${args.url}`,\n        cause,\n      });\n    }\n    throw new ProviderHttpError({\n      providerName: args.providerName,\n      status: 0,\n      message: `network error reaching ${args.url}`,\n      cause,\n    });\n  } finally {\n    if (timer !== undefined) clearTimeout(timer);\n  }\n  if (!resp.ok) {\n    const detail = await safeReadText(resp);\n    throw new ProviderHttpError({\n      providerName: args.providerName,\n      status: resp.status,\n      message: detail.length > 0 ? detail : resp.statusText,\n    });\n  }\n  return resp;\n}\n\nasync function safeReadText(resp: Response): Promise<string> {\n  try {\n    return await resp.text();\n  } catch {\n    return '';\n  }\n}\n\n/**\n * Yield a `stream-start` `ProviderEvent` for an HTTP adapter.\n *\n * @internal\n */\nexport function makeStreamStartEvent(args: {\n  readonly providerName: string;\n  readonly modelId: string;\n  readonly responseId?: string;\n}): ProviderEvent {\n  return {\n    type: 'stream-start',\n    metadata: {\n      providerName: args.providerName,\n      modelId: args.modelId,\n      ...(args.responseId !== undefined ? { responseId: args.responseId } : {}),\n      createdAt: new Date().toISOString(),\n    },\n  };\n}\n"],"mappings":";;;;;;;;;;;AAkBA,SAAgB,qBACd,UAUwC;AACxC,QAAO,SAAS,KAAK,QAAQ;EAC3B,MAAMA,MAA+B;GACnC,MAAM,IAAI;GACV,SAAS,OAAO,IAAI,YAAY,WAAW,IAAI,UAAU,eAAe,IAAI,QAAQ;GACrF;AACD,MAAI,IAAI,cAAc,UAAa,IAAI,UAAU,SAAS,EACxD,KAAI,aAAa,IAAI,UAAU,KAAK,QAAQ;GAC1C,IAAI,GAAG;GACP,MAAM;GACN,UAAU;IACR,MAAM,GAAG;IACT,WAAW,OAAO,GAAG,SAAS,WAAW,GAAG,OAAO,KAAK,UAAU,GAAG,KAAK;IAC3E;GACF,EAAE;AAEL,MAAI,IAAI,SAAS,UAAU,IAAI,eAAe,OAC5C,KAAI,eAAe,IAAI;AAEzB,SAAO;GACP;;;;;;;;;;;AAYJ,SAAgB,qBACd,UAUwC;AACxC,QAAO,SAAS,KAAK,QAAQ;EAC3B,MAAMA,MAA+B;GACnC,MAAM,IAAI;GACV,SAAS,OAAO,IAAI,YAAY,WAAW,IAAI,UAAU,eAAe,IAAI,QAAQ;GACrF;AACD,MAAI,IAAI,cAAc,UAAa,IAAI,UAAU,SAAS,EACxD,KAAI,aAAa,IAAI,UAAU,KAAK,QAAQ,EAC1C,UAAU;GACR,MAAM,GAAG;GACT,WAAW,aAAa,GAAG,KAAK;GACjC,EACF,EAAE;AAEL,SAAO;GACP;;;;;;;AAQJ,SAAS,aAAa,MAAwC;AAC5D,KAAI,OAAO,SAAS,SAClB,KAAI;EACF,MAAMC,SAAkB,KAAK,MAAM,KAAK;AACxC,SAAO,OAAO,WAAW,YAAY,WAAW,OAC3C,SACD,EAAE;SACA;AACN,SAAO,EAAE;;AAGb,QAAO,OAAO,SAAS,YAAY,SAAS,OAAQ,OAAmC,EAAE;;AAG3F,SAAS,eAAe,OAAuC;CAC7D,MAAMC,SAAmB,EAAE;AAC3B,MAAK,MAAM,QAAQ,OAAO;AACxB,MAAI,OAAO,SAAS,UAAU;AAC5B,UAAO,KAAK,KAAK;AACjB;;AAEF,MAAI,OAAO,SAAS,YAAY,SAAS,MAAM;GAC7C,MAAM,MAAM;AACZ,OAAI,IAAI,SAAS,UAAU,OAAO,IAAI,SAAS,SAC7C,QAAO,KAAK,IAAI,KAAK;;;AAI3B,QAAO,OAAO,KAAK,GAAG;;;;;;;;;;;;;;;;;;AAmBxB,MAAa,6BAA6B;AAE1C,eAAsB,aAAa,MAYb;CACpB,MAAM,YAAY,KAAK,aAAa,WAAW,MAAM,KAAK,WAAW;CACrE,MAAM,YAAY,KAAK,aAAa;CACpC,MAAM,aAAa,YAAY,IAAI,IAAI,iBAAiB,GAAG;CAC3D,MAAM,QACJ,eAAe,SAAY,iBAAiB,WAAW,OAAO,EAAE,UAAU,GAAG;CAC/E,MAAM,SACJ,KAAK,WAAW,UAAa,eAAe,SACxC,YAAY,IAAI,CAAC,KAAK,QAAQ,WAAW,OAAO,CAAC,GAChD,KAAK,UAAU,YAAY;CAClC,IAAIC;AACJ,KAAI;AACF,SAAO,MAAM,UAAU,KAAK,KAAK;GAC/B,QAAQ;GACR,SAAS,KAAK;GACd,MAAM,KAAK,UAAU,KAAK,KAAK;GAC/B,GAAI,WAAW,SAAY,EAAE,QAAQ,GAAG,EAAE;GAC3C,CAAC;UACK,OAAO;AACd,MAAI,YAAY,OAAO,YAAY,QAAQ,KAAK,QAAQ,YAAY,KAClE,OAAM,IAAI,kBAAkB;GAC1B,cAAc,KAAK;GACnB,QAAQ;GACR,SAAS,2BAA2B,UAAU,cAAc,KAAK;GACjE;GACD,CAAC;AAEJ,QAAM,IAAI,kBAAkB;GAC1B,cAAc,KAAK;GACnB,QAAQ;GACR,SAAS,0BAA0B,KAAK;GACxC;GACD,CAAC;WACM;AACR,MAAI,UAAU,OAAW,cAAa,MAAM;;AAE9C,KAAI,CAAC,KAAK,IAAI;EACZ,MAAM,SAAS,MAAM,aAAa,KAAK;AACvC,QAAM,IAAI,kBAAkB;GAC1B,cAAc,KAAK;GACnB,QAAQ,KAAK;GACb,SAAS,OAAO,SAAS,IAAI,SAAS,KAAK;GAC5C,CAAC;;AAEJ,QAAO;;AAGT,eAAe,aAAa,MAAiC;AAC3D,KAAI;AACF,SAAO,MAAM,KAAK,MAAM;SAClB;AACN,SAAO;;;;;;;;AASX,SAAgB,qBAAqB,MAInB;AAChB,QAAO;EACL,MAAM;EACN,UAAU;GACR,cAAc,KAAK;GACnB,SAAS,KAAK;GACd,GAAI,KAAK,eAAe,SAAY,EAAE,YAAY,KAAK,YAAY,GAAG,EAAE;GACxE,4BAAW,IAAI,MAAM,EAAC,aAAa;GACpC;EACF"}
+{"version":3,"file":"http.js","names":["out: Record<string, unknown>","parsed: unknown","buffer: string[]","resp: Response","picked: Record<string, string>"],"sources":["../../src/internal/http.ts"],"sourcesContent":["/**\n * Internal HTTP helpers shared across the local-LLM adapters.\n *\n * @internal\n */\n\nimport type { ProviderEvent } from '@graphorin/core';\n\nimport { ProviderHttpError } from '../errors/errors.js';\n\n/**\n * Convert a graphorin `Message` to the OpenAI-compatible chat-completion\n * shape. The shape is the lingua-franca of the bundled local adapters\n * (`llamaCppServerAdapter` and `openAICompatibleAdapter`); the\n * native-Ollama path uses its own conversion.\n *\n * @internal\n */\nexport function toOpenAIChatMessages(\n  messages: ReadonlyArray<{\n    readonly role: 'system' | 'user' | 'assistant' | 'tool';\n    readonly content: string | ReadonlyArray<unknown>;\n    readonly toolCalls?: ReadonlyArray<{\n      readonly toolCallId: string;\n      readonly toolName: string;\n      readonly args: unknown;\n    }>;\n    readonly toolCallId?: string;\n  }>,\n): ReadonlyArray<Record<string, unknown>> {\n  return messages.map((msg) => {\n    const out: Record<string, unknown> = {\n      role: msg.role,\n      content: typeof msg.content === 'string' ? msg.content : flattenContent(msg.content),\n    };\n    if (msg.toolCalls !== undefined && msg.toolCalls.length > 0) {\n      out.tool_calls = msg.toolCalls.map((tc) => ({\n        id: tc.toolCallId,\n        type: 'function',\n        function: {\n          name: tc.toolName,\n          arguments: typeof tc.args === 'string' ? tc.args : JSON.stringify(tc.args),\n        },\n      }));\n    }\n    if (msg.role === 'tool' && msg.toolCallId !== undefined) {\n      out.tool_call_id = msg.toolCallId;\n    }\n    return out;\n  });\n}\n\n/**\n * Convert a graphorin `Message` to Ollama's **native** `/api/chat` shape\n * (PS-13). Unlike the OpenAI form, Ollama's Go server expects `tool_calls`\n * with object `arguments` (a `map[string]any`, never a JSON string) and no\n * `id` / `type` fields - sending those breaks its unmarshaller and any\n * multi-turn replay of assistant tool calls.\n *\n * @internal\n */\nexport function toOllamaChatMessages(\n  messages: ReadonlyArray<{\n    readonly role: 'system' | 'user' | 'assistant' | 'tool';\n    readonly content: string | ReadonlyArray<unknown>;\n    readonly toolCalls?: ReadonlyArray<{\n      readonly toolCallId: string;\n      readonly toolName: string;\n      readonly args: unknown;\n    }>;\n    readonly toolCallId?: string;\n  }>,\n): ReadonlyArray<Record<string, unknown>> {\n  return messages.map((msg) => {\n    const out: Record<string, unknown> = {\n      role: msg.role,\n      content: typeof msg.content === 'string' ? msg.content : flattenContent(msg.content),\n    };\n    if (msg.toolCalls !== undefined && msg.toolCalls.length > 0) {\n      out.tool_calls = msg.toolCalls.map((tc) => ({\n        function: {\n          name: tc.toolName,\n          arguments: toArgsObject(tc.args),\n        },\n      }));\n    }\n    return out;\n  });\n}\n\n/**\n * Coerce a tool-call `args` value into the object map Ollama expects. Strings\n * (e.g. an OpenAI-style JSON blob produced upstream) are parsed leniently;\n * anything that isn't a JSON object becomes `{}`.\n */\nfunction toArgsObject(args: unknown): Record<string, unknown> {\n  if (typeof args === 'string') {\n    try {\n      const parsed: unknown = JSON.parse(args);\n      return typeof parsed === 'object' && parsed !== null\n        ? (parsed as Record<string, unknown>)\n        : {};\n    } catch {\n      return {};\n    }\n  }\n  return typeof args === 'object' && args !== null ? (args as Record<string, unknown>) : {};\n}\n\nfunction flattenContent(parts: ReadonlyArray<unknown>): string {\n  const buffer: string[] = [];\n  for (const part of parts) {\n    if (typeof part === 'string') {\n      buffer.push(part);\n      continue;\n    }\n    if (typeof part === 'object' && part !== null) {\n      const obj = part as { type?: string; text?: string };\n      if (obj.type === 'text' && typeof obj.text === 'string') {\n        buffer.push(obj.text);\n      }\n    }\n  }\n  return buffer.join('');\n}\n\n/**\n * Wrap a `fetch` call with HTTP error mapping. The helper does not\n * assume any particular streaming format - callers receive the raw\n * `Response` and dispatch on its body.\n *\n * @internal\n */\n/**\n * Default per-request timeout for the baseUrl adapters (PS-24). Scoped\n * to time-to-response (headers): the timer is cleared the moment the\n * server answers, so long streaming bodies are never killed - only a\n * hung server that never responds. Generous because a cold local\n * llama-server can take tens of seconds to load a model.\n *\n * @stable\n */\nexport const DEFAULT_REQUEST_TIMEOUT_MS = 120_000;\n\nexport async function callJsonHttp(args: {\n  readonly providerName: string;\n  readonly url: string;\n  readonly headers: Record<string, string>;\n  readonly body: unknown;\n  readonly signal?: AbortSignal;\n  readonly fetchImpl?: typeof fetch;\n  /**\n   * Time-to-response budget (PS-24). Default\n   * {@link DEFAULT_REQUEST_TIMEOUT_MS}; `0` disables.\n   */\n  readonly timeoutMs?: number;\n}): Promise<Response> {\n  const fetchImpl = args.fetchImpl ?? globalThis.fetch.bind(globalThis);\n  const timeoutMs = args.timeoutMs ?? DEFAULT_REQUEST_TIMEOUT_MS;\n  const timeoutCtl = timeoutMs > 0 ? new AbortController() : undefined;\n  const timer =\n    timeoutCtl !== undefined ? setTimeout(() => timeoutCtl.abort(), timeoutMs) : undefined;\n  const signal =\n    args.signal !== undefined && timeoutCtl !== undefined\n      ? AbortSignal.any([args.signal, timeoutCtl.signal])\n      : (args.signal ?? timeoutCtl?.signal);\n  let resp: Response;\n  try {\n    resp = await fetchImpl(args.url, {\n      method: 'POST',\n      headers: args.headers,\n      body: JSON.stringify(args.body),\n      ...(signal !== undefined ? { signal } : {}),\n    });\n  } catch (cause) {\n    if (timeoutCtl?.signal.aborted === true && args.signal?.aborted !== true) {\n      throw new ProviderHttpError({\n        providerName: args.providerName,\n        status: 0,\n        message: `request timed out after ${timeoutMs}ms reaching ${args.url}`,\n        cause,\n      });\n    }\n    throw new ProviderHttpError({\n      providerName: args.providerName,\n      status: 0,\n      message: `network error reaching ${args.url}`,\n      cause,\n    });\n  } finally {\n    if (timer !== undefined) clearTimeout(timer);\n  }\n  if (!resp.ok) {\n    const detail = await safeReadText(resp);\n    const headers = pickBackoffHeaders(resp.headers);\n    throw new ProviderHttpError({\n      providerName: args.providerName,\n      status: resp.status,\n      message: detail.length > 0 ? detail : resp.statusText,\n      ...(headers !== undefined ? { headers } : {}),\n    });\n  }\n  return resp;\n}\n\n/**\n * Capture the backoff-relevant response headers (`retry-after`,\n * `x-ratelimit-*`) so `withRetry`'s Retry-After hint reader can honour\n * server-provided delays. Returns `undefined` when none are present.\n */\nfunction pickBackoffHeaders(headers: Headers): Readonly<Record<string, string>> | undefined {\n  const picked: Record<string, string> = {};\n  headers.forEach((value, key) => {\n    const lower = key.toLowerCase();\n    if (lower === 'retry-after' || lower.startsWith('x-ratelimit-')) {\n      picked[lower] = value;\n    }\n  });\n  return Object.keys(picked).length > 0 ? picked : undefined;\n}\n\nasync function safeReadText(resp: Response): Promise<string> {\n  try {\n    return await resp.text();\n  } catch {\n    return '';\n  }\n}\n\n/**\n * Yield a `stream-start` `ProviderEvent` for an HTTP adapter.\n *\n * @internal\n */\nexport function makeStreamStartEvent(args: {\n  readonly providerName: string;\n  readonly modelId: string;\n  readonly responseId?: string;\n}): ProviderEvent {\n  return {\n    type: 'stream-start',\n    metadata: {\n      providerName: args.providerName,\n      modelId: args.modelId,\n      ...(args.responseId !== undefined ? { responseId: args.responseId } : {}),\n      createdAt: new Date().toISOString(),\n    },\n  };\n}\n"],"mappings":";;;;;;;;;;;AAkBA,SAAgB,qBACd,UAUwC;AACxC,QAAO,SAAS,KAAK,QAAQ;EAC3B,MAAMA,MAA+B;GACnC,MAAM,IAAI;GACV,SAAS,OAAO,IAAI,YAAY,WAAW,IAAI,UAAU,eAAe,IAAI,QAAQ;GACrF;AACD,MAAI,IAAI,cAAc,UAAa,IAAI,UAAU,SAAS,EACxD,KAAI,aAAa,IAAI,UAAU,KAAK,QAAQ;GAC1C,IAAI,GAAG;GACP,MAAM;GACN,UAAU;IACR,MAAM,GAAG;IACT,WAAW,OAAO,GAAG,SAAS,WAAW,GAAG,OAAO,KAAK,UAAU,GAAG,KAAK;IAC3E;GACF,EAAE;AAEL,MAAI,IAAI,SAAS,UAAU,IAAI,eAAe,OAC5C,KAAI,eAAe,IAAI;AAEzB,SAAO;GACP;;;;;;;;;;;AAYJ,SAAgB,qBACd,UAUwC;AACxC,QAAO,SAAS,KAAK,QAAQ;EAC3B,MAAMA,MAA+B;GACnC,MAAM,IAAI;GACV,SAAS,OAAO,IAAI,YAAY,WAAW,IAAI,UAAU,eAAe,IAAI,QAAQ;GACrF;AACD,MAAI,IAAI,cAAc,UAAa,IAAI,UAAU,SAAS,EACxD,KAAI,aAAa,IAAI,UAAU,KAAK,QAAQ,EAC1C,UAAU;GACR,MAAM,GAAG;GACT,WAAW,aAAa,GAAG,KAAK;GACjC,EACF,EAAE;AAEL,SAAO;GACP;;;;;;;AAQJ,SAAS,aAAa,MAAwC;AAC5D,KAAI,OAAO,SAAS,SAClB,KAAI;EACF,MAAMC,SAAkB,KAAK,MAAM,KAAK;AACxC,SAAO,OAAO,WAAW,YAAY,WAAW,OAC3C,SACD,EAAE;SACA;AACN,SAAO,EAAE;;AAGb,QAAO,OAAO,SAAS,YAAY,SAAS,OAAQ,OAAmC,EAAE;;AAG3F,SAAS,eAAe,OAAuC;CAC7D,MAAMC,SAAmB,EAAE;AAC3B,MAAK,MAAM,QAAQ,OAAO;AACxB,MAAI,OAAO,SAAS,UAAU;AAC5B,UAAO,KAAK,KAAK;AACjB;;AAEF,MAAI,OAAO,SAAS,YAAY,SAAS,MAAM;GAC7C,MAAM,MAAM;AACZ,OAAI,IAAI,SAAS,UAAU,OAAO,IAAI,SAAS,SAC7C,QAAO,KAAK,IAAI,KAAK;;;AAI3B,QAAO,OAAO,KAAK,GAAG;;;;;;;;;;;;;;;;;;AAmBxB,MAAa,6BAA6B;AAE1C,eAAsB,aAAa,MAYb;CACpB,MAAM,YAAY,KAAK,aAAa,WAAW,MAAM,KAAK,WAAW;CACrE,MAAM,YAAY,KAAK,aAAa;CACpC,MAAM,aAAa,YAAY,IAAI,IAAI,iBAAiB,GAAG;CAC3D,MAAM,QACJ,eAAe,SAAY,iBAAiB,WAAW,OAAO,EAAE,UAAU,GAAG;CAC/E,MAAM,SACJ,KAAK,WAAW,UAAa,eAAe,SACxC,YAAY,IAAI,CAAC,KAAK,QAAQ,WAAW,OAAO,CAAC,GAChD,KAAK,UAAU,YAAY;CAClC,IAAIC;AACJ,KAAI;AACF,SAAO,MAAM,UAAU,KAAK,KAAK;GAC/B,QAAQ;GACR,SAAS,KAAK;GACd,MAAM,KAAK,UAAU,KAAK,KAAK;GAC/B,GAAI,WAAW,SAAY,EAAE,QAAQ,GAAG,EAAE;GAC3C,CAAC;UACK,OAAO;AACd,MAAI,YAAY,OAAO,YAAY,QAAQ,KAAK,QAAQ,YAAY,KAClE,OAAM,IAAI,kBAAkB;GAC1B,cAAc,KAAK;GACnB,QAAQ;GACR,SAAS,2BAA2B,UAAU,cAAc,KAAK;GACjE;GACD,CAAC;AAEJ,QAAM,IAAI,kBAAkB;GAC1B,cAAc,KAAK;GACnB,QAAQ;GACR,SAAS,0BAA0B,KAAK;GACxC;GACD,CAAC;WACM;AACR,MAAI,UAAU,OAAW,cAAa,MAAM;;AAE9C,KAAI,CAAC,KAAK,IAAI;EACZ,MAAM,SAAS,MAAM,aAAa,KAAK;EACvC,MAAM,UAAU,mBAAmB,KAAK,QAAQ;AAChD,QAAM,IAAI,kBAAkB;GAC1B,cAAc,KAAK;GACnB,QAAQ,KAAK;GACb,SAAS,OAAO,SAAS,IAAI,SAAS,KAAK;GAC3C,GAAI,YAAY,SAAY,EAAE,SAAS,GAAG,EAAE;GAC7C,CAAC;;AAEJ,QAAO;;;;;;;AAQT,SAAS,mBAAmB,SAAgE;CAC1F,MAAMC,SAAiC,EAAE;AACzC,SAAQ,SAAS,OAAO,QAAQ;EAC9B,MAAM,QAAQ,IAAI,aAAa;AAC/B,MAAI,UAAU,iBAAiB,MAAM,WAAW,eAAe,CAC7D,QAAO,SAAS;GAElB;AACF,QAAO,OAAO,KAAK,OAAO,CAAC,SAAS,IAAI,SAAS;;AAGnD,eAAe,aAAa,MAAiC;AAC3D,KAAI;AACF,SAAO,MAAM,KAAK,MAAM;SAClB;AACN,SAAO;;;;;;;;AASX,SAAgB,qBAAqB,MAInB;AAChB,QAAO;EACL,MAAM;EACN,UAAU;GACR,cAAc,KAAK;GACnB,SAAS,KAAK;GACd,GAAI,KAAK,eAAe,SAAY,EAAE,YAAY,KAAK,YAAY,GAAG,EAAE;GACxE,4BAAW,IAAI,MAAM,EAAC,aAAa;GACpC;EACF"}

package/dist/internal/openai-shaped.js

@@ -1,6 +1,7 @@
 import { LocalProviderInsecureTransportError, ProviderStreamParseError } from "../errors/errors.js";
 import { applyReasoningPolicy } from "../reasoning/apply-policy.js";
 import { resolveReasoningRetention } from "../reasoning/retention.js";
+import { foldToolExamples } from "../tool-examples.js";
 import { classifyLocalProvider } from "../trust/classify-local-provider.js";
 import { callJsonHttp, makeStreamStartEvent, toOpenAIChatMessages } from "./http.js";
 import { parseEventStream } from "./sse.js";
@@ -220,9 +221,10 @@ function buildBody(model, req, stream, structuredOutput) {
 		}, ...req.messages] : req.messages),
 		stream
 	};
+	if (stream) body.stream_options = { include_usage: true };
 	if (req.temperature !== void 0) body.temperature = req.temperature;
 	if (req.maxTokens !== void 0) body.max_tokens = req.maxTokens;
-	if (req.tools !== void 0 && req.tools.length > 0) body.tools = req.tools.map((t) => ({
+	if (req.tools !== void 0 && req.tools.length > 0) body.tools = foldToolExamples(req.tools).map((t) => ({
 		type: "function",
 		function: {
 			name: t.name,
@@ -277,11 +279,14 @@ function mapOpenAIUsage(input) {
 	const u = input;
 	const promptTokens = u.prompt_tokens ?? 0;
 	const completionTokens = u.completion_tokens ?? 0;
+	const totalTokens = u.total_tokens ?? promptTokens + completionTokens;
+	const cachedTokens = u.prompt_tokens_details?.cached_tokens;
 	return {
 		promptTokens,
 		completionTokens,
-		totalTokens: u.total_tokens ?? promptTokens + completionTokens,
-		...u.reasoning_tokens !== void 0 ? { reasoningTokens: u.reasoning_tokens } : {}
+		totalTokens,
+		...u.reasoning_tokens !== void 0 ? { reasoningTokens: u.reasoning_tokens } : {},
+		...typeof cachedTokens === "number" && Number.isFinite(cachedTokens) && cachedTokens > 0 ? { cachedReadTokens: cachedTokens } : {}
 	};
 }
 function parseFinalArgs(raw) {

package/dist/internal/openai-shaped.js.map

@@ -1 +1 @@
-{"version":3,"file":"openai-shaped.js","names":["DEFAULT_CAPABILITIES: ProviderCapabilities","capabilities: ProviderCapabilities","finishReason: FinishReason","usage: Usage | undefined","parsed: OpenAIChunk","json: OpenAIChunk","body: Record<string, unknown>","headers: Record<string, string>"],"sources":["../../src/internal/openai-shaped.ts"],"sourcesContent":["/**\n * Shared implementation backing the two OpenAI-compatible adapters\n * (`llamaCppServerAdapter` and `openAICompatibleAdapter`). The wire\n * shape is identical for the `/v1/chat/completions` endpoint, so we\n * factor the streaming + one-shot logic here and the adapters become\n * thin shells declaring trust class, defaults, and capability flags.\n *\n * @internal\n */\n\nimport type {\n  FinishReason,\n  Provider,\n  ProviderCapabilities,\n  ProviderEvent,\n  ProviderRequest,\n  ProviderResponse,\n  Sensitivity,\n  Usage,\n} from '@graphorin/core';\n\nimport { LocalProviderInsecureTransportError, ProviderStreamParseError } from '../errors/errors.js';\nimport { applyReasoningPolicy } from '../reasoning/apply-policy.js';\nimport { resolveReasoningRetention } from '../reasoning/retention.js';\nimport {\n  classifyLocalProvider,\n  type LocalProviderClassification,\n} from '../trust/classify-local-provider.js';\nimport { callJsonHttp, makeStreamStartEvent, toOpenAIChatMessages } from './http.js';\nimport { parseEventStream } from './sse.js';\nimport { stripTrailingSlashes } from './url-utils.js';\n\n/**\n * Options shared by every OpenAI-compatible adapter.\n *\n * @internal\n */\nexport interface OpenAIShapedOptions {\n  readonly providerName: string;\n  readonly model: string;\n  readonly baseUrl: string;\n  readonly chatPath?: string;\n  readonly apiKey?: string;\n  readonly headers?: Readonly<Record<string, string>>;\n  readonly fetchImpl?: typeof fetch;\n  readonly allowInsecureTransport?: boolean;\n  readonly capabilities?: Partial<ProviderCapabilities>;\n  /**\n   * Time-to-response budget per request (PS-24). Default\n   * `DEFAULT_REQUEST_TIMEOUT_MS` (120s); `0` disables.\n   */\n  readonly timeoutMs?: number;\n  /**\n   * Overrides the default `acceptsSensitivity` value derived from the\n   * trust classifier. Loud users can opt out of the conservative\n   * default, but the framework still emits one WARN per process so\n   * the override is visible.\n   */\n  readonly acceptsSensitivity?: ReadonlyArray<Sensitivity>;\n  /**\n   * Optional log sink. When unset, the adapter falls back to\n   * `console.warn`. Tests pass a fixture-driven sink to silence the\n   * console.\n   */\n  readonly logger?: (level: 'warn' | 'info', message: string, meta?: object) => void;\n}\n\nconst DEFAULT_CAPABILITIES: ProviderCapabilities = {\n  streaming: true,\n  toolCalling: true,\n  parallelToolCalls: false,\n  multimodal: false,\n  structuredOutput: true,\n  reasoning: false,\n  contextWindow: 8_192,\n  maxOutput: 4_096,\n  reasoningContract: 'optional',\n};\n\nconst KNOWN_LOOPBACK_OVERRIDES_WARNED = new Set<string>();\n\n/**\n * Build a Graphorin `Provider` backed by an OpenAI-compatible HTTP\n * server. Returns the wrapped object plus the resolved trust\n * classification so the caller can attach it to span attributes if\n * desired.\n *\n * @internal\n */\nexport function buildOpenAIShapedProvider(opts: OpenAIShapedOptions): {\n  readonly provider: Provider;\n  readonly classification: LocalProviderClassification;\n} {\n  const classification = classifyLocalProvider(opts.baseUrl);\n  if (classification.trust === 'public-cleartext' && opts.allowInsecureTransport !== true) {\n    throw new LocalProviderInsecureTransportError(opts.baseUrl);\n  }\n  const log = opts.logger ?? defaultLogger;\n  if (classification.trust === 'public-cleartext') {\n    log('warn', `[${opts.providerName}] allowInsecureTransport=true accepted for ${opts.baseUrl}`, {\n      baseUrl: opts.baseUrl,\n    });\n  } else if (classification.trust === 'public-tls') {\n    log('warn', `[${opts.providerName}] public-TLS endpoint; treating as cloud-tier`, {\n      baseUrl: opts.baseUrl,\n    });\n  } else if (classification.trust === 'private') {\n    log(\n      'warn',\n      `[${opts.providerName}] private-network endpoint detected (${classification.reason})`,\n      {\n        baseUrl: opts.baseUrl,\n        acceptsSensitivity: classification.acceptsSensitivity,\n      },\n    );\n  }\n\n  const acceptsSensitivity = opts.acceptsSensitivity ?? classification.acceptsSensitivity;\n  if (\n    opts.acceptsSensitivity !== undefined &&\n    !sameSensitivity(opts.acceptsSensitivity, classification.acceptsSensitivity) &&\n    !KNOWN_LOOPBACK_OVERRIDES_WARNED.has(opts.baseUrl)\n  ) {\n    KNOWN_LOOPBACK_OVERRIDES_WARNED.add(opts.baseUrl);\n    log(\n      'info',\n      `[${opts.providerName}] sensitivity override accepted; default would be [${classification.acceptsSensitivity.join(', ')}]`,\n      {\n        baseUrl: opts.baseUrl,\n      },\n    );\n  }\n\n  const capabilities: ProviderCapabilities = {\n    ...DEFAULT_CAPABILITIES,\n    ...opts.capabilities,\n  };\n  const chatPath = opts.chatPath ?? '/v1/chat/completions';\n  const url = `${stripTrailingSlashes(opts.baseUrl)}${chatPath}`;\n\n  const provider: Provider = {\n    name: opts.providerName,\n    modelId: opts.model,\n    capabilities,\n    acceptsSensitivity,\n    stream(req) {\n      return streamOpenAIShaped(opts, url, applyOpenAIShapedPreflight(req, capabilities));\n    },\n    async generate(req) {\n      return generateOpenAIShaped(opts, url, applyOpenAIShapedPreflight(req, capabilities));\n    },\n  };\n\n  return { provider, classification };\n}\n\nfunction applyOpenAIShapedPreflight(\n  req: ProviderRequest,\n  capabilities: ProviderCapabilities,\n): ProviderRequest {\n  const retention = resolveReasoningRetention({\n    ...(req.reasoningRetention !== undefined ? { requested: req.reasoningRetention } : {}),\n    ...(capabilities.reasoningContract !== undefined\n      ? { contract: capabilities.reasoningContract }\n      : {}),\n  });\n  if (retention === 'pass-through-all') return req;\n  const filtered = applyReasoningPolicy({ messages: req.messages, retention });\n  if (filtered === req.messages) return req;\n  return { ...req, messages: filtered };\n}\n\nasync function* streamOpenAIShaped(\n  opts: OpenAIShapedOptions,\n  url: string,\n  req: ProviderRequest,\n): AsyncIterable<ProviderEvent> {\n  const body = buildBody(opts.model, req, true, opts.capabilities?.structuredOutput ?? true);\n  const resp = await callJsonHttp({\n    providerName: opts.providerName,\n    url,\n    headers: buildHeaders(opts),\n    body,\n    ...(req.signal !== undefined ? { signal: req.signal } : {}),\n    ...(opts.fetchImpl !== undefined ? { fetchImpl: opts.fetchImpl } : {}),\n    ...(opts.timeoutMs !== undefined ? { timeoutMs: opts.timeoutMs } : {}),\n  });\n\n  yield makeStreamStartEvent({ providerName: opts.providerName, modelId: opts.model });\n\n  let finishReason: FinishReason = 'stop';\n  let usage: Usage | undefined;\n  const toolCallBuffer = new Map<\n    number,\n    { id: string; name: string; args: string; emittedStart: boolean }\n  >();\n\n  for await (const payload of parseEventStream(\n    resp.body,\n    req.signal !== undefined ? { signal: req.signal } : {},\n  )) {\n    if (req.signal?.aborted) {\n      // PS-12: report the abort honestly instead of the default 'stop'.\n      finishReason = 'aborted';\n      break;\n    }\n    let parsed: OpenAIChunk;\n    try {\n      parsed = JSON.parse(payload) as OpenAIChunk;\n    } catch (cause) {\n      throw new ProviderStreamParseError(\n        opts.providerName,\n        `failed to parse SSE chunk as JSON: ${(cause as Error).message}`,\n        cause,\n      );\n    }\n    const choice = parsed.choices?.[0];\n    if (choice === undefined) {\n      if (parsed.usage !== undefined) usage = mapOpenAIUsage(parsed.usage);\n      continue;\n    }\n    const delta = choice.delta ?? {};\n    if (typeof delta.content === 'string' && delta.content.length > 0) {\n      yield { type: 'text-delta', delta: delta.content };\n    }\n    if (typeof (delta as { reasoning_content?: unknown }).reasoning_content === 'string') {\n      const r = (delta as { reasoning_content: string }).reasoning_content;\n      if (r.length > 0) yield { type: 'reasoning-delta', delta: r };\n    }\n    if (Array.isArray(delta.tool_calls)) {\n      for (const tc of delta.tool_calls) {\n        const idx = tc.index ?? 0;\n        const slot = toolCallBuffer.get(idx) ?? {\n          id: '',\n          name: '',\n          args: '',\n          emittedStart: false,\n        };\n        if (typeof tc.id === 'string' && tc.id.length > 0) slot.id = tc.id;\n        if (tc.function?.name !== undefined) slot.name = tc.function.name;\n        if (typeof tc.function?.arguments === 'string') slot.args += tc.function.arguments;\n        toolCallBuffer.set(idx, slot);\n        if (!slot.emittedStart && slot.id.length > 0 && slot.name.length > 0) {\n          yield {\n            type: 'tool-call-start',\n            toolCallId: slot.id,\n            toolName: slot.name,\n          };\n          slot.emittedStart = true;\n        }\n        if (typeof tc.function?.arguments === 'string' && tc.function.arguments.length > 0) {\n          yield {\n            type: 'tool-call-input-delta',\n            toolCallId: slot.id,\n            argsDelta: tc.function.arguments,\n          };\n        }\n      }\n    }\n    if (typeof choice.finish_reason === 'string') {\n      finishReason = mapFinishReason(choice.finish_reason);\n      for (const slot of toolCallBuffer.values()) {\n        const finalArgs = parseFinalArgs(slot.args);\n        yield {\n          type: 'tool-call-end',\n          toolCallId: slot.id,\n          finalArgs,\n        };\n      }\n    }\n    if (parsed.usage !== undefined) usage = mapOpenAIUsage(parsed.usage);\n  }\n\n  yield {\n    type: 'finish',\n    finishReason,\n    usage: usage ?? { promptTokens: 0, completionTokens: 0, totalTokens: 0 },\n  };\n}\n\nasync function generateOpenAIShaped(\n  opts: OpenAIShapedOptions,\n  url: string,\n  req: ProviderRequest,\n): Promise<ProviderResponse> {\n  const body = buildBody(opts.model, req, false, opts.capabilities?.structuredOutput ?? true);\n  const resp = await callJsonHttp({\n    providerName: opts.providerName,\n    url,\n    headers: buildHeaders(opts),\n    body,\n    ...(req.signal !== undefined ? { signal: req.signal } : {}),\n    ...(opts.fetchImpl !== undefined ? { fetchImpl: opts.fetchImpl } : {}),\n    ...(opts.timeoutMs !== undefined ? { timeoutMs: opts.timeoutMs } : {}),\n  });\n  let json: OpenAIChunk;\n  try {\n    json = (await resp.json()) as OpenAIChunk;\n  } catch (cause) {\n    throw new ProviderStreamParseError(\n      opts.providerName,\n      'response body was not valid JSON',\n      cause,\n    );\n  }\n  const choice = json.choices?.[0];\n  const message = choice?.message ?? {};\n  const text = typeof message.content === 'string' ? message.content : undefined;\n  const toolCalls = Array.isArray(message.tool_calls)\n    ? message.tool_calls.map((tc) => ({\n        toolCallId: tc.id ?? '',\n        toolName: tc.function?.name ?? '',\n        args: parseFinalArgs(\n          typeof tc.function?.arguments === 'string' ? tc.function.arguments : '',\n        ),\n      }))\n    : undefined;\n  const usage = mapOpenAIUsage(json.usage) ?? {\n    promptTokens: 0,\n    completionTokens: 0,\n    totalTokens: 0,\n  };\n  return {\n    usage,\n    finishReason: mapFinishReason(choice?.finish_reason),\n    ...(text !== undefined ? { text } : {}),\n    ...(toolCalls !== undefined ? { toolCalls } : {}),\n  };\n}\n\nfunction buildBody(\n  model: string,\n  req: ProviderRequest,\n  stream: boolean,\n  structuredOutput: boolean,\n): Record<string, unknown> {\n  const messages =\n    req.systemMessage !== undefined\n      ? [{ role: 'system' as const, content: req.systemMessage }, ...req.messages]\n      : req.messages;\n  const body: Record<string, unknown> = {\n    model,\n    messages: toOpenAIChatMessages(messages),\n    stream,\n  };\n  if (req.temperature !== undefined) body.temperature = req.temperature;\n  if (req.maxTokens !== undefined) body.max_tokens = req.maxTokens;\n  if (req.tools !== undefined && req.tools.length > 0) {\n    body.tools = req.tools.map((t) => ({\n      type: 'function',\n      function: {\n        name: t.name,\n        ...(t.description !== undefined ? { description: t.description } : {}),\n        parameters: t.inputSchema,\n      },\n    }));\n  }\n  if (req.toolChoice !== undefined) {\n    body.tool_choice = mapToolChoice(req.toolChoice);\n  }\n  // PS-24: structured output finally reaches the wire — gated on the\n  // declared capability so a structuredOutput:false override keeps the\n  // request clean for servers that reject response_format.\n  if (structuredOutput && req.outputType?.kind === 'structured') {\n    body.response_format =\n      req.outputType.jsonSchema !== undefined\n        ? {\n            type: 'json_schema',\n            json_schema: { name: 'output', schema: req.outputType.jsonSchema, strict: true },\n          }\n        : { type: 'json_object' };\n  }\n  if (req.providerOptions !== undefined) {\n    Object.assign(body, req.providerOptions);\n  }\n  return body;\n}\n\nfunction buildHeaders(opts: OpenAIShapedOptions): Record<string, string> {\n  const headers: Record<string, string> = {\n    'content-type': 'application/json',\n    accept: 'application/json',\n    ...opts.headers,\n  };\n  if (opts.apiKey !== undefined && opts.apiKey.length > 0) {\n    headers.authorization = `Bearer ${opts.apiKey}`;\n  }\n  return headers;\n}\n\nfunction mapToolChoice(value: unknown): unknown {\n  if (value === 'auto' || value === 'none' || value === 'required') return value;\n  if (typeof value === 'object' && value !== null && 'tool' in value) {\n    return { type: 'function', function: { name: (value as { tool: string }).tool } };\n  }\n  return 'auto';\n}\n\nfunction mapFinishReason(value: string | null | undefined): FinishReason {\n  switch (value) {\n    case 'stop':\n    case 'length':\n    case 'content-filter':\n      return value;\n    // PS-12: the OpenAI wire format spells it with an underscore; the dashed\n    // form alone never matched, so content-filtered completions reported 'stop'.\n    case 'content_filter':\n      return 'content-filter';\n    case 'tool_calls':\n    case 'function_call':\n      return 'tool-calls';\n    case null:\n    case undefined:\n      return 'stop';\n    default:\n      return 'stop';\n  }\n}\n\nfunction mapOpenAIUsage(input: unknown): Usage | undefined {\n  if (input === undefined || input === null || typeof input !== 'object') return undefined;\n  const u = input as {\n    prompt_tokens?: number;\n    completion_tokens?: number;\n    total_tokens?: number;\n    reasoning_tokens?: number;\n  };\n  const promptTokens = u.prompt_tokens ?? 0;\n  const completionTokens = u.completion_tokens ?? 0;\n  const totalTokens = u.total_tokens ?? promptTokens + completionTokens;\n  const usage: Usage = {\n    promptTokens,\n    completionTokens,\n    totalTokens,\n    ...(u.reasoning_tokens !== undefined ? { reasoningTokens: u.reasoning_tokens } : {}),\n  };\n  return usage;\n}\n\nfunction parseFinalArgs(raw: string): unknown {\n  if (raw.length === 0) return {};\n  try {\n    return JSON.parse(raw);\n  } catch {\n    return raw;\n  }\n}\n\nfunction sameSensitivity(a: ReadonlyArray<Sensitivity>, b: ReadonlyArray<Sensitivity>): boolean {\n  if (a.length !== b.length) return false;\n  for (let i = 0; i < a.length; i++) {\n    if (a[i] !== b[i]) return false;\n  }\n  return true;\n}\n\nfunction defaultLogger(level: 'warn' | 'info', message: string, meta?: object): void {\n  const fn = level === 'warn' ? console.warn : console.info;\n  if (meta !== undefined) {\n    fn(`[graphorin/provider] ${message}`, meta);\n  } else {\n    fn(`[graphorin/provider] ${message}`);\n  }\n}\n\ninterface OpenAIChunk {\n  readonly id?: string;\n  readonly choices?: ReadonlyArray<{\n    readonly delta?: {\n      readonly content?: string;\n      readonly tool_calls?: ReadonlyArray<{\n        readonly index?: number;\n        readonly id?: string;\n        readonly type?: string;\n        readonly function?: { readonly name?: string; readonly arguments?: string };\n      }>;\n      readonly reasoning_content?: string;\n    };\n    readonly message?: {\n      readonly content?: string | null;\n      readonly tool_calls?: ReadonlyArray<{\n        readonly id?: string;\n        readonly type?: string;\n        readonly function?: { readonly name?: string; readonly arguments?: string };\n      }>;\n    };\n    readonly finish_reason?: string | null;\n  }>;\n  readonly usage?: {\n    readonly prompt_tokens?: number;\n    readonly completion_tokens?: number;\n    readonly total_tokens?: number;\n    readonly reasoning_tokens?: number;\n  };\n}\n"],"mappings":";;;;;;;;;AAmEA,MAAMA,uBAA6C;CACjD,WAAW;CACX,aAAa;CACb,mBAAmB;CACnB,YAAY;CACZ,kBAAkB;CAClB,WAAW;CACX,eAAe;CACf,WAAW;CACX,mBAAmB;CACpB;AAED,MAAM,kDAAkC,IAAI,KAAa;;;;;;;;;AAUzD,SAAgB,0BAA0B,MAGxC;CACA,MAAM,iBAAiB,sBAAsB,KAAK,QAAQ;AAC1D,KAAI,eAAe,UAAU,sBAAsB,KAAK,2BAA2B,KACjF,OAAM,IAAI,oCAAoC,KAAK,QAAQ;CAE7D,MAAM,MAAM,KAAK,UAAU;AAC3B,KAAI,eAAe,UAAU,mBAC3B,KAAI,QAAQ,IAAI,KAAK,aAAa,6CAA6C,KAAK,WAAW,EAC7F,SAAS,KAAK,SACf,CAAC;UACO,eAAe,UAAU,aAClC,KAAI,QAAQ,IAAI,KAAK,aAAa,gDAAgD,EAChF,SAAS,KAAK,SACf,CAAC;UACO,eAAe,UAAU,UAClC,KACE,QACA,IAAI,KAAK,aAAa,uCAAuC,eAAe,OAAO,IACnF;EACE,SAAS,KAAK;EACd,oBAAoB,eAAe;EACpC,CACF;CAGH,MAAM,qBAAqB,KAAK,sBAAsB,eAAe;AACrE,KACE,KAAK,uBAAuB,UAC5B,CAAC,gBAAgB,KAAK,oBAAoB,eAAe,mBAAmB,IAC5E,CAAC,gCAAgC,IAAI,KAAK,QAAQ,EAClD;AACA,kCAAgC,IAAI,KAAK,QAAQ;AACjD,MACE,QACA,IAAI,KAAK,aAAa,qDAAqD,eAAe,mBAAmB,KAAK,KAAK,CAAC,IACxH,EACE,SAAS,KAAK,SACf,CACF;;CAGH,MAAMC,eAAqC;EACzC,GAAG;EACH,GAAG,KAAK;EACT;CACD,MAAM,WAAW,KAAK,YAAY;CAClC,MAAM,MAAM,GAAG,qBAAqB,KAAK,QAAQ,GAAG;AAepD,QAAO;EAAE,UAbkB;GACzB,MAAM,KAAK;GACX,SAAS,KAAK;GACd;GACA;GACA,OAAO,KAAK;AACV,WAAO,mBAAmB,MAAM,KAAK,2BAA2B,KAAK,aAAa,CAAC;;GAErF,MAAM,SAAS,KAAK;AAClB,WAAO,qBAAqB,MAAM,KAAK,2BAA2B,KAAK,aAAa,CAAC;;GAExF;EAEkB;EAAgB;;AAGrC,SAAS,2BACP,KACA,cACiB;CACjB,MAAM,YAAY,0BAA0B;EAC1C,GAAI,IAAI,uBAAuB,SAAY,EAAE,WAAW,IAAI,oBAAoB,GAAG,EAAE;EACrF,GAAI,aAAa,sBAAsB,SACnC,EAAE,UAAU,aAAa,mBAAmB,GAC5C,EAAE;EACP,CAAC;AACF,KAAI,cAAc,mBAAoB,QAAO;CAC7C,MAAM,WAAW,qBAAqB;EAAE,UAAU,IAAI;EAAU;EAAW,CAAC;AAC5E,KAAI,aAAa,IAAI,SAAU,QAAO;AACtC,QAAO;EAAE,GAAG;EAAK,UAAU;EAAU;;AAGvC,gBAAgB,mBACd,MACA,KACA,KAC8B;CAC9B,MAAM,OAAO,UAAU,KAAK,OAAO,KAAK,MAAM,KAAK,cAAc,oBAAoB,KAAK;CAC1F,MAAM,OAAO,MAAM,aAAa;EAC9B,cAAc,KAAK;EACnB;EACA,SAAS,aAAa,KAAK;EAC3B;EACA,GAAI,IAAI,WAAW,SAAY,EAAE,QAAQ,IAAI,QAAQ,GAAG,EAAE;EAC1D,GAAI,KAAK,cAAc,SAAY,EAAE,WAAW,KAAK,WAAW,GAAG,EAAE;EACrE,GAAI,KAAK,cAAc,SAAY,EAAE,WAAW,KAAK,WAAW,GAAG,EAAE;EACtE,CAAC;AAEF,OAAM,qBAAqB;EAAE,cAAc,KAAK;EAAc,SAAS,KAAK;EAAO,CAAC;CAEpF,IAAIC,eAA6B;CACjC,IAAIC;CACJ,MAAM,iCAAiB,IAAI,KAGxB;AAEH,YAAW,MAAM,WAAW,iBAC1B,KAAK,MACL,IAAI,WAAW,SAAY,EAAE,QAAQ,IAAI,QAAQ,GAAG,EAAE,CACvD,EAAE;AACD,MAAI,IAAI,QAAQ,SAAS;AAEvB,kBAAe;AACf;;EAEF,IAAIC;AACJ,MAAI;AACF,YAAS,KAAK,MAAM,QAAQ;WACrB,OAAO;AACd,SAAM,IAAI,yBACR,KAAK,cACL,sCAAuC,MAAgB,WACvD,MACD;;EAEH,MAAM,SAAS,OAAO,UAAU;AAChC,MAAI,WAAW,QAAW;AACxB,OAAI,OAAO,UAAU,OAAW,SAAQ,eAAe,OAAO,MAAM;AACpE;;EAEF,MAAM,QAAQ,OAAO,SAAS,EAAE;AAChC,MAAI,OAAO,MAAM,YAAY,YAAY,MAAM,QAAQ,SAAS,EAC9D,OAAM;GAAE,MAAM;GAAc,OAAO,MAAM;GAAS;AAEpD,MAAI,OAAQ,MAA0C,sBAAsB,UAAU;GACpF,MAAM,IAAK,MAAwC;AACnD,OAAI,EAAE,SAAS,EAAG,OAAM;IAAE,MAAM;IAAmB,OAAO;IAAG;;AAE/D,MAAI,MAAM,QAAQ,MAAM,WAAW,CACjC,MAAK,MAAM,MAAM,MAAM,YAAY;GACjC,MAAM,MAAM,GAAG,SAAS;GACxB,MAAM,OAAO,eAAe,IAAI,IAAI,IAAI;IACtC,IAAI;IACJ,MAAM;IACN,MAAM;IACN,cAAc;IACf;AACD,OAAI,OAAO,GAAG,OAAO,YAAY,GAAG,GAAG,SAAS,EAAG,MAAK,KAAK,GAAG;AAChE,OAAI,GAAG,UAAU,SAAS,OAAW,MAAK,OAAO,GAAG,SAAS;AAC7D,OAAI,OAAO,GAAG,UAAU,cAAc,SAAU,MAAK,QAAQ,GAAG,SAAS;AACzE,kBAAe,IAAI,KAAK,KAAK;AAC7B,OAAI,CAAC,KAAK,gBAAgB,KAAK,GAAG,SAAS,KAAK,KAAK,KAAK,SAAS,GAAG;AACpE,UAAM;KACJ,MAAM;KACN,YAAY,KAAK;KACjB,UAAU,KAAK;KAChB;AACD,SAAK,eAAe;;AAEtB,OAAI,OAAO,GAAG,UAAU,cAAc,YAAY,GAAG,SAAS,UAAU,SAAS,EAC/E,OAAM;IACJ,MAAM;IACN,YAAY,KAAK;IACjB,WAAW,GAAG,SAAS;IACxB;;AAIP,MAAI,OAAO,OAAO,kBAAkB,UAAU;AAC5C,kBAAe,gBAAgB,OAAO,cAAc;AACpD,QAAK,MAAM,QAAQ,eAAe,QAAQ,EAAE;IAC1C,MAAM,YAAY,eAAe,KAAK,KAAK;AAC3C,UAAM;KACJ,MAAM;KACN,YAAY,KAAK;KACjB;KACD;;;AAGL,MAAI,OAAO,UAAU,OAAW,SAAQ,eAAe,OAAO,MAAM;;AAGtE,OAAM;EACJ,MAAM;EACN;EACA,OAAO,SAAS;GAAE,cAAc;GAAG,kBAAkB;GAAG,aAAa;GAAG;EACzE;;AAGH,eAAe,qBACb,MACA,KACA,KAC2B;CAC3B,MAAM,OAAO,UAAU,KAAK,OAAO,KAAK,OAAO,KAAK,cAAc,oBAAoB,KAAK;CAC3F,MAAM,OAAO,MAAM,aAAa;EAC9B,cAAc,KAAK;EACnB;EACA,SAAS,aAAa,KAAK;EAC3B;EACA,GAAI,IAAI,WAAW,SAAY,EAAE,QAAQ,IAAI,QAAQ,GAAG,EAAE;EAC1D,GAAI,KAAK,cAAc,SAAY,EAAE,WAAW,KAAK,WAAW,GAAG,EAAE;EACrE,GAAI,KAAK,cAAc,SAAY,EAAE,WAAW,KAAK,WAAW,GAAG,EAAE;EACtE,CAAC;CACF,IAAIC;AACJ,KAAI;AACF,SAAQ,MAAM,KAAK,MAAM;UAClB,OAAO;AACd,QAAM,IAAI,yBACR,KAAK,cACL,oCACA,MACD;;CAEH,MAAM,SAAS,KAAK,UAAU;CAC9B,MAAM,UAAU,QAAQ,WAAW,EAAE;CACrC,MAAM,OAAO,OAAO,QAAQ,YAAY,WAAW,QAAQ,UAAU;CACrE,MAAM,YAAY,MAAM,QAAQ,QAAQ,WAAW,GAC/C,QAAQ,WAAW,KAAK,QAAQ;EAC9B,YAAY,GAAG,MAAM;EACrB,UAAU,GAAG,UAAU,QAAQ;EAC/B,MAAM,eACJ,OAAO,GAAG,UAAU,cAAc,WAAW,GAAG,SAAS,YAAY,GACtE;EACF,EAAE,GACH;AAMJ,QAAO;EACL,OANY,eAAe,KAAK,MAAM,IAAI;GAC1C,cAAc;GACd,kBAAkB;GAClB,aAAa;GACd;EAGC,cAAc,gBAAgB,QAAQ,cAAc;EACpD,GAAI,SAAS,SAAY,EAAE,MAAM,GAAG,EAAE;EACtC,GAAI,cAAc,SAAY,EAAE,WAAW,GAAG,EAAE;EACjD;;AAGH,SAAS,UACP,OACA,KACA,QACA,kBACyB;CAKzB,MAAMC,OAAgC;EACpC;EACA,UAAU,qBALV,IAAI,kBAAkB,SAClB,CAAC;GAAE,MAAM;GAAmB,SAAS,IAAI;GAAe,EAAE,GAAG,IAAI,SAAS,GAC1E,IAAI,SAGgC;EACxC;EACD;AACD,KAAI,IAAI,gBAAgB,OAAW,MAAK,cAAc,IAAI;AAC1D,KAAI,IAAI,cAAc,OAAW,MAAK,aAAa,IAAI;AACvD,KAAI,IAAI,UAAU,UAAa,IAAI,MAAM,SAAS,EAChD,MAAK,QAAQ,IAAI,MAAM,KAAK,OAAO;EACjC,MAAM;EACN,UAAU;GACR,MAAM,EAAE;GACR,GAAI,EAAE,gBAAgB,SAAY,EAAE,aAAa,EAAE,aAAa,GAAG,EAAE;GACrE,YAAY,EAAE;GACf;EACF,EAAE;AAEL,KAAI,IAAI,eAAe,OACrB,MAAK,cAAc,cAAc,IAAI,WAAW;AAKlD,KAAI,oBAAoB,IAAI,YAAY,SAAS,aAC/C,MAAK,kBACH,IAAI,WAAW,eAAe,SAC1B;EACE,MAAM;EACN,aAAa;GAAE,MAAM;GAAU,QAAQ,IAAI,WAAW;GAAY,QAAQ;GAAM;EACjF,GACD,EAAE,MAAM,eAAe;AAE/B,KAAI,IAAI,oBAAoB,OAC1B,QAAO,OAAO,MAAM,IAAI,gBAAgB;AAE1C,QAAO;;AAGT,SAAS,aAAa,MAAmD;CACvE,MAAMC,UAAkC;EACtC,gBAAgB;EAChB,QAAQ;EACR,GAAG,KAAK;EACT;AACD,KAAI,KAAK,WAAW,UAAa,KAAK,OAAO,SAAS,EACpD,SAAQ,gBAAgB,UAAU,KAAK;AAEzC,QAAO;;AAGT,SAAS,cAAc,OAAyB;AAC9C,KAAI,UAAU,UAAU,UAAU,UAAU,UAAU,WAAY,QAAO;AACzE,KAAI,OAAO,UAAU,YAAY,UAAU,QAAQ,UAAU,MAC3D,QAAO;EAAE,MAAM;EAAY,UAAU,EAAE,MAAO,MAA2B,MAAM;EAAE;AAEnF,QAAO;;AAGT,SAAS,gBAAgB,OAAgD;AACvE,SAAQ,OAAR;EACE,KAAK;EACL,KAAK;EACL,KAAK,iBACH,QAAO;EAGT,KAAK,iBACH,QAAO;EACT,KAAK;EACL,KAAK,gBACH,QAAO;EACT,KAAK;EACL,KAAK,OACH,QAAO;EACT,QACE,QAAO;;;AAIb,SAAS,eAAe,OAAmC;AACzD,KAAI,UAAU,UAAa,UAAU,QAAQ,OAAO,UAAU,SAAU,QAAO;CAC/E,MAAM,IAAI;CAMV,MAAM,eAAe,EAAE,iBAAiB;CACxC,MAAM,mBAAmB,EAAE,qBAAqB;AAQhD,QANqB;EACnB;EACA;EACA,aAJkB,EAAE,gBAAgB,eAAe;EAKnD,GAAI,EAAE,qBAAqB,SAAY,EAAE,iBAAiB,EAAE,kBAAkB,GAAG,EAAE;EACpF;;AAIH,SAAS,eAAe,KAAsB;AAC5C,KAAI,IAAI,WAAW,EAAG,QAAO,EAAE;AAC/B,KAAI;AACF,SAAO,KAAK,MAAM,IAAI;SAChB;AACN,SAAO;;;AAIX,SAAS,gBAAgB,GAA+B,GAAwC;AAC9F,KAAI,EAAE,WAAW,EAAE,OAAQ,QAAO;AAClC,MAAK,IAAI,IAAI,GAAG,IAAI,EAAE,QAAQ,IAC5B,KAAI,EAAE,OAAO,EAAE,GAAI,QAAO;AAE5B,QAAO;;AAGT,SAAS,cAAc,OAAwB,SAAiB,MAAqB;CACnF,MAAM,KAAK,UAAU,SAAS,QAAQ,OAAO,QAAQ;AACrD,KAAI,SAAS,OACX,IAAG,wBAAwB,WAAW,KAAK;KAE3C,IAAG,wBAAwB,UAAU"}
+{"version":3,"file":"openai-shaped.js","names":["DEFAULT_CAPABILITIES: ProviderCapabilities","capabilities: ProviderCapabilities","finishReason: FinishReason","usage: Usage | undefined","parsed: OpenAIChunk","json: OpenAIChunk","body: Record<string, unknown>","headers: Record<string, string>"],"sources":["../../src/internal/openai-shaped.ts"],"sourcesContent":["/**\n * Shared implementation backing the two OpenAI-compatible adapters\n * (`llamaCppServerAdapter` and `openAICompatibleAdapter`). The wire\n * shape is identical for the `/v1/chat/completions` endpoint, so we\n * factor the streaming + one-shot logic here and the adapters become\n * thin shells declaring trust class, defaults, and capability flags.\n *\n * @internal\n */\n\nimport type {\n  FinishReason,\n  Provider,\n  ProviderCapabilities,\n  ProviderEvent,\n  ProviderRequest,\n  ProviderResponse,\n  Sensitivity,\n  Usage,\n} from '@graphorin/core';\n\nimport { LocalProviderInsecureTransportError, ProviderStreamParseError } from '../errors/errors.js';\nimport { applyReasoningPolicy } from '../reasoning/apply-policy.js';\nimport { resolveReasoningRetention } from '../reasoning/retention.js';\nimport { foldToolExamples } from '../tool-examples.js';\nimport {\n  classifyLocalProvider,\n  type LocalProviderClassification,\n} from '../trust/classify-local-provider.js';\nimport { callJsonHttp, makeStreamStartEvent, toOpenAIChatMessages } from './http.js';\nimport { parseEventStream } from './sse.js';\nimport { stripTrailingSlashes } from './url-utils.js';\n\n/**\n * Options shared by every OpenAI-compatible adapter.\n *\n * @internal\n */\nexport interface OpenAIShapedOptions {\n  readonly providerName: string;\n  readonly model: string;\n  readonly baseUrl: string;\n  readonly chatPath?: string;\n  readonly apiKey?: string;\n  readonly headers?: Readonly<Record<string, string>>;\n  readonly fetchImpl?: typeof fetch;\n  readonly allowInsecureTransport?: boolean;\n  readonly capabilities?: Partial<ProviderCapabilities>;\n  /**\n   * Time-to-response budget per request (PS-24). Default\n   * `DEFAULT_REQUEST_TIMEOUT_MS` (120s); `0` disables.\n   */\n  readonly timeoutMs?: number;\n  /**\n   * Overrides the default `acceptsSensitivity` value derived from the\n   * trust classifier. Loud users can opt out of the conservative\n   * default, but the framework still emits one WARN per process so\n   * the override is visible.\n   */\n  readonly acceptsSensitivity?: ReadonlyArray<Sensitivity>;\n  /**\n   * Optional log sink. When unset, the adapter falls back to\n   * `console.warn`. Tests pass a fixture-driven sink to silence the\n   * console.\n   */\n  readonly logger?: (level: 'warn' | 'info', message: string, meta?: object) => void;\n}\n\nconst DEFAULT_CAPABILITIES: ProviderCapabilities = {\n  streaming: true,\n  toolCalling: true,\n  parallelToolCalls: false,\n  multimodal: false,\n  structuredOutput: true,\n  reasoning: false,\n  contextWindow: 8_192,\n  maxOutput: 4_096,\n  reasoningContract: 'optional',\n};\n\nconst KNOWN_LOOPBACK_OVERRIDES_WARNED = new Set<string>();\n\n/**\n * Build a Graphorin `Provider` backed by an OpenAI-compatible HTTP\n * server. Returns the wrapped object plus the resolved trust\n * classification so the caller can attach it to span attributes if\n * desired.\n *\n * @internal\n */\nexport function buildOpenAIShapedProvider(opts: OpenAIShapedOptions): {\n  readonly provider: Provider;\n  readonly classification: LocalProviderClassification;\n} {\n  const classification = classifyLocalProvider(opts.baseUrl);\n  if (classification.trust === 'public-cleartext' && opts.allowInsecureTransport !== true) {\n    throw new LocalProviderInsecureTransportError(opts.baseUrl);\n  }\n  const log = opts.logger ?? defaultLogger;\n  if (classification.trust === 'public-cleartext') {\n    log('warn', `[${opts.providerName}] allowInsecureTransport=true accepted for ${opts.baseUrl}`, {\n      baseUrl: opts.baseUrl,\n    });\n  } else if (classification.trust === 'public-tls') {\n    log('warn', `[${opts.providerName}] public-TLS endpoint; treating as cloud-tier`, {\n      baseUrl: opts.baseUrl,\n    });\n  } else if (classification.trust === 'private') {\n    log(\n      'warn',\n      `[${opts.providerName}] private-network endpoint detected (${classification.reason})`,\n      {\n        baseUrl: opts.baseUrl,\n        acceptsSensitivity: classification.acceptsSensitivity,\n      },\n    );\n  }\n\n  const acceptsSensitivity = opts.acceptsSensitivity ?? classification.acceptsSensitivity;\n  if (\n    opts.acceptsSensitivity !== undefined &&\n    !sameSensitivity(opts.acceptsSensitivity, classification.acceptsSensitivity) &&\n    !KNOWN_LOOPBACK_OVERRIDES_WARNED.has(opts.baseUrl)\n  ) {\n    KNOWN_LOOPBACK_OVERRIDES_WARNED.add(opts.baseUrl);\n    log(\n      'info',\n      `[${opts.providerName}] sensitivity override accepted; default would be [${classification.acceptsSensitivity.join(', ')}]`,\n      {\n        baseUrl: opts.baseUrl,\n      },\n    );\n  }\n\n  const capabilities: ProviderCapabilities = {\n    ...DEFAULT_CAPABILITIES,\n    ...opts.capabilities,\n  };\n  const chatPath = opts.chatPath ?? '/v1/chat/completions';\n  const url = `${stripTrailingSlashes(opts.baseUrl)}${chatPath}`;\n\n  const provider: Provider = {\n    name: opts.providerName,\n    modelId: opts.model,\n    capabilities,\n    acceptsSensitivity,\n    stream(req) {\n      return streamOpenAIShaped(opts, url, applyOpenAIShapedPreflight(req, capabilities));\n    },\n    async generate(req) {\n      return generateOpenAIShaped(opts, url, applyOpenAIShapedPreflight(req, capabilities));\n    },\n  };\n\n  return { provider, classification };\n}\n\nfunction applyOpenAIShapedPreflight(\n  req: ProviderRequest,\n  capabilities: ProviderCapabilities,\n): ProviderRequest {\n  const retention = resolveReasoningRetention({\n    ...(req.reasoningRetention !== undefined ? { requested: req.reasoningRetention } : {}),\n    ...(capabilities.reasoningContract !== undefined\n      ? { contract: capabilities.reasoningContract }\n      : {}),\n  });\n  if (retention === 'pass-through-all') return req;\n  const filtered = applyReasoningPolicy({ messages: req.messages, retention });\n  if (filtered === req.messages) return req;\n  return { ...req, messages: filtered };\n}\n\nasync function* streamOpenAIShaped(\n  opts: OpenAIShapedOptions,\n  url: string,\n  req: ProviderRequest,\n): AsyncIterable<ProviderEvent> {\n  const body = buildBody(opts.model, req, true, opts.capabilities?.structuredOutput ?? true);\n  const resp = await callJsonHttp({\n    providerName: opts.providerName,\n    url,\n    headers: buildHeaders(opts),\n    body,\n    ...(req.signal !== undefined ? { signal: req.signal } : {}),\n    ...(opts.fetchImpl !== undefined ? { fetchImpl: opts.fetchImpl } : {}),\n    ...(opts.timeoutMs !== undefined ? { timeoutMs: opts.timeoutMs } : {}),\n  });\n\n  yield makeStreamStartEvent({ providerName: opts.providerName, modelId: opts.model });\n\n  let finishReason: FinishReason = 'stop';\n  let usage: Usage | undefined;\n  const toolCallBuffer = new Map<\n    number,\n    { id: string; name: string; args: string; emittedStart: boolean }\n  >();\n\n  for await (const payload of parseEventStream(\n    resp.body,\n    req.signal !== undefined ? { signal: req.signal } : {},\n  )) {\n    if (req.signal?.aborted) {\n      // PS-12: report the abort honestly instead of the default 'stop'.\n      finishReason = 'aborted';\n      break;\n    }\n    let parsed: OpenAIChunk;\n    try {\n      parsed = JSON.parse(payload) as OpenAIChunk;\n    } catch (cause) {\n      throw new ProviderStreamParseError(\n        opts.providerName,\n        `failed to parse SSE chunk as JSON: ${(cause as Error).message}`,\n        cause,\n      );\n    }\n    const choice = parsed.choices?.[0];\n    if (choice === undefined) {\n      if (parsed.usage !== undefined) usage = mapOpenAIUsage(parsed.usage);\n      continue;\n    }\n    const delta = choice.delta ?? {};\n    if (typeof delta.content === 'string' && delta.content.length > 0) {\n      yield { type: 'text-delta', delta: delta.content };\n    }\n    if (typeof (delta as { reasoning_content?: unknown }).reasoning_content === 'string') {\n      const r = (delta as { reasoning_content: string }).reasoning_content;\n      if (r.length > 0) yield { type: 'reasoning-delta', delta: r };\n    }\n    if (Array.isArray(delta.tool_calls)) {\n      for (const tc of delta.tool_calls) {\n        const idx = tc.index ?? 0;\n        const slot = toolCallBuffer.get(idx) ?? {\n          id: '',\n          name: '',\n          args: '',\n          emittedStart: false,\n        };\n        if (typeof tc.id === 'string' && tc.id.length > 0) slot.id = tc.id;\n        if (tc.function?.name !== undefined) slot.name = tc.function.name;\n        if (typeof tc.function?.arguments === 'string') slot.args += tc.function.arguments;\n        toolCallBuffer.set(idx, slot);\n        if (!slot.emittedStart && slot.id.length > 0 && slot.name.length > 0) {\n          yield {\n            type: 'tool-call-start',\n            toolCallId: slot.id,\n            toolName: slot.name,\n          };\n          slot.emittedStart = true;\n        }\n        if (typeof tc.function?.arguments === 'string' && tc.function.arguments.length > 0) {\n          yield {\n            type: 'tool-call-input-delta',\n            toolCallId: slot.id,\n            argsDelta: tc.function.arguments,\n          };\n        }\n      }\n    }\n    if (typeof choice.finish_reason === 'string') {\n      finishReason = mapFinishReason(choice.finish_reason);\n      for (const slot of toolCallBuffer.values()) {\n        const finalArgs = parseFinalArgs(slot.args);\n        yield {\n          type: 'tool-call-end',\n          toolCallId: slot.id,\n          finalArgs,\n        };\n      }\n    }\n    if (parsed.usage !== undefined) usage = mapOpenAIUsage(parsed.usage);\n  }\n\n  yield {\n    type: 'finish',\n    finishReason,\n    usage: usage ?? { promptTokens: 0, completionTokens: 0, totalTokens: 0 },\n  };\n}\n\nasync function generateOpenAIShaped(\n  opts: OpenAIShapedOptions,\n  url: string,\n  req: ProviderRequest,\n): Promise<ProviderResponse> {\n  const body = buildBody(opts.model, req, false, opts.capabilities?.structuredOutput ?? true);\n  const resp = await callJsonHttp({\n    providerName: opts.providerName,\n    url,\n    headers: buildHeaders(opts),\n    body,\n    ...(req.signal !== undefined ? { signal: req.signal } : {}),\n    ...(opts.fetchImpl !== undefined ? { fetchImpl: opts.fetchImpl } : {}),\n    ...(opts.timeoutMs !== undefined ? { timeoutMs: opts.timeoutMs } : {}),\n  });\n  let json: OpenAIChunk;\n  try {\n    json = (await resp.json()) as OpenAIChunk;\n  } catch (cause) {\n    throw new ProviderStreamParseError(\n      opts.providerName,\n      'response body was not valid JSON',\n      cause,\n    );\n  }\n  const choice = json.choices?.[0];\n  const message = choice?.message ?? {};\n  const text = typeof message.content === 'string' ? message.content : undefined;\n  const toolCalls = Array.isArray(message.tool_calls)\n    ? message.tool_calls.map((tc) => ({\n        toolCallId: tc.id ?? '',\n        toolName: tc.function?.name ?? '',\n        args: parseFinalArgs(\n          typeof tc.function?.arguments === 'string' ? tc.function.arguments : '',\n        ),\n      }))\n    : undefined;\n  const usage = mapOpenAIUsage(json.usage) ?? {\n    promptTokens: 0,\n    completionTokens: 0,\n    totalTokens: 0,\n  };\n  return {\n    usage,\n    finishReason: mapFinishReason(choice?.finish_reason),\n    ...(text !== undefined ? { text } : {}),\n    ...(toolCalls !== undefined ? { toolCalls } : {}),\n  };\n}\n\nfunction buildBody(\n  model: string,\n  req: ProviderRequest,\n  stream: boolean,\n  structuredOutput: boolean,\n): Record<string, unknown> {\n  const messages =\n    req.systemMessage !== undefined\n      ? [{ role: 'system' as const, content: req.systemMessage }, ...req.messages]\n      : req.messages;\n  const body: Record<string, unknown> = {\n    model,\n    messages: toOpenAIChatMessages(messages),\n    stream,\n  };\n  if (stream) {\n    // core-provider-09: vanilla OpenAI-shaped servers (vLLM, Together,\n    // OpenAI proper) only emit the final usage chunk when asked;\n    // without it streamed calls finish with {0,0,0} usage and cost\n    // tracking silently zeroes. llama.cpp accepts the flag too.\n    // `providerOptions` is merged last, so callers can override it for\n    // servers that reject the field.\n    body.stream_options = { include_usage: true };\n  }\n  if (req.temperature !== undefined) body.temperature = req.temperature;\n  if (req.maxTokens !== undefined) body.max_tokens = req.maxTokens;\n  if (req.tools !== undefined && req.tools.length > 0) {\n    // C2: fold worked examples in the adapter itself (idempotent when an\n    // upstream createProvider fold already ran).\n    body.tools = foldToolExamples(req.tools).map((t) => ({\n      type: 'function',\n      function: {\n        name: t.name,\n        ...(t.description !== undefined ? { description: t.description } : {}),\n        parameters: t.inputSchema,\n      },\n    }));\n  }\n  if (req.toolChoice !== undefined) {\n    body.tool_choice = mapToolChoice(req.toolChoice);\n  }\n  // PS-24: structured output finally reaches the wire - gated on the\n  // declared capability so a structuredOutput:false override keeps the\n  // request clean for servers that reject response_format.\n  if (structuredOutput && req.outputType?.kind === 'structured') {\n    body.response_format =\n      req.outputType.jsonSchema !== undefined\n        ? {\n            type: 'json_schema',\n            json_schema: { name: 'output', schema: req.outputType.jsonSchema, strict: true },\n          }\n        : { type: 'json_object' };\n  }\n  if (req.providerOptions !== undefined) {\n    Object.assign(body, req.providerOptions);\n  }\n  return body;\n}\n\nfunction buildHeaders(opts: OpenAIShapedOptions): Record<string, string> {\n  const headers: Record<string, string> = {\n    'content-type': 'application/json',\n    accept: 'application/json',\n    ...opts.headers,\n  };\n  if (opts.apiKey !== undefined && opts.apiKey.length > 0) {\n    headers.authorization = `Bearer ${opts.apiKey}`;\n  }\n  return headers;\n}\n\nfunction mapToolChoice(value: unknown): unknown {\n  if (value === 'auto' || value === 'none' || value === 'required') return value;\n  if (typeof value === 'object' && value !== null && 'tool' in value) {\n    return { type: 'function', function: { name: (value as { tool: string }).tool } };\n  }\n  return 'auto';\n}\n\nfunction mapFinishReason(value: string | null | undefined): FinishReason {\n  switch (value) {\n    case 'stop':\n    case 'length':\n    case 'content-filter':\n      return value;\n    // PS-12: the OpenAI wire format spells it with an underscore; the dashed\n    // form alone never matched, so content-filtered completions reported 'stop'.\n    case 'content_filter':\n      return 'content-filter';\n    case 'tool_calls':\n    case 'function_call':\n      return 'tool-calls';\n    case null:\n    case undefined:\n      return 'stop';\n    default:\n      return 'stop';\n  }\n}\n\nfunction mapOpenAIUsage(input: unknown): Usage | undefined {\n  if (input === undefined || input === null || typeof input !== 'object') return undefined;\n  const u = input as {\n    prompt_tokens?: number;\n    completion_tokens?: number;\n    total_tokens?: number;\n    reasoning_tokens?: number;\n    // OpenAI reports cache reads as a SUBSET of prompt_tokens\n    // (core-provider-02); there is no cache-write leg on this wire.\n    prompt_tokens_details?: { cached_tokens?: number };\n  };\n  const promptTokens = u.prompt_tokens ?? 0;\n  const completionTokens = u.completion_tokens ?? 0;\n  const totalTokens = u.total_tokens ?? promptTokens + completionTokens;\n  const cachedTokens = u.prompt_tokens_details?.cached_tokens;\n  const usage: Usage = {\n    promptTokens,\n    completionTokens,\n    totalTokens,\n    ...(u.reasoning_tokens !== undefined ? { reasoningTokens: u.reasoning_tokens } : {}),\n    ...(typeof cachedTokens === 'number' && Number.isFinite(cachedTokens) && cachedTokens > 0\n      ? { cachedReadTokens: cachedTokens }\n      : {}),\n  };\n  return usage;\n}\n\nfunction parseFinalArgs(raw: string): unknown {\n  if (raw.length === 0) return {};\n  try {\n    return JSON.parse(raw);\n  } catch {\n    return raw;\n  }\n}\n\nfunction sameSensitivity(a: ReadonlyArray<Sensitivity>, b: ReadonlyArray<Sensitivity>): boolean {\n  if (a.length !== b.length) return false;\n  for (let i = 0; i < a.length; i++) {\n    if (a[i] !== b[i]) return false;\n  }\n  return true;\n}\n\nfunction defaultLogger(level: 'warn' | 'info', message: string, meta?: object): void {\n  const fn = level === 'warn' ? console.warn : console.info;\n  if (meta !== undefined) {\n    fn(`[graphorin/provider] ${message}`, meta);\n  } else {\n    fn(`[graphorin/provider] ${message}`);\n  }\n}\n\ninterface OpenAIChunk {\n  readonly id?: string;\n  readonly choices?: ReadonlyArray<{\n    readonly delta?: {\n      readonly content?: string;\n      readonly tool_calls?: ReadonlyArray<{\n        readonly index?: number;\n        readonly id?: string;\n        readonly type?: string;\n        readonly function?: { readonly name?: string; readonly arguments?: string };\n      }>;\n      readonly reasoning_content?: string;\n    };\n    readonly message?: {\n      readonly content?: string | null;\n      readonly tool_calls?: ReadonlyArray<{\n        readonly id?: string;\n        readonly type?: string;\n        readonly function?: { readonly name?: string; readonly arguments?: string };\n      }>;\n    };\n    readonly finish_reason?: string | null;\n  }>;\n  readonly usage?: {\n    readonly prompt_tokens?: number;\n    readonly completion_tokens?: number;\n    readonly total_tokens?: number;\n    readonly reasoning_tokens?: number;\n    readonly prompt_tokens_details?: { readonly cached_tokens?: number };\n  };\n}\n"],"mappings":";;;;;;;;;;AAoEA,MAAMA,uBAA6C;CACjD,WAAW;CACX,aAAa;CACb,mBAAmB;CACnB,YAAY;CACZ,kBAAkB;CAClB,WAAW;CACX,eAAe;CACf,WAAW;CACX,mBAAmB;CACpB;AAED,MAAM,kDAAkC,IAAI,KAAa;;;;;;;;;AAUzD,SAAgB,0BAA0B,MAGxC;CACA,MAAM,iBAAiB,sBAAsB,KAAK,QAAQ;AAC1D,KAAI,eAAe,UAAU,sBAAsB,KAAK,2BAA2B,KACjF,OAAM,IAAI,oCAAoC,KAAK,QAAQ;CAE7D,MAAM,MAAM,KAAK,UAAU;AAC3B,KAAI,eAAe,UAAU,mBAC3B,KAAI,QAAQ,IAAI,KAAK,aAAa,6CAA6C,KAAK,WAAW,EAC7F,SAAS,KAAK,SACf,CAAC;UACO,eAAe,UAAU,aAClC,KAAI,QAAQ,IAAI,KAAK,aAAa,gDAAgD,EAChF,SAAS,KAAK,SACf,CAAC;UACO,eAAe,UAAU,UAClC,KACE,QACA,IAAI,KAAK,aAAa,uCAAuC,eAAe,OAAO,IACnF;EACE,SAAS,KAAK;EACd,oBAAoB,eAAe;EACpC,CACF;CAGH,MAAM,qBAAqB,KAAK,sBAAsB,eAAe;AACrE,KACE,KAAK,uBAAuB,UAC5B,CAAC,gBAAgB,KAAK,oBAAoB,eAAe,mBAAmB,IAC5E,CAAC,gCAAgC,IAAI,KAAK,QAAQ,EAClD;AACA,kCAAgC,IAAI,KAAK,QAAQ;AACjD,MACE,QACA,IAAI,KAAK,aAAa,qDAAqD,eAAe,mBAAmB,KAAK,KAAK,CAAC,IACxH,EACE,SAAS,KAAK,SACf,CACF;;CAGH,MAAMC,eAAqC;EACzC,GAAG;EACH,GAAG,KAAK;EACT;CACD,MAAM,WAAW,KAAK,YAAY;CAClC,MAAM,MAAM,GAAG,qBAAqB,KAAK,QAAQ,GAAG;AAepD,QAAO;EAAE,UAbkB;GACzB,MAAM,KAAK;GACX,SAAS,KAAK;GACd;GACA;GACA,OAAO,KAAK;AACV,WAAO,mBAAmB,MAAM,KAAK,2BAA2B,KAAK,aAAa,CAAC;;GAErF,MAAM,SAAS,KAAK;AAClB,WAAO,qBAAqB,MAAM,KAAK,2BAA2B,KAAK,aAAa,CAAC;;GAExF;EAEkB;EAAgB;;AAGrC,SAAS,2BACP,KACA,cACiB;CACjB,MAAM,YAAY,0BAA0B;EAC1C,GAAI,IAAI,uBAAuB,SAAY,EAAE,WAAW,IAAI,oBAAoB,GAAG,EAAE;EACrF,GAAI,aAAa,sBAAsB,SACnC,EAAE,UAAU,aAAa,mBAAmB,GAC5C,EAAE;EACP,CAAC;AACF,KAAI,cAAc,mBAAoB,QAAO;CAC7C,MAAM,WAAW,qBAAqB;EAAE,UAAU,IAAI;EAAU;EAAW,CAAC;AAC5E,KAAI,aAAa,IAAI,SAAU,QAAO;AACtC,QAAO;EAAE,GAAG;EAAK,UAAU;EAAU;;AAGvC,gBAAgB,mBACd,MACA,KACA,KAC8B;CAC9B,MAAM,OAAO,UAAU,KAAK,OAAO,KAAK,MAAM,KAAK,cAAc,oBAAoB,KAAK;CAC1F,MAAM,OAAO,MAAM,aAAa;EAC9B,cAAc,KAAK;EACnB;EACA,SAAS,aAAa,KAAK;EAC3B;EACA,GAAI,IAAI,WAAW,SAAY,EAAE,QAAQ,IAAI,QAAQ,GAAG,EAAE;EAC1D,GAAI,KAAK,cAAc,SAAY,EAAE,WAAW,KAAK,WAAW,GAAG,EAAE;EACrE,GAAI,KAAK,cAAc,SAAY,EAAE,WAAW,KAAK,WAAW,GAAG,EAAE;EACtE,CAAC;AAEF,OAAM,qBAAqB;EAAE,cAAc,KAAK;EAAc,SAAS,KAAK;EAAO,CAAC;CAEpF,IAAIC,eAA6B;CACjC,IAAIC;CACJ,MAAM,iCAAiB,IAAI,KAGxB;AAEH,YAAW,MAAM,WAAW,iBAC1B,KAAK,MACL,IAAI,WAAW,SAAY,EAAE,QAAQ,IAAI,QAAQ,GAAG,EAAE,CACvD,EAAE;AACD,MAAI,IAAI,QAAQ,SAAS;AAEvB,kBAAe;AACf;;EAEF,IAAIC;AACJ,MAAI;AACF,YAAS,KAAK,MAAM,QAAQ;WACrB,OAAO;AACd,SAAM,IAAI,yBACR,KAAK,cACL,sCAAuC,MAAgB,WACvD,MACD;;EAEH,MAAM,SAAS,OAAO,UAAU;AAChC,MAAI,WAAW,QAAW;AACxB,OAAI,OAAO,UAAU,OAAW,SAAQ,eAAe,OAAO,MAAM;AACpE;;EAEF,MAAM,QAAQ,OAAO,SAAS,EAAE;AAChC,MAAI,OAAO,MAAM,YAAY,YAAY,MAAM,QAAQ,SAAS,EAC9D,OAAM;GAAE,MAAM;GAAc,OAAO,MAAM;GAAS;AAEpD,MAAI,OAAQ,MAA0C,sBAAsB,UAAU;GACpF,MAAM,IAAK,MAAwC;AACnD,OAAI,EAAE,SAAS,EAAG,OAAM;IAAE,MAAM;IAAmB,OAAO;IAAG;;AAE/D,MAAI,MAAM,QAAQ,MAAM,WAAW,CACjC,MAAK,MAAM,MAAM,MAAM,YAAY;GACjC,MAAM,MAAM,GAAG,SAAS;GACxB,MAAM,OAAO,eAAe,IAAI,IAAI,IAAI;IACtC,IAAI;IACJ,MAAM;IACN,MAAM;IACN,cAAc;IACf;AACD,OAAI,OAAO,GAAG,OAAO,YAAY,GAAG,GAAG,SAAS,EAAG,MAAK,KAAK,GAAG;AAChE,OAAI,GAAG,UAAU,SAAS,OAAW,MAAK,OAAO,GAAG,SAAS;AAC7D,OAAI,OAAO,GAAG,UAAU,cAAc,SAAU,MAAK,QAAQ,GAAG,SAAS;AACzE,kBAAe,IAAI,KAAK,KAAK;AAC7B,OAAI,CAAC,KAAK,gBAAgB,KAAK,GAAG,SAAS,KAAK,KAAK,KAAK,SAAS,GAAG;AACpE,UAAM;KACJ,MAAM;KACN,YAAY,KAAK;KACjB,UAAU,KAAK;KAChB;AACD,SAAK,eAAe;;AAEtB,OAAI,OAAO,GAAG,UAAU,cAAc,YAAY,GAAG,SAAS,UAAU,SAAS,EAC/E,OAAM;IACJ,MAAM;IACN,YAAY,KAAK;IACjB,WAAW,GAAG,SAAS;IACxB;;AAIP,MAAI,OAAO,OAAO,kBAAkB,UAAU;AAC5C,kBAAe,gBAAgB,OAAO,cAAc;AACpD,QAAK,MAAM,QAAQ,eAAe,QAAQ,EAAE;IAC1C,MAAM,YAAY,eAAe,KAAK,KAAK;AAC3C,UAAM;KACJ,MAAM;KACN,YAAY,KAAK;KACjB;KACD;;;AAGL,MAAI,OAAO,UAAU,OAAW,SAAQ,eAAe,OAAO,MAAM;;AAGtE,OAAM;EACJ,MAAM;EACN;EACA,OAAO,SAAS;GAAE,cAAc;GAAG,kBAAkB;GAAG,aAAa;GAAG;EACzE;;AAGH,eAAe,qBACb,MACA,KACA,KAC2B;CAC3B,MAAM,OAAO,UAAU,KAAK,OAAO,KAAK,OAAO,KAAK,cAAc,oBAAoB,KAAK;CAC3F,MAAM,OAAO,MAAM,aAAa;EAC9B,cAAc,KAAK;EACnB;EACA,SAAS,aAAa,KAAK;EAC3B;EACA,GAAI,IAAI,WAAW,SAAY,EAAE,QAAQ,IAAI,QAAQ,GAAG,EAAE;EAC1D,GAAI,KAAK,cAAc,SAAY,EAAE,WAAW,KAAK,WAAW,GAAG,EAAE;EACrE,GAAI,KAAK,cAAc,SAAY,EAAE,WAAW,KAAK,WAAW,GAAG,EAAE;EACtE,CAAC;CACF,IAAIC;AACJ,KAAI;AACF,SAAQ,MAAM,KAAK,MAAM;UAClB,OAAO;AACd,QAAM,IAAI,yBACR,KAAK,cACL,oCACA,MACD;;CAEH,MAAM,SAAS,KAAK,UAAU;CAC9B,MAAM,UAAU,QAAQ,WAAW,EAAE;CACrC,MAAM,OAAO,OAAO,QAAQ,YAAY,WAAW,QAAQ,UAAU;CACrE,MAAM,YAAY,MAAM,QAAQ,QAAQ,WAAW,GAC/C,QAAQ,WAAW,KAAK,QAAQ;EAC9B,YAAY,GAAG,MAAM;EACrB,UAAU,GAAG,UAAU,QAAQ;EAC/B,MAAM,eACJ,OAAO,GAAG,UAAU,cAAc,WAAW,GAAG,SAAS,YAAY,GACtE;EACF,EAAE,GACH;AAMJ,QAAO;EACL,OANY,eAAe,KAAK,MAAM,IAAI;GAC1C,cAAc;GACd,kBAAkB;GAClB,aAAa;GACd;EAGC,cAAc,gBAAgB,QAAQ,cAAc;EACpD,GAAI,SAAS,SAAY,EAAE,MAAM,GAAG,EAAE;EACtC,GAAI,cAAc,SAAY,EAAE,WAAW,GAAG,EAAE;EACjD;;AAGH,SAAS,UACP,OACA,KACA,QACA,kBACyB;CAKzB,MAAMC,OAAgC;EACpC;EACA,UAAU,qBALV,IAAI,kBAAkB,SAClB,CAAC;GAAE,MAAM;GAAmB,SAAS,IAAI;GAAe,EAAE,GAAG,IAAI,SAAS,GAC1E,IAAI,SAGgC;EACxC;EACD;AACD,KAAI,OAOF,MAAK,iBAAiB,EAAE,eAAe,MAAM;AAE/C,KAAI,IAAI,gBAAgB,OAAW,MAAK,cAAc,IAAI;AAC1D,KAAI,IAAI,cAAc,OAAW,MAAK,aAAa,IAAI;AACvD,KAAI,IAAI,UAAU,UAAa,IAAI,MAAM,SAAS,EAGhD,MAAK,QAAQ,iBAAiB,IAAI,MAAM,CAAC,KAAK,OAAO;EACnD,MAAM;EACN,UAAU;GACR,MAAM,EAAE;GACR,GAAI,EAAE,gBAAgB,SAAY,EAAE,aAAa,EAAE,aAAa,GAAG,EAAE;GACrE,YAAY,EAAE;GACf;EACF,EAAE;AAEL,KAAI,IAAI,eAAe,OACrB,MAAK,cAAc,cAAc,IAAI,WAAW;AAKlD,KAAI,oBAAoB,IAAI,YAAY,SAAS,aAC/C,MAAK,kBACH,IAAI,WAAW,eAAe,SAC1B;EACE,MAAM;EACN,aAAa;GAAE,MAAM;GAAU,QAAQ,IAAI,WAAW;GAAY,QAAQ;GAAM;EACjF,GACD,EAAE,MAAM,eAAe;AAE/B,KAAI,IAAI,oBAAoB,OAC1B,QAAO,OAAO,MAAM,IAAI,gBAAgB;AAE1C,QAAO;;AAGT,SAAS,aAAa,MAAmD;CACvE,MAAMC,UAAkC;EACtC,gBAAgB;EAChB,QAAQ;EACR,GAAG,KAAK;EACT;AACD,KAAI,KAAK,WAAW,UAAa,KAAK,OAAO,SAAS,EACpD,SAAQ,gBAAgB,UAAU,KAAK;AAEzC,QAAO;;AAGT,SAAS,cAAc,OAAyB;AAC9C,KAAI,UAAU,UAAU,UAAU,UAAU,UAAU,WAAY,QAAO;AACzE,KAAI,OAAO,UAAU,YAAY,UAAU,QAAQ,UAAU,MAC3D,QAAO;EAAE,MAAM;EAAY,UAAU,EAAE,MAAO,MAA2B,MAAM;EAAE;AAEnF,QAAO;;AAGT,SAAS,gBAAgB,OAAgD;AACvE,SAAQ,OAAR;EACE,KAAK;EACL,KAAK;EACL,KAAK,iBACH,QAAO;EAGT,KAAK,iBACH,QAAO;EACT,KAAK;EACL,KAAK,gBACH,QAAO;EACT,KAAK;EACL,KAAK,OACH,QAAO;EACT,QACE,QAAO;;;AAIb,SAAS,eAAe,OAAmC;AACzD,KAAI,UAAU,UAAa,UAAU,QAAQ,OAAO,UAAU,SAAU,QAAO;CAC/E,MAAM,IAAI;CASV,MAAM,eAAe,EAAE,iBAAiB;CACxC,MAAM,mBAAmB,EAAE,qBAAqB;CAChD,MAAM,cAAc,EAAE,gBAAgB,eAAe;CACrD,MAAM,eAAe,EAAE,uBAAuB;AAU9C,QATqB;EACnB;EACA;EACA;EACA,GAAI,EAAE,qBAAqB,SAAY,EAAE,iBAAiB,EAAE,kBAAkB,GAAG,EAAE;EACnF,GAAI,OAAO,iBAAiB,YAAY,OAAO,SAAS,aAAa,IAAI,eAAe,IACpF,EAAE,kBAAkB,cAAc,GAClC,EAAE;EACP;;AAIH,SAAS,eAAe,KAAsB;AAC5C,KAAI,IAAI,WAAW,EAAG,QAAO,EAAE;AAC/B,KAAI;AACF,SAAO,KAAK,MAAM,IAAI;SAChB;AACN,SAAO;;;AAIX,SAAS,gBAAgB,GAA+B,GAAwC;AAC9F,KAAI,EAAE,WAAW,EAAE,OAAQ,QAAO;AAClC,MAAK,IAAI,IAAI,GAAG,IAAI,EAAE,QAAQ,IAC5B,KAAI,EAAE,OAAO,EAAE,GAAI,QAAO;AAE5B,QAAO;;AAGT,SAAS,cAAc,OAAwB,SAAiB,MAAqB;CACnF,MAAM,KAAK,UAAU,SAAS,QAAQ,OAAO,QAAQ;AACrD,KAAI,SAAS,OACX,IAAG,wBAAwB,WAAW,KAAK;KAE3C,IAAG,wBAAwB,UAAU"}

package/dist/middleware/compose.d.ts

@@ -11,7 +11,7 @@ import { Provider, ProviderMiddleware } from "@graphorin/core";
  */
 declare const MIDDLEWARE_KIND: unique symbol;
 /**
- * Canonical middleware ordering — outermost → innermost. The table
+ * Canonical middleware ordering - outermost → innermost. The table
  * is enforced by {@link composeProviderMiddleware} and is part of the
  * provider layer's public contract (DEC-145 / ADR-039).
  *
@@ -20,8 +20,8 @@ declare const MIDDLEWARE_KIND: unique symbol;
  * Composition is outermost-first, so a request flows top→bottom and a
  * response flows bottom→top:
  *
- * - `withTracing` is outermost so the span wraps everything below —
- *   including retries — and records true end-to-end latency.
+ * - `withTracing` is outermost so the span wraps everything below -
+ *   including retries - and records true end-to-end latency.
  * - `withRetry` sits above the rate/cost limiters so each retry
  *   attempt is independently counted and throttled.
  * - `withRateLimit` → `withCostLimit` → `withCostTracking` form the
@@ -31,7 +31,7 @@ declare const MIDDLEWARE_KIND: unique symbol;
  *   secondary provider still passes through the redactor.
  * - `withRedaction` is **innermost** (closest to the provider) so it
  *   is the last thing to touch the outbound payload and the first to
- *   touch the inbound stream — guaranteeing every retry, fallback, and
+ *   touch the inbound stream - guaranteeing every retry, fallback, and
  *   cost-tracked request sees an already-redacted payload and no
  *   secret can bypass it.
  *
@@ -56,13 +56,13 @@ declare function providerHasMiddleware(provider: Provider, name: string): boolea
 /**
  * Wrap an adapter in a middleware chain whose order is validated
  * against {@link CANONICAL_MIDDLEWARE_ORDER}. The argument array MUST
- * be ordered outermost → innermost — the same way the layers appear
+ * be ordered outermost → innermost - the same way the layers appear
  * in the documented composition example. The composer validates that
  * every kind known to the canonical order is monotonically non-
  * decreasing in index, throws otherwise.
  *
  * Custom middleware whose kind is NOT in the canonical order is
- * silently allowed at any position — operators registering bespoke
+ * silently allowed at any position - operators registering bespoke
  * layers via {@link defineProviderMiddleware} carry the
  * responsibility of placing them sensibly.
  *

package/dist/middleware/compose.js

@@ -10,7 +10,7 @@ import { MiddlewareOrderingError } from "../errors/errors.js";
 */
 const MIDDLEWARE_KIND = Symbol.for("graphorin.provider.middleware.kind");
 /**
-* Canonical middleware ordering — outermost → innermost. The table
+* Canonical middleware ordering - outermost → innermost. The table
 * is enforced by {@link composeProviderMiddleware} and is part of the
 * provider layer's public contract (DEC-145 / ADR-039).
 *
@@ -19,8 +19,8 @@ const MIDDLEWARE_KIND = Symbol.for("graphorin.provider.middleware.kind");
 * Composition is outermost-first, so a request flows top→bottom and a
 * response flows bottom→top:
 *
-* - `withTracing` is outermost so the span wraps everything below —
-*   including retries — and records true end-to-end latency.
+* - `withTracing` is outermost so the span wraps everything below -
+*   including retries - and records true end-to-end latency.
 * - `withRetry` sits above the rate/cost limiters so each retry
 *   attempt is independently counted and throttled.
 * - `withRateLimit` → `withCostLimit` → `withCostTracking` form the
@@ -30,7 +30,7 @@ const MIDDLEWARE_KIND = Symbol.for("graphorin.provider.middleware.kind");
 *   secondary provider still passes through the redactor.
 * - `withRedaction` is **innermost** (closest to the provider) so it
 *   is the last thing to touch the outbound payload and the first to
-*   touch the inbound stream — guaranteeing every retry, fallback, and
+*   touch the inbound stream - guaranteeing every retry, fallback, and
 *   cost-tracked request sees an already-redacted payload and no
 *   secret can bypass it.
 *
@@ -50,7 +50,7 @@ const CANONICAL_INDEX = Object.freeze(CANONICAL_MIDDLEWARE_ORDER.reduce((acc, na
 	return acc;
 }, {}));
 /**
-* Symbol used to walk the chain — every wrapper exposes the inner
+* Symbol used to walk the chain - every wrapper exposes the inner
 * provider so the composer can introspect the full stack at startup.
 *
 * @stable
@@ -94,13 +94,13 @@ function providerHasMiddleware(provider, name) {
 /**
 * Wrap an adapter in a middleware chain whose order is validated
 * against {@link CANONICAL_MIDDLEWARE_ORDER}. The argument array MUST
-* be ordered outermost → innermost — the same way the layers appear
+* be ordered outermost → innermost - the same way the layers appear
 * in the documented composition example. The composer validates that
 * every kind known to the canonical order is monotonically non-
 * decreasing in index, throws otherwise.
 *
 * Custom middleware whose kind is NOT in the canonical order is
-* silently allowed at any position — operators registering bespoke
+* silently allowed at any position - operators registering bespoke
 * layers via {@link defineProviderMiddleware} carry the
 * responsibility of placing them sensibly.
 *

package/dist/middleware/compose.js.map

@@ -1 +1 @@
-{"version":3,"file":"compose.js","names":["MIDDLEWARE_KIND: unique symbol","CANONICAL_MIDDLEWARE_ORDER: readonly string[]","CANONICAL_INDEX: Readonly<Record<string, number>>","INNER_PROVIDER: unique symbol","kinds: string[]","cur: KindedProvider | undefined","chain: Provider","recognised: { kind: string; index: number }[]","branded: Provider"],"sources":["../../src/middleware/compose.ts"],"sourcesContent":["/**\n * Canonical-order middleware composer. The composer enforces the\n * documented ordering at startup and throws\n * {@link MiddlewareOrderingError} on violation. Manual\n * `withRetry(withRedaction(p))` style still works but emits a one-time\n * WARN unless the consumer flips `strict: true`, in which case the\n * composer also throws on manual nesting that violates the canonical\n * order.\n *\n * @packageDocumentation\n */\n\nimport type { Provider, ProviderMiddleware } from '@graphorin/core';\n\nimport { MiddlewareOrderingError } from '../errors/errors.js';\n\n/**\n * Symbol attached to every middleware-produced provider so the\n * composer can detect and validate the chain. The symbol is opaque\n * and cross-realm safe via `Symbol.for`.\n *\n * @stable\n */\nexport const MIDDLEWARE_KIND: unique symbol = Symbol.for('graphorin.provider.middleware.kind');\n\n/**\n * Canonical middleware ordering — outermost → innermost. The table\n * is enforced by {@link composeProviderMiddleware} and is part of the\n * provider layer's public contract (DEC-145 / ADR-039).\n *\n * ## Why this order\n *\n * Composition is outermost-first, so a request flows top→bottom and a\n * response flows bottom→top:\n *\n * - `withTracing` is outermost so the span wraps everything below —\n *   including retries — and records true end-to-end latency.\n * - `withRetry` sits above the rate/cost limiters so each retry\n *   attempt is independently counted and throttled.\n * - `withRateLimit` → `withCostLimit` → `withCostTracking` form the\n *   budget stack: throttle before admitting, reject over-budget calls\n *   before spending, then meter what actually went through.\n * - `withFallback` is just above redaction so a fallback to a\n *   secondary provider still passes through the redactor.\n * - `withRedaction` is **innermost** (closest to the provider) so it\n *   is the last thing to touch the outbound payload and the first to\n *   touch the inbound stream — guaranteeing every retry, fallback, and\n *   cost-tracked request sees an already-redacted payload and no\n *   secret can bypass it.\n *\n * @stable\n */\nexport const CANONICAL_MIDDLEWARE_ORDER: readonly string[] = [\n  'withTracing',\n  'withRetry',\n  'withRateLimit',\n  'withCostLimit',\n  'withCostTracking',\n  'withFallback',\n  'withRedaction',\n];\n\nconst CANONICAL_INDEX: Readonly<Record<string, number>> = Object.freeze(\n  CANONICAL_MIDDLEWARE_ORDER.reduce<Record<string, number>>((acc, name, idx) => {\n    acc[name] = idx;\n    return acc;\n  }, {}),\n);\n\ninterface KindedProvider extends Provider {\n  readonly [MIDDLEWARE_KIND]?: string;\n  readonly [INNER_PROVIDER]?: Provider;\n}\n\n/**\n * Symbol used to walk the chain — every wrapper exposes the inner\n * provider so the composer can introspect the full stack at startup.\n *\n * @stable\n */\nexport const INNER_PROVIDER: unique symbol = Symbol.for('graphorin.provider.middleware.inner');\n\n/**\n * Read the discriminant kind attached to a middleware-produced\n * provider. Returns `undefined` if the provider is the bare adapter\n * or a custom wrapper that does not declare a kind.\n *\n * @stable\n */\nexport function getMiddlewareKind(provider: Provider): string | undefined {\n  return (provider as KindedProvider)[MIDDLEWARE_KIND];\n}\n\n/**\n * Walk the middleware chain inside `provider` and return the array of\n * declared kinds (outer → inner).\n *\n * @stable\n */\nexport function listMiddlewareKinds(provider: Provider): readonly string[] {\n  const kinds: string[] = [];\n  let cur: KindedProvider | undefined = provider as KindedProvider;\n  while (cur !== undefined) {\n    const kind = cur[MIDDLEWARE_KIND];\n    if (kind !== undefined) kinds.push(kind);\n    cur = cur[INNER_PROVIDER] as KindedProvider | undefined;\n  }\n  return kinds;\n}\n\n/**\n * Return `true` iff the chain rooted at `provider` contains a\n * middleware whose kind matches `name`.\n *\n * @stable\n */\nexport function providerHasMiddleware(provider: Provider, name: string): boolean {\n  return listMiddlewareKinds(provider).includes(name);\n}\n\n/**\n * Wrap an adapter in a middleware chain whose order is validated\n * against {@link CANONICAL_MIDDLEWARE_ORDER}. The argument array MUST\n * be ordered outermost → innermost — the same way the layers appear\n * in the documented composition example. The composer validates that\n * every kind known to the canonical order is monotonically non-\n * decreasing in index, throws otherwise.\n *\n * Custom middleware whose kind is NOT in the canonical order is\n * silently allowed at any position — operators registering bespoke\n * layers via {@link defineProviderMiddleware} carry the\n * responsibility of placing them sensibly.\n *\n * @stable\n */\nexport function composeProviderMiddleware(\n  middlewares: ReadonlyArray<ProviderMiddleware>,\n): ProviderMiddleware {\n  return (next: Provider): Provider => {\n    let chain: Provider = next;\n    // Apply innermost → outermost so the composed call order matches\n    // the array's outer → inner declaration.\n    for (let i = middlewares.length - 1; i >= 0; i--) {\n      const factory = middlewares[i];\n      if (typeof factory !== 'function') {\n        throw new TypeError(\n          `composeProviderMiddleware: entry at index ${i} is not a ProviderMiddleware function.`,\n        );\n      }\n      chain = factory(chain);\n    }\n    validateMiddlewareOrder(chain);\n    return chain;\n  };\n}\n\n/**\n * Validate the kinds present in `provider` against the canonical\n * order. Exposed for tests; the composer calls it on every chain.\n *\n * @stable\n */\nexport function validateMiddlewareOrder(provider: Provider): void {\n  const kinds = listMiddlewareKinds(provider);\n  // Filter to the kinds we recognise.\n  const recognised: { kind: string; index: number }[] = [];\n  for (const kind of kinds) {\n    const idx = CANONICAL_INDEX[kind];\n    if (idx !== undefined) recognised.push({ kind, index: idx });\n  }\n  // The recognised list runs outer → inner; canonical indices must\n  // increase monotonically. The first decreasing pair is the\n  // violation.\n  for (let i = 1; i < recognised.length; i++) {\n    const prev = recognised[i - 1];\n    const cur = recognised[i];\n    if (prev !== undefined && cur !== undefined && cur.index < prev.index) {\n      throw new MiddlewareOrderingError({\n        offendingPair: [prev.kind, cur.kind],\n        canonicalOrder: CANONICAL_MIDDLEWARE_ORDER,\n      });\n    }\n  }\n}\n\n/**\n * Decorator factory used internally by every built-in middleware. The\n * returned function attaches the canonical kind discriminator and the\n * inner-provider symbol so the composer can introspect chains.\n *\n * @stable\n */\nexport function defineProviderMiddleware<T>(args: {\n  readonly kind: string;\n  readonly factory: (opts: T) => ProviderMiddleware;\n}): (opts: T) => ProviderMiddleware {\n  return (opts: T): ProviderMiddleware => {\n    const built = args.factory(opts);\n    return (next: Provider): Provider => {\n      const wrapped = built(next);\n      const branded: Provider = Object.create(wrapped) as Provider;\n      Object.defineProperty(branded, MIDDLEWARE_KIND, {\n        value: args.kind,\n        enumerable: false,\n      });\n      Object.defineProperty(branded, INNER_PROVIDER, {\n        value: next,\n        enumerable: false,\n      });\n      return branded;\n    };\n  };\n}\n"],"mappings":";;;;;;;;;;AAuBA,MAAaA,kBAAiC,OAAO,IAAI,qCAAqC;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA6B9F,MAAaC,6BAAgD;CAC3D;CACA;CACA;CACA;CACA;CACA;CACA;CACD;AAED,MAAMC,kBAAoD,OAAO,OAC/D,2BAA2B,QAAgC,KAAK,MAAM,QAAQ;AAC5E,KAAI,QAAQ;AACZ,QAAO;GACN,EAAE,CAAC,CACP;;;;;;;AAaD,MAAaC,iBAAgC,OAAO,IAAI,sCAAsC;;;;;;;;AAS9F,SAAgB,kBAAkB,UAAwC;AACxE,QAAQ,SAA4B;;;;;;;;AAStC,SAAgB,oBAAoB,UAAuC;CACzE,MAAMC,QAAkB,EAAE;CAC1B,IAAIC,MAAkC;AACtC,QAAO,QAAQ,QAAW;EACxB,MAAM,OAAO,IAAI;AACjB,MAAI,SAAS,OAAW,OAAM,KAAK,KAAK;AACxC,QAAM,IAAI;;AAEZ,QAAO;;;;;;;;AAST,SAAgB,sBAAsB,UAAoB,MAAuB;AAC/E,QAAO,oBAAoB,SAAS,CAAC,SAAS,KAAK;;;;;;;;;;;;;;;;;AAkBrD,SAAgB,0BACd,aACoB;AACpB,SAAQ,SAA6B;EACnC,IAAIC,QAAkB;AAGtB,OAAK,IAAI,IAAI,YAAY,SAAS,GAAG,KAAK,GAAG,KAAK;GAChD,MAAM,UAAU,YAAY;AAC5B,OAAI,OAAO,YAAY,WACrB,OAAM,IAAI,UACR,6CAA6C,EAAE,wCAChD;AAEH,WAAQ,QAAQ,MAAM;;AAExB,0BAAwB,MAAM;AAC9B,SAAO;;;;;;;;;AAUX,SAAgB,wBAAwB,UAA0B;CAChE,MAAM,QAAQ,oBAAoB,SAAS;CAE3C,MAAMC,aAAgD,EAAE;AACxD,MAAK,MAAM,QAAQ,OAAO;EACxB,MAAM,MAAM,gBAAgB;AAC5B,MAAI,QAAQ,OAAW,YAAW,KAAK;GAAE;GAAM,OAAO;GAAK,CAAC;;AAK9D,MAAK,IAAI,IAAI,GAAG,IAAI,WAAW,QAAQ,KAAK;EAC1C,MAAM,OAAO,WAAW,IAAI;EAC5B,MAAM,MAAM,WAAW;AACvB,MAAI,SAAS,UAAa,QAAQ,UAAa,IAAI,QAAQ,KAAK,MAC9D,OAAM,IAAI,wBAAwB;GAChC,eAAe,CAAC,KAAK,MAAM,IAAI,KAAK;GACpC,gBAAgB;GACjB,CAAC;;;;;;;;;;AAYR,SAAgB,yBAA4B,MAGR;AAClC,SAAQ,SAAgC;EACtC,MAAM,QAAQ,KAAK,QAAQ,KAAK;AAChC,UAAQ,SAA6B;GACnC,MAAM,UAAU,MAAM,KAAK;GAC3B,MAAMC,UAAoB,OAAO,OAAO,QAAQ;AAChD,UAAO,eAAe,SAAS,iBAAiB;IAC9C,OAAO,KAAK;IACZ,YAAY;IACb,CAAC;AACF,UAAO,eAAe,SAAS,gBAAgB;IAC7C,OAAO;IACP,YAAY;IACb,CAAC;AACF,UAAO"}
+{"version":3,"file":"compose.js","names":["MIDDLEWARE_KIND: unique symbol","CANONICAL_MIDDLEWARE_ORDER: readonly string[]","CANONICAL_INDEX: Readonly<Record<string, number>>","INNER_PROVIDER: unique symbol","kinds: string[]","cur: KindedProvider | undefined","chain: Provider","recognised: { kind: string; index: number }[]","branded: Provider"],"sources":["../../src/middleware/compose.ts"],"sourcesContent":["/**\n * Canonical-order middleware composer. The composer enforces the\n * documented ordering at startup and throws\n * {@link MiddlewareOrderingError} on violation. Manual\n * `withRetry(withRedaction(p))` style still works but emits a one-time\n * WARN unless the consumer flips `strict: true`, in which case the\n * composer also throws on manual nesting that violates the canonical\n * order.\n *\n * @packageDocumentation\n */\n\nimport type { Provider, ProviderMiddleware } from '@graphorin/core';\n\nimport { MiddlewareOrderingError } from '../errors/errors.js';\n\n/**\n * Symbol attached to every middleware-produced provider so the\n * composer can detect and validate the chain. The symbol is opaque\n * and cross-realm safe via `Symbol.for`.\n *\n * @stable\n */\nexport const MIDDLEWARE_KIND: unique symbol = Symbol.for('graphorin.provider.middleware.kind');\n\n/**\n * Canonical middleware ordering - outermost → innermost. The table\n * is enforced by {@link composeProviderMiddleware} and is part of the\n * provider layer's public contract (DEC-145 / ADR-039).\n *\n * ## Why this order\n *\n * Composition is outermost-first, so a request flows top→bottom and a\n * response flows bottom→top:\n *\n * - `withTracing` is outermost so the span wraps everything below -\n *   including retries - and records true end-to-end latency.\n * - `withRetry` sits above the rate/cost limiters so each retry\n *   attempt is independently counted and throttled.\n * - `withRateLimit` → `withCostLimit` → `withCostTracking` form the\n *   budget stack: throttle before admitting, reject over-budget calls\n *   before spending, then meter what actually went through.\n * - `withFallback` is just above redaction so a fallback to a\n *   secondary provider still passes through the redactor.\n * - `withRedaction` is **innermost** (closest to the provider) so it\n *   is the last thing to touch the outbound payload and the first to\n *   touch the inbound stream - guaranteeing every retry, fallback, and\n *   cost-tracked request sees an already-redacted payload and no\n *   secret can bypass it.\n *\n * @stable\n */\nexport const CANONICAL_MIDDLEWARE_ORDER: readonly string[] = [\n  'withTracing',\n  'withRetry',\n  'withRateLimit',\n  'withCostLimit',\n  'withCostTracking',\n  'withFallback',\n  'withRedaction',\n];\n\nconst CANONICAL_INDEX: Readonly<Record<string, number>> = Object.freeze(\n  CANONICAL_MIDDLEWARE_ORDER.reduce<Record<string, number>>((acc, name, idx) => {\n    acc[name] = idx;\n    return acc;\n  }, {}),\n);\n\ninterface KindedProvider extends Provider {\n  readonly [MIDDLEWARE_KIND]?: string;\n  readonly [INNER_PROVIDER]?: Provider;\n}\n\n/**\n * Symbol used to walk the chain - every wrapper exposes the inner\n * provider so the composer can introspect the full stack at startup.\n *\n * @stable\n */\nexport const INNER_PROVIDER: unique symbol = Symbol.for('graphorin.provider.middleware.inner');\n\n/**\n * Read the discriminant kind attached to a middleware-produced\n * provider. Returns `undefined` if the provider is the bare adapter\n * or a custom wrapper that does not declare a kind.\n *\n * @stable\n */\nexport function getMiddlewareKind(provider: Provider): string | undefined {\n  return (provider as KindedProvider)[MIDDLEWARE_KIND];\n}\n\n/**\n * Walk the middleware chain inside `provider` and return the array of\n * declared kinds (outer → inner).\n *\n * @stable\n */\nexport function listMiddlewareKinds(provider: Provider): readonly string[] {\n  const kinds: string[] = [];\n  let cur: KindedProvider | undefined = provider as KindedProvider;\n  while (cur !== undefined) {\n    const kind = cur[MIDDLEWARE_KIND];\n    if (kind !== undefined) kinds.push(kind);\n    cur = cur[INNER_PROVIDER] as KindedProvider | undefined;\n  }\n  return kinds;\n}\n\n/**\n * Return `true` iff the chain rooted at `provider` contains a\n * middleware whose kind matches `name`.\n *\n * @stable\n */\nexport function providerHasMiddleware(provider: Provider, name: string): boolean {\n  return listMiddlewareKinds(provider).includes(name);\n}\n\n/**\n * Wrap an adapter in a middleware chain whose order is validated\n * against {@link CANONICAL_MIDDLEWARE_ORDER}. The argument array MUST\n * be ordered outermost → innermost - the same way the layers appear\n * in the documented composition example. The composer validates that\n * every kind known to the canonical order is monotonically non-\n * decreasing in index, throws otherwise.\n *\n * Custom middleware whose kind is NOT in the canonical order is\n * silently allowed at any position - operators registering bespoke\n * layers via {@link defineProviderMiddleware} carry the\n * responsibility of placing them sensibly.\n *\n * @stable\n */\nexport function composeProviderMiddleware(\n  middlewares: ReadonlyArray<ProviderMiddleware>,\n): ProviderMiddleware {\n  return (next: Provider): Provider => {\n    let chain: Provider = next;\n    // Apply innermost → outermost so the composed call order matches\n    // the array's outer → inner declaration.\n    for (let i = middlewares.length - 1; i >= 0; i--) {\n      const factory = middlewares[i];\n      if (typeof factory !== 'function') {\n        throw new TypeError(\n          `composeProviderMiddleware: entry at index ${i} is not a ProviderMiddleware function.`,\n        );\n      }\n      chain = factory(chain);\n    }\n    validateMiddlewareOrder(chain);\n    return chain;\n  };\n}\n\n/**\n * Validate the kinds present in `provider` against the canonical\n * order. Exposed for tests; the composer calls it on every chain.\n *\n * @stable\n */\nexport function validateMiddlewareOrder(provider: Provider): void {\n  const kinds = listMiddlewareKinds(provider);\n  // Filter to the kinds we recognise.\n  const recognised: { kind: string; index: number }[] = [];\n  for (const kind of kinds) {\n    const idx = CANONICAL_INDEX[kind];\n    if (idx !== undefined) recognised.push({ kind, index: idx });\n  }\n  // The recognised list runs outer → inner; canonical indices must\n  // increase monotonically. The first decreasing pair is the\n  // violation.\n  for (let i = 1; i < recognised.length; i++) {\n    const prev = recognised[i - 1];\n    const cur = recognised[i];\n    if (prev !== undefined && cur !== undefined && cur.index < prev.index) {\n      throw new MiddlewareOrderingError({\n        offendingPair: [prev.kind, cur.kind],\n        canonicalOrder: CANONICAL_MIDDLEWARE_ORDER,\n      });\n    }\n  }\n}\n\n/**\n * Decorator factory used internally by every built-in middleware. The\n * returned function attaches the canonical kind discriminator and the\n * inner-provider symbol so the composer can introspect chains.\n *\n * @stable\n */\nexport function defineProviderMiddleware<T>(args: {\n  readonly kind: string;\n  readonly factory: (opts: T) => ProviderMiddleware;\n}): (opts: T) => ProviderMiddleware {\n  return (opts: T): ProviderMiddleware => {\n    const built = args.factory(opts);\n    return (next: Provider): Provider => {\n      const wrapped = built(next);\n      const branded: Provider = Object.create(wrapped) as Provider;\n      Object.defineProperty(branded, MIDDLEWARE_KIND, {\n        value: args.kind,\n        enumerable: false,\n      });\n      Object.defineProperty(branded, INNER_PROVIDER, {\n        value: next,\n        enumerable: false,\n      });\n      return branded;\n    };\n  };\n}\n"],"mappings":";;;;;;;;;;AAuBA,MAAaA,kBAAiC,OAAO,IAAI,qCAAqC;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA6B9F,MAAaC,6BAAgD;CAC3D;CACA;CACA;CACA;CACA;CACA;CACA;CACD;AAED,MAAMC,kBAAoD,OAAO,OAC/D,2BAA2B,QAAgC,KAAK,MAAM,QAAQ;AAC5E,KAAI,QAAQ;AACZ,QAAO;GACN,EAAE,CAAC,CACP;;;;;;;AAaD,MAAaC,iBAAgC,OAAO,IAAI,sCAAsC;;;;;;;;AAS9F,SAAgB,kBAAkB,UAAwC;AACxE,QAAQ,SAA4B;;;;;;;;AAStC,SAAgB,oBAAoB,UAAuC;CACzE,MAAMC,QAAkB,EAAE;CAC1B,IAAIC,MAAkC;AACtC,QAAO,QAAQ,QAAW;EACxB,MAAM,OAAO,IAAI;AACjB,MAAI,SAAS,OAAW,OAAM,KAAK,KAAK;AACxC,QAAM,IAAI;;AAEZ,QAAO;;;;;;;;AAST,SAAgB,sBAAsB,UAAoB,MAAuB;AAC/E,QAAO,oBAAoB,SAAS,CAAC,SAAS,KAAK;;;;;;;;;;;;;;;;;AAkBrD,SAAgB,0BACd,aACoB;AACpB,SAAQ,SAA6B;EACnC,IAAIC,QAAkB;AAGtB,OAAK,IAAI,IAAI,YAAY,SAAS,GAAG,KAAK,GAAG,KAAK;GAChD,MAAM,UAAU,YAAY;AAC5B,OAAI,OAAO,YAAY,WACrB,OAAM,IAAI,UACR,6CAA6C,EAAE,wCAChD;AAEH,WAAQ,QAAQ,MAAM;;AAExB,0BAAwB,MAAM;AAC9B,SAAO;;;;;;;;;AAUX,SAAgB,wBAAwB,UAA0B;CAChE,MAAM,QAAQ,oBAAoB,SAAS;CAE3C,MAAMC,aAAgD,EAAE;AACxD,MAAK,MAAM,QAAQ,OAAO;EACxB,MAAM,MAAM,gBAAgB;AAC5B,MAAI,QAAQ,OAAW,YAAW,KAAK;GAAE;GAAM,OAAO;GAAK,CAAC;;AAK9D,MAAK,IAAI,IAAI,GAAG,IAAI,WAAW,QAAQ,KAAK;EAC1C,MAAM,OAAO,WAAW,IAAI;EAC5B,MAAM,MAAM,WAAW;AACvB,MAAI,SAAS,UAAa,QAAQ,UAAa,IAAI,QAAQ,KAAK,MAC9D,OAAM,IAAI,wBAAwB;GAChC,eAAe,CAAC,KAAK,MAAM,IAAI,KAAK;GACpC,gBAAgB;GACjB,CAAC;;;;;;;;;;AAYR,SAAgB,yBAA4B,MAGR;AAClC,SAAQ,SAAgC;EACtC,MAAM,QAAQ,KAAK,QAAQ,KAAK;AAChC,UAAQ,SAA6B;GACnC,MAAM,UAAU,MAAM,KAAK;GAC3B,MAAMC,UAAoB,OAAO,OAAO,QAAQ;AAChD,UAAO,eAAe,SAAS,iBAAiB;IAC9C,OAAO,KAAK;IACZ,YAAY;IACb,CAAC;AACF,UAAO,eAAe,SAAS,gBAAgB;IAC7C,OAAO;IACP,YAAY;IACb,CAAC;AACF,UAAO"}

package/dist/middleware/production-hook.js.map

@@ -1 +1 @@
-{"version":3,"file":"production-hook.js","names":[],"sources":["../../src/middleware/production-hook.ts"],"sourcesContent":["/**\n * Production startup hook — fails fast when a security-critical\n * middleware is missing from the composed chain. The default\n * configuration enforces `withRedaction`; consumers can extend the\n * `requiredKinds` list to lock additional middlewares.\n *\n * @packageDocumentation\n */\n\nimport type { Provider } from '@graphorin/core';\n\nimport { MissingProductionMiddlewareError } from '../errors/errors.js';\nimport { providerHasMiddleware } from './compose.js';\n\n/**\n * Options for {@link assertProductionMiddleware}.\n *\n * @stable\n */\nexport interface ProductionStartupHookOptions {\n  /** Middleware kinds that must be present. Defaults to `['withRedaction']`. */\n  readonly requiredKinds?: ReadonlyArray<string>;\n  /** Force the check regardless of `NODE_ENV`. */\n  readonly force?: boolean;\n}\n\n/**\n * Throw {@link MissingProductionMiddlewareError} if a required\n * middleware is missing from the chain rooted at `provider`. The\n * check runs only when `NODE_ENV === 'production'` unless `force` is\n * `true`.\n *\n * @stable\n */\nexport function assertProductionMiddleware(\n  provider: Provider,\n  options: ProductionStartupHookOptions = {},\n): void {\n  const isProd = process.env.NODE_ENV === 'production';\n  if (!isProd && options.force !== true) return;\n  const required = options.requiredKinds ?? ['withRedaction'];\n  for (const kind of required) {\n    if (!providerHasMiddleware(provider, kind)) {\n      throw new MissingProductionMiddlewareError(kind);\n    }\n  }\n}\n"],"mappings":";;;;;;;;;;;;AAkCA,SAAgB,2BACd,UACA,UAAwC,EAAE,EACpC;AAEN,KAAI,EADW,QAAQ,IAAI,aAAa,iBACzB,QAAQ,UAAU,KAAM;CACvC,MAAM,WAAW,QAAQ,iBAAiB,CAAC,gBAAgB;AAC3D,MAAK,MAAM,QAAQ,SACjB,KAAI,CAAC,sBAAsB,UAAU,KAAK,CACxC,OAAM,IAAI,iCAAiC,KAAK"}
+{"version":3,"file":"production-hook.js","names":[],"sources":["../../src/middleware/production-hook.ts"],"sourcesContent":["/**\n * Production startup hook - fails fast when a security-critical\n * middleware is missing from the composed chain. The default\n * configuration enforces `withRedaction`; consumers can extend the\n * `requiredKinds` list to lock additional middlewares.\n *\n * @packageDocumentation\n */\n\nimport type { Provider } from '@graphorin/core';\n\nimport { MissingProductionMiddlewareError } from '../errors/errors.js';\nimport { providerHasMiddleware } from './compose.js';\n\n/**\n * Options for {@link assertProductionMiddleware}.\n *\n * @stable\n */\nexport interface ProductionStartupHookOptions {\n  /** Middleware kinds that must be present. Defaults to `['withRedaction']`. */\n  readonly requiredKinds?: ReadonlyArray<string>;\n  /** Force the check regardless of `NODE_ENV`. */\n  readonly force?: boolean;\n}\n\n/**\n * Throw {@link MissingProductionMiddlewareError} if a required\n * middleware is missing from the chain rooted at `provider`. The\n * check runs only when `NODE_ENV === 'production'` unless `force` is\n * `true`.\n *\n * @stable\n */\nexport function assertProductionMiddleware(\n  provider: Provider,\n  options: ProductionStartupHookOptions = {},\n): void {\n  const isProd = process.env.NODE_ENV === 'production';\n  if (!isProd && options.force !== true) return;\n  const required = options.requiredKinds ?? ['withRedaction'];\n  for (const kind of required) {\n    if (!providerHasMiddleware(provider, kind)) {\n      throw new MissingProductionMiddlewareError(kind);\n    }\n  }\n}\n"],"mappings":";;;;;;;;;;;;AAkCA,SAAgB,2BACd,UACA,UAAwC,EAAE,EACpC;AAEN,KAAI,EADW,QAAQ,IAAI,aAAa,iBACzB,QAAQ,UAAU,KAAM;CACvC,MAAM,WAAW,QAAQ,iBAAiB,CAAC,gBAAgB;AAC3D,MAAK,MAAM,QAAQ,SACjB,KAAI,CAAC,sBAAsB,UAAU,KAAK,CACxC,OAAM,IAAI,iCAAiC,KAAK"}

package/dist/middleware/with-cost-limit.d.ts

@@ -3,7 +3,7 @@ import * as _graphorin_core1 from "@graphorin/core";
 //#region src/middleware/with-cost-limit.d.ts
 
 /**
- * `withCostLimit` — enforce per-session / per-run / per-agent /
+ * `withCostLimit` - enforce per-session / per-run / per-agent /
  * per-hour cost ceilings. Couples with {@link withCostTracking} for
  * the underlying accumulator. The middleware is positioned between
  * `withRateLimit` and `withCostTracking` per the canonical order so a

package/dist/middleware/with-cost-limit.js.map

@@ -1 +1 @@
-{"version":3,"file":"with-cost-limit.js","names":["checks: Array<['session' | 'run' | 'hour', number | undefined]>"],"sources":["../../src/middleware/with-cost-limit.ts"],"sourcesContent":["/**\n * `withCostLimit` — enforce per-session / per-run / per-agent /\n * per-hour cost ceilings. Couples with {@link withCostTracking} for\n * the underlying accumulator. The middleware is positioned between\n * `withRateLimit` and `withCostTracking` per the canonical order so a\n * budget breach trips before the request hits the underlying\n * provider.\n *\n * @packageDocumentation\n */\n\nimport type { Provider } from '@graphorin/core';\n\nimport { CostBudgetExceededError } from '../errors/errors.js';\nimport { defineProviderMiddleware } from './compose.js';\n\n/**\n * Options for {@link withCostLimit}.\n *\n * @stable\n */\nexport interface WithCostLimitOptions {\n  /** Maximum cumulative USD cost per session. */\n  readonly maxPerSession?: number;\n  /** Maximum cumulative USD cost per run. */\n  readonly maxPerRun?: number;\n  /** Maximum cumulative USD cost per hour. */\n  readonly maxPerHour?: number;\n  /** What to do on breach. Default `'throw'`. */\n  readonly onExceed?: 'throw' | 'warn';\n  /**\n   * Resolver returning the current observed cost for the relevant\n   * scope. The resolver lets consumers wire any accumulator (the\n   * shipped `@graphorin/observability/cost.CostTracker` works out of\n   * the box). When unset, the middleware is a no-op (a placeholder\n   * for tooling that wires the accumulator later).\n   */\n  readonly resolveObservedCost?: (\n    scope: 'session' | 'run' | 'hour',\n    metadata: Readonly<Record<string, unknown>> | undefined,\n  ) => number;\n  /** Optional sink for `'warn'` mode. Defaults to `console.warn`. */\n  readonly logger?: (message: string, meta?: object) => void;\n}\n\n/**\n * @stable\n */\nexport const withCostLimit = defineProviderMiddleware<WithCostLimitOptions>({\n  kind: 'withCostLimit',\n  factory: (opts: WithCostLimitOptions) => {\n    const onExceed = opts.onExceed ?? 'throw';\n    const resolver = opts.resolveObservedCost;\n    const logger = opts.logger ?? defaultLogger;\n    // PS-8: a ceiling without a resolver is a silent no-op — the limits look\n    // enforced but never trip. Warn loudly so the gap is visible. (No ceiling\n    // + no resolver is the documented inert placeholder, so it stays quiet.)\n    const hasCeiling =\n      opts.maxPerSession !== undefined ||\n      opts.maxPerRun !== undefined ||\n      opts.maxPerHour !== undefined;\n    if (hasCeiling && resolver === undefined) {\n      logger(\n        '[graphorin/provider] withCostLimit: a cost ceiling is configured but no ' +\n          '`resolveObservedCost` was supplied, so the limit is UNENFORCED (no-op). ' +\n          'Wire a resolver (e.g. from createCostAccumulator / @graphorin/observability).',\n      );\n    }\n    return (next: Provider): Provider => ({\n      name: next.name,\n      modelId: next.modelId,\n      capabilities: next.capabilities,\n      ...(next.acceptsSensitivity !== undefined\n        ? { acceptsSensitivity: next.acceptsSensitivity }\n        : {}),\n      stream(req) {\n        if (resolver !== undefined) {\n          check(asRecord(req.metadata), resolver, opts, onExceed, logger);\n        }\n        return next.stream(req);\n      },\n      async generate(req) {\n        if (resolver !== undefined) {\n          check(asRecord(req.metadata), resolver, opts, onExceed, logger);\n        }\n        return next.generate(req);\n      },\n      ...(next.countTokens ? { countTokens: next.countTokens.bind(next) } : {}),\n    });\n  },\n});\n\nfunction check(\n  metadata: Readonly<Record<string, unknown>> | undefined,\n  resolver: (\n    scope: 'session' | 'run' | 'hour',\n    metadata?: Readonly<Record<string, unknown>>,\n  ) => number,\n  opts: WithCostLimitOptions,\n  onExceed: 'throw' | 'warn',\n  logger: (message: string, meta?: object) => void,\n): void {\n  const checks: Array<['session' | 'run' | 'hour', number | undefined]> = [\n    ['session', opts.maxPerSession],\n    ['run', opts.maxPerRun],\n    ['hour', opts.maxPerHour],\n  ];\n  for (const [scope, limit] of checks) {\n    if (limit === undefined) continue;\n    const observed = resolver(scope, metadata);\n    if (observed > limit) {\n      if (onExceed === 'throw') {\n        throw new CostBudgetExceededError({ scope, limit, observed });\n      }\n      logger(\n        `[graphorin/provider] withCostLimit: ${scope} budget breach (${observed.toFixed(4)} > ${limit.toFixed(4)} USD).`,\n        { scope, limit, observed },\n      );\n    }\n  }\n}\n\nfunction defaultLogger(message: string, meta?: object): void {\n  if (meta !== undefined) console.warn(message, meta);\n  else console.warn(message);\n}\n\nfunction asRecord(value: unknown): Readonly<Record<string, unknown>> | undefined {\n  if (value === undefined || value === null) return undefined;\n  return value as Readonly<Record<string, unknown>>;\n}\n"],"mappings":";;;;;;;AAgDA,MAAa,gBAAgB,yBAA+C;CAC1E,MAAM;CACN,UAAU,SAA+B;EACvC,MAAM,WAAW,KAAK,YAAY;EAClC,MAAM,WAAW,KAAK;EACtB,MAAM,SAAS,KAAK,UAAU;AAQ9B,OAHE,KAAK,kBAAkB,UACvB,KAAK,cAAc,UACnB,KAAK,eAAe,WACJ,aAAa,OAC7B,QACE,gOAGD;AAEH,UAAQ,UAA8B;GACpC,MAAM,KAAK;GACX,SAAS,KAAK;GACd,cAAc,KAAK;GACnB,GAAI,KAAK,uBAAuB,SAC5B,EAAE,oBAAoB,KAAK,oBAAoB,GAC/C,EAAE;GACN,OAAO,KAAK;AACV,QAAI,aAAa,OACf,OAAM,SAAS,IAAI,SAAS,EAAE,UAAU,MAAM,UAAU,OAAO;AAEjE,WAAO,KAAK,OAAO,IAAI;;GAEzB,MAAM,SAAS,KAAK;AAClB,QAAI,aAAa,OACf,OAAM,SAAS,IAAI,SAAS,EAAE,UAAU,MAAM,UAAU,OAAO;AAEjE,WAAO,KAAK,SAAS,IAAI;;GAE3B,GAAI,KAAK,cAAc,EAAE,aAAa,KAAK,YAAY,KAAK,KAAK,EAAE,GAAG,EAAE;GACzE;;CAEJ,CAAC;AAEF,SAAS,MACP,UACA,UAIA,MACA,UACA,QACM;CACN,MAAMA,SAAkE;EACtE,CAAC,WAAW,KAAK,cAAc;EAC/B,CAAC,OAAO,KAAK,UAAU;EACvB,CAAC,QAAQ,KAAK,WAAW;EAC1B;AACD,MAAK,MAAM,CAAC,OAAO,UAAU,QAAQ;AACnC,MAAI,UAAU,OAAW;EACzB,MAAM,WAAW,SAAS,OAAO,SAAS;AAC1C,MAAI,WAAW,OAAO;AACpB,OAAI,aAAa,QACf,OAAM,IAAI,wBAAwB;IAAE;IAAO;IAAO;IAAU,CAAC;AAE/D,UACE,uCAAuC,MAAM,kBAAkB,SAAS,QAAQ,EAAE,CAAC,KAAK,MAAM,QAAQ,EAAE,CAAC,SACzG;IAAE;IAAO;IAAO;IAAU,CAC3B;;;;AAKP,SAAS,cAAc,SAAiB,MAAqB;AAC3D,KAAI,SAAS,OAAW,SAAQ,KAAK,SAAS,KAAK;KAC9C,SAAQ,KAAK,QAAQ;;AAG5B,SAAS,SAAS,OAA+D;AAC/E,KAAI,UAAU,UAAa,UAAU,KAAM,QAAO;AAClD,QAAO"}
+{"version":3,"file":"with-cost-limit.js","names":["checks: Array<['session' | 'run' | 'hour', number | undefined]>"],"sources":["../../src/middleware/with-cost-limit.ts"],"sourcesContent":["/**\n * `withCostLimit` - enforce per-session / per-run / per-agent /\n * per-hour cost ceilings. Couples with {@link withCostTracking} for\n * the underlying accumulator. The middleware is positioned between\n * `withRateLimit` and `withCostTracking` per the canonical order so a\n * budget breach trips before the request hits the underlying\n * provider.\n *\n * @packageDocumentation\n */\n\nimport type { Provider } from '@graphorin/core';\n\nimport { CostBudgetExceededError } from '../errors/errors.js';\nimport { defineProviderMiddleware } from './compose.js';\n\n/**\n * Options for {@link withCostLimit}.\n *\n * @stable\n */\nexport interface WithCostLimitOptions {\n  /** Maximum cumulative USD cost per session. */\n  readonly maxPerSession?: number;\n  /** Maximum cumulative USD cost per run. */\n  readonly maxPerRun?: number;\n  /** Maximum cumulative USD cost per hour. */\n  readonly maxPerHour?: number;\n  /** What to do on breach. Default `'throw'`. */\n  readonly onExceed?: 'throw' | 'warn';\n  /**\n   * Resolver returning the current observed cost for the relevant\n   * scope. The resolver lets consumers wire any accumulator (the\n   * shipped `@graphorin/observability/cost.CostTracker` works out of\n   * the box). When unset, the middleware is a no-op (a placeholder\n   * for tooling that wires the accumulator later).\n   */\n  readonly resolveObservedCost?: (\n    scope: 'session' | 'run' | 'hour',\n    metadata: Readonly<Record<string, unknown>> | undefined,\n  ) => number;\n  /** Optional sink for `'warn'` mode. Defaults to `console.warn`. */\n  readonly logger?: (message: string, meta?: object) => void;\n}\n\n/**\n * @stable\n */\nexport const withCostLimit = defineProviderMiddleware<WithCostLimitOptions>({\n  kind: 'withCostLimit',\n  factory: (opts: WithCostLimitOptions) => {\n    const onExceed = opts.onExceed ?? 'throw';\n    const resolver = opts.resolveObservedCost;\n    const logger = opts.logger ?? defaultLogger;\n    // PS-8: a ceiling without a resolver is a silent no-op - the limits look\n    // enforced but never trip. Warn loudly so the gap is visible. (No ceiling\n    // + no resolver is the documented inert placeholder, so it stays quiet.)\n    const hasCeiling =\n      opts.maxPerSession !== undefined ||\n      opts.maxPerRun !== undefined ||\n      opts.maxPerHour !== undefined;\n    if (hasCeiling && resolver === undefined) {\n      logger(\n        '[graphorin/provider] withCostLimit: a cost ceiling is configured but no ' +\n          '`resolveObservedCost` was supplied, so the limit is UNENFORCED (no-op). ' +\n          'Wire a resolver (e.g. from createCostAccumulator / @graphorin/observability).',\n      );\n    }\n    return (next: Provider): Provider => ({\n      name: next.name,\n      modelId: next.modelId,\n      capabilities: next.capabilities,\n      ...(next.acceptsSensitivity !== undefined\n        ? { acceptsSensitivity: next.acceptsSensitivity }\n        : {}),\n      stream(req) {\n        if (resolver !== undefined) {\n          check(asRecord(req.metadata), resolver, opts, onExceed, logger);\n        }\n        return next.stream(req);\n      },\n      async generate(req) {\n        if (resolver !== undefined) {\n          check(asRecord(req.metadata), resolver, opts, onExceed, logger);\n        }\n        return next.generate(req);\n      },\n      ...(next.countTokens ? { countTokens: next.countTokens.bind(next) } : {}),\n    });\n  },\n});\n\nfunction check(\n  metadata: Readonly<Record<string, unknown>> | undefined,\n  resolver: (\n    scope: 'session' | 'run' | 'hour',\n    metadata?: Readonly<Record<string, unknown>>,\n  ) => number,\n  opts: WithCostLimitOptions,\n  onExceed: 'throw' | 'warn',\n  logger: (message: string, meta?: object) => void,\n): void {\n  const checks: Array<['session' | 'run' | 'hour', number | undefined]> = [\n    ['session', opts.maxPerSession],\n    ['run', opts.maxPerRun],\n    ['hour', opts.maxPerHour],\n  ];\n  for (const [scope, limit] of checks) {\n    if (limit === undefined) continue;\n    const observed = resolver(scope, metadata);\n    if (observed > limit) {\n      if (onExceed === 'throw') {\n        throw new CostBudgetExceededError({ scope, limit, observed });\n      }\n      logger(\n        `[graphorin/provider] withCostLimit: ${scope} budget breach (${observed.toFixed(4)} > ${limit.toFixed(4)} USD).`,\n        { scope, limit, observed },\n      );\n    }\n  }\n}\n\nfunction defaultLogger(message: string, meta?: object): void {\n  if (meta !== undefined) console.warn(message, meta);\n  else console.warn(message);\n}\n\nfunction asRecord(value: unknown): Readonly<Record<string, unknown>> | undefined {\n  if (value === undefined || value === null) return undefined;\n  return value as Readonly<Record<string, unknown>>;\n}\n"],"mappings":";;;;;;;AAgDA,MAAa,gBAAgB,yBAA+C;CAC1E,MAAM;CACN,UAAU,SAA+B;EACvC,MAAM,WAAW,KAAK,YAAY;EAClC,MAAM,WAAW,KAAK;EACtB,MAAM,SAAS,KAAK,UAAU;AAQ9B,OAHE,KAAK,kBAAkB,UACvB,KAAK,cAAc,UACnB,KAAK,eAAe,WACJ,aAAa,OAC7B,QACE,gOAGD;AAEH,UAAQ,UAA8B;GACpC,MAAM,KAAK;GACX,SAAS,KAAK;GACd,cAAc,KAAK;GACnB,GAAI,KAAK,uBAAuB,SAC5B,EAAE,oBAAoB,KAAK,oBAAoB,GAC/C,EAAE;GACN,OAAO,KAAK;AACV,QAAI,aAAa,OACf,OAAM,SAAS,IAAI,SAAS,EAAE,UAAU,MAAM,UAAU,OAAO;AAEjE,WAAO,KAAK,OAAO,IAAI;;GAEzB,MAAM,SAAS,KAAK;AAClB,QAAI,aAAa,OACf,OAAM,SAAS,IAAI,SAAS,EAAE,UAAU,MAAM,UAAU,OAAO;AAEjE,WAAO,KAAK,SAAS,IAAI;;GAE3B,GAAI,KAAK,cAAc,EAAE,aAAa,KAAK,YAAY,KAAK,KAAK,EAAE,GAAG,EAAE;GACzE;;CAEJ,CAAC;AAEF,SAAS,MACP,UACA,UAIA,MACA,UACA,QACM;CACN,MAAMA,SAAkE;EACtE,CAAC,WAAW,KAAK,cAAc;EAC/B,CAAC,OAAO,KAAK,UAAU;EACvB,CAAC,QAAQ,KAAK,WAAW;EAC1B;AACD,MAAK,MAAM,CAAC,OAAO,UAAU,QAAQ;AACnC,MAAI,UAAU,OAAW;EACzB,MAAM,WAAW,SAAS,OAAO,SAAS;AAC1C,MAAI,WAAW,OAAO;AACpB,OAAI,aAAa,QACf,OAAM,IAAI,wBAAwB;IAAE;IAAO;IAAO;IAAU,CAAC;AAE/D,UACE,uCAAuC,MAAM,kBAAkB,SAAS,QAAQ,EAAE,CAAC,KAAK,MAAM,QAAQ,EAAE,CAAC,SACzG;IAAE;IAAO;IAAO;IAAU,CAC3B;;;;AAKP,SAAS,cAAc,SAAiB,MAAqB;AAC3D,KAAI,SAAS,OAAW,SAAQ,KAAK,SAAS,KAAK;KAC9C,SAAQ,KAAK,QAAQ;;AAG5B,SAAS,SAAS,OAA+D;AAC/E,KAAI,UAAU,UAAa,UAAU,KAAM,QAAO;AAClD,QAAO"}

package/dist/middleware/with-cost-tracking.d.ts

@@ -3,7 +3,7 @@ import * as _graphorin_core2 from "@graphorin/core";
 //#region src/middleware/with-cost-tracking.d.ts
 
 /**
- * `withCostTracking` — reports `tokensUsed` + `cost` per `provider × model`
+ * `withCostTracking` - reports `tokensUsed` + `cost` per `provider × model`
  * from the `finish` event of every stream / one-shot generation via the
  * `onUsage` hook. For a ready-made process-local accumulator, wire
  * {@link createCostAccumulator}'s `onUsage` and read its `.totals()`; consumers
@@ -23,12 +23,16 @@ interface CostTrackingTotals {
   readonly promptTokens: number;
   readonly completionTokens: number;
   readonly totalTokens: number;
+  /** Prompt tokens served from the provider cache (subset of promptTokens). */
+  readonly cachedReadTokens: number;
+  /** Prompt tokens written to the provider cache (subset of promptTokens). */
+  readonly cacheWriteTokens: number;
   readonly costUsd: number;
 }
 /**
  * A process-local cost accumulator (PS-8). Wire {@link CostAccumulator.onUsage}
- * into {@link withCostTracking} and read the running totals — keyed by
- * `provider × model` — back via {@link CostAccumulator.totals} /
+ * into {@link withCostTracking} and read the running totals - keyed by
+ * `provider × model` - back via {@link CostAccumulator.totals} /
  * {@link CostAccumulator.totalFor}.
  *
  * @stable
@@ -44,7 +48,7 @@ interface CostAccumulator {
   reset(): void;
 }
 /**
- * Create a {@link CostAccumulator} — the process-local accumulator described on
+ * Create a {@link CostAccumulator} - the process-local accumulator described on
  * {@link withCostTracking}. Keys totals by `'<providerName>::<modelId>'`.
  *
  * @stable
@@ -67,12 +71,19 @@ interface WithCostTrackingOptions {
     readonly promptTokens: number;
     readonly completionTokens: number;
     readonly totalTokens: number;
+    readonly cachedReadTokens?: number;
+    readonly cacheWriteTokens?: number;
     readonly costUsd: number;
     readonly metadata: Readonly<Record<string, unknown>> | undefined;
   }) => void;
   /**
    * Optional pricing lookup. When set, the middleware computes
    * `costUsd` from the returned price and surfaces it on the hook.
+   *
+   * `cachedReadPerMtok` / `cacheWritePerMtok` price the prompt-cache legs
+   * (core-provider-02); when omitted, cache tokens are billed at the full
+   * input rate (never cheaper than reality, so absent price data degrades
+   * to the pre-cache behaviour).
    */
   readonly priceLookup?: (info: {
     readonly providerName: string;
@@ -80,6 +91,8 @@ interface WithCostTrackingOptions {
   }) => {
     readonly inputPerMtok?: number;
     readonly outputPerMtok?: number;
+    readonly cachedReadPerMtok?: number;
+    readonly cacheWritePerMtok?: number;
   } | null;
 }
 /**

package/dist/middleware/with-cost-tracking.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"with-cost-tracking.d.ts","names":[],"sources":["../../src/middleware/with-cost-tracking.ts"],"sourcesContent":[],"mappings":";;;;;;;AAqBA;AAgBA;;;;;;;AAyBA;AA0BA;AA4BA;;;UA/FiB,kBAAA;;;;;;;;;;;;;;;UAgBA,eAAA;;oBAEG,YAAY;;YAEpB,oBAAoB;;mDAEmB;;;;;;;;;;iBAmBnC,qBAAA,CAAA,GAAyB;;;;;;UA0BxB,uBAAA;;;;;;;;;;;;;uBAaM,SAAS;;;;;;;;;;;;;;;;;cAenB,yBAAgB,4BAAA,gBAAA,CAAA"}
+{"version":3,"file":"with-cost-tracking.d.ts","names":[],"sources":["../../src/middleware/with-cost-tracking.ts"],"sourcesContent":[],"mappings":";;;;;;;AAqBA;AAoBA;;;;;;;AA2BA;AA4BA;AAqCA;;;UAhHiB,kBAAA;;;;;;;;;;;;;;;;;;;UAoBA,eAAA;;oBAEG,YAAY;;YAEpB,oBAAoB;;mDAEmB;;;;;;;;;;iBAqBnC,qBAAA,CAAA,GAAyB;;;;;;UA4BxB,uBAAA;;;;;;;;;;;;;;;uBAeM,SAAS;;;;;;;;;;;;;;;;;;;;;;;;cAsBnB,yBAAgB,4BAAA,gBAAA,CAAA"}

package/dist/middleware/with-cost-tracking.js

@@ -6,10 +6,12 @@ const ZERO_TOTALS = Object.freeze({
 	promptTokens: 0,
 	completionTokens: 0,
 	totalTokens: 0,
+	cachedReadTokens: 0,
+	cacheWriteTokens: 0,
 	costUsd: 0
 });
 /**
-* Create a {@link CostAccumulator} — the process-local accumulator described on
+* Create a {@link CostAccumulator} - the process-local accumulator described on
 * {@link withCostTracking}. Keys totals by `'<providerName>::<modelId>'`.
 *
 * @stable
@@ -26,6 +28,8 @@ function createCostAccumulator() {
 				promptTokens: prev.promptTokens + info.promptTokens,
 				completionTokens: prev.completionTokens + info.completionTokens,
 				totalTokens: prev.totalTokens + info.totalTokens,
+				cachedReadTokens: prev.cachedReadTokens + (info.cachedReadTokens ?? 0),
+				cacheWriteTokens: prev.cacheWriteTokens + (info.cacheWriteTokens ?? 0),
 				costUsd: prev.costUsd + info.costUsd
 			});
 		},
@@ -78,13 +82,19 @@ function emitUsage(result, next, opts, metadata) {
 		providerName: next.name,
 		modelId: next.modelId
 	}) ?? null;
-	const costUsd = price !== null ? usage.promptTokens * (price.inputPerMtok ?? 0) / 1e6 + usage.completionTokens * (price.outputPerMtok ?? 0) / 1e6 : usage.cost?.amount ?? 0;
+	const cachedRead = usage.cachedReadTokens ?? 0;
+	const cacheWrite = usage.cacheWriteTokens ?? 0;
+	const basePromptTokens = Math.max(0, usage.promptTokens - cachedRead - cacheWrite);
+	const inputRate = price?.inputPerMtok ?? 0;
+	const costUsd = price !== null ? (basePromptTokens * inputRate + cachedRead * (price.cachedReadPerMtok ?? inputRate) + cacheWrite * (price.cacheWritePerMtok ?? inputRate) + (usage.completionTokens + (usage.reasoningTokens ?? 0)) * (price.outputPerMtok ?? 0)) / 1e6 : usage.cost?.amount ?? 0;
 	opts.onUsage({
 		providerName: next.name,
 		modelId: next.modelId,
 		promptTokens: usage.promptTokens,
 		completionTokens: usage.completionTokens,
 		totalTokens: usage.totalTokens,
+		...usage.cachedReadTokens !== void 0 ? { cachedReadTokens: usage.cachedReadTokens } : {},
+		...usage.cacheWriteTokens !== void 0 ? { cacheWriteTokens: usage.cacheWriteTokens } : {},
 		costUsd,
 		metadata
 	});

package/dist/middleware/with-cost-tracking.js.map

@@ -1 +1 @@
-{"version":3,"file":"with-cost-tracking.js","names":["ZERO_TOTALS: CostTrackingTotals"],"sources":["../../src/middleware/with-cost-tracking.ts"],"sourcesContent":["/**\n * `withCostTracking` — reports `tokensUsed` + `cost` per `provider × model`\n * from the `finish` event of every stream / one-shot generation via the\n * `onUsage` hook. For a ready-made process-local accumulator, wire\n * {@link createCostAccumulator}'s `onUsage` and read its `.totals()`; consumers\n * that prefer the framework's `CostTracker` from `@graphorin/observability`\n * pass their own delegate.\n *\n * @packageDocumentation\n */\n\nimport type { Provider, ProviderEvent, ProviderResponse } from '@graphorin/core';\n\nimport { defineProviderMiddleware } from './compose.js';\n\n/**\n * Aggregated totals for one `provider × model`, returned by\n * {@link CostAccumulator.totalFor} / {@link CostAccumulator.totals}.\n *\n * @stable\n */\nexport interface CostTrackingTotals {\n  readonly callCount: number;\n  readonly promptTokens: number;\n  readonly completionTokens: number;\n  readonly totalTokens: number;\n  readonly costUsd: number;\n}\n\n/**\n * A process-local cost accumulator (PS-8). Wire {@link CostAccumulator.onUsage}\n * into {@link withCostTracking} and read the running totals — keyed by\n * `provider × model` — back via {@link CostAccumulator.totals} /\n * {@link CostAccumulator.totalFor}.\n *\n * @stable\n */\nexport interface CostAccumulator {\n  /** Pass this to {@link withCostTracking}'s `onUsage`. */\n  readonly onUsage: NonNullable<WithCostTrackingOptions['onUsage']>;\n  /** Snapshot of every tracked `provider::model` → totals. */\n  totals(): ReadonlyMap<string, CostTrackingTotals>;\n  /** Running totals for one `provider × model` (zeros when unseen). */\n  totalFor(providerName: string, modelId: string): CostTrackingTotals;\n  /** Clear all accumulated totals. */\n  reset(): void;\n}\n\nconst ZERO_TOTALS: CostTrackingTotals = Object.freeze({\n  callCount: 0,\n  promptTokens: 0,\n  completionTokens: 0,\n  totalTokens: 0,\n  costUsd: 0,\n});\n\n/**\n * Create a {@link CostAccumulator} — the process-local accumulator described on\n * {@link withCostTracking}. Keys totals by `'<providerName>::<modelId>'`.\n *\n * @stable\n */\nexport function createCostAccumulator(): CostAccumulator {\n  const map = new Map<string, CostTrackingTotals>();\n  const keyOf = (providerName: string, modelId: string): string => `${providerName}::${modelId}`;\n  return {\n    onUsage: (info) => {\n      const key = keyOf(info.providerName, info.modelId);\n      const prev = map.get(key) ?? ZERO_TOTALS;\n      map.set(key, {\n        callCount: prev.callCount + 1,\n        promptTokens: prev.promptTokens + info.promptTokens,\n        completionTokens: prev.completionTokens + info.completionTokens,\n        totalTokens: prev.totalTokens + info.totalTokens,\n        costUsd: prev.costUsd + info.costUsd,\n      });\n    },\n    totals: () => new Map(map),\n    totalFor: (providerName, modelId) => map.get(keyOf(providerName, modelId)) ?? ZERO_TOTALS,\n    reset: () => map.clear(),\n  };\n}\n\n/**\n * Options for {@link withCostTracking}.\n *\n * @stable\n */\nexport interface WithCostTrackingOptions {\n  /**\n   * Hook fired on every `finish` event with the parsed usage. The\n   * hook receives the underlying provider's name + modelId so the\n   * caller can route into a per-model accumulator.\n   */\n  readonly onUsage?: (info: {\n    readonly providerName: string;\n    readonly modelId: string;\n    readonly promptTokens: number;\n    readonly completionTokens: number;\n    readonly totalTokens: number;\n    readonly costUsd: number;\n    readonly metadata: Readonly<Record<string, unknown>> | undefined;\n  }) => void;\n  /**\n   * Optional pricing lookup. When set, the middleware computes\n   * `costUsd` from the returned price and surfaces it on the hook.\n   */\n  readonly priceLookup?: (info: {\n    readonly providerName: string;\n    readonly modelId: string;\n  }) => { readonly inputPerMtok?: number; readonly outputPerMtok?: number } | null;\n}\n\n/**\n * @stable\n */\nexport const withCostTracking = defineProviderMiddleware<WithCostTrackingOptions>({\n  kind: 'withCostTracking',\n  factory: (opts: WithCostTrackingOptions) => {\n    return (next: Provider): Provider => ({\n      name: next.name,\n      modelId: next.modelId,\n      capabilities: next.capabilities,\n      ...(next.acceptsSensitivity !== undefined\n        ? { acceptsSensitivity: next.acceptsSensitivity }\n        : {}),\n      stream(req) {\n        return trackingStream(next, req, opts);\n      },\n      async generate(req) {\n        const result = await next.generate(req);\n        emitUsage(result, next, opts, asRecord(req.metadata));\n        return result;\n      },\n      ...(next.countTokens ? { countTokens: next.countTokens.bind(next) } : {}),\n    });\n  },\n});\n\nasync function* trackingStream(\n  next: Provider,\n  req: import('@graphorin/core').ProviderRequest,\n  opts: WithCostTrackingOptions,\n): AsyncIterable<ProviderEvent> {\n  const metadata = asRecord(req.metadata);\n  for await (const event of next.stream(req)) {\n    yield event;\n    if (event.type === 'finish') {\n      emitUsage({ usage: event.usage, finishReason: event.finishReason }, next, opts, metadata);\n    }\n  }\n}\n\nfunction asRecord(value: unknown): Readonly<Record<string, unknown>> | undefined {\n  if (value === undefined || value === null) return undefined;\n  return value as Readonly<Record<string, unknown>>;\n}\n\nfunction emitUsage(\n  result:\n    | ProviderResponse\n    | { usage: ProviderResponse['usage']; finishReason: ProviderResponse['finishReason'] },\n  next: Provider,\n  opts: WithCostTrackingOptions,\n  metadata: Readonly<Record<string, unknown>> | undefined,\n): void {\n  if (opts.onUsage === undefined) return;\n  const usage = result.usage;\n  const price = opts.priceLookup?.({ providerName: next.name, modelId: next.modelId }) ?? null;\n  const costUsd =\n    price !== null\n      ? (usage.promptTokens * (price.inputPerMtok ?? 0)) / 1_000_000 +\n        (usage.completionTokens * (price.outputPerMtok ?? 0)) / 1_000_000\n      : (usage.cost?.amount ?? 0);\n  opts.onUsage({\n    providerName: next.name,\n    modelId: next.modelId,\n    promptTokens: usage.promptTokens,\n    completionTokens: usage.completionTokens,\n    totalTokens: usage.totalTokens,\n    costUsd,\n    metadata,\n  });\n}\n"],"mappings":";;;AAgDA,MAAMA,cAAkC,OAAO,OAAO;CACpD,WAAW;CACX,cAAc;CACd,kBAAkB;CAClB,aAAa;CACb,SAAS;CACV,CAAC;;;;;;;AAQF,SAAgB,wBAAyC;CACvD,MAAM,sBAAM,IAAI,KAAiC;CACjD,MAAM,SAAS,cAAsB,YAA4B,GAAG,aAAa,IAAI;AACrF,QAAO;EACL,UAAU,SAAS;GACjB,MAAM,MAAM,MAAM,KAAK,cAAc,KAAK,QAAQ;GAClD,MAAM,OAAO,IAAI,IAAI,IAAI,IAAI;AAC7B,OAAI,IAAI,KAAK;IACX,WAAW,KAAK,YAAY;IAC5B,cAAc,KAAK,eAAe,KAAK;IACvC,kBAAkB,KAAK,mBAAmB,KAAK;IAC/C,aAAa,KAAK,cAAc,KAAK;IACrC,SAAS,KAAK,UAAU,KAAK;IAC9B,CAAC;;EAEJ,cAAc,IAAI,IAAI,IAAI;EAC1B,WAAW,cAAc,YAAY,IAAI,IAAI,MAAM,cAAc,QAAQ,CAAC,IAAI;EAC9E,aAAa,IAAI,OAAO;EACzB;;;;;AAoCH,MAAa,mBAAmB,yBAAkD;CAChF,MAAM;CACN,UAAU,SAAkC;AAC1C,UAAQ,UAA8B;GACpC,MAAM,KAAK;GACX,SAAS,KAAK;GACd,cAAc,KAAK;GACnB,GAAI,KAAK,uBAAuB,SAC5B,EAAE,oBAAoB,KAAK,oBAAoB,GAC/C,EAAE;GACN,OAAO,KAAK;AACV,WAAO,eAAe,MAAM,KAAK,KAAK;;GAExC,MAAM,SAAS,KAAK;IAClB,MAAM,SAAS,MAAM,KAAK,SAAS,IAAI;AACvC,cAAU,QAAQ,MAAM,MAAM,SAAS,IAAI,SAAS,CAAC;AACrD,WAAO;;GAET,GAAI,KAAK,cAAc,EAAE,aAAa,KAAK,YAAY,KAAK,KAAK,EAAE,GAAG,EAAE;GACzE;;CAEJ,CAAC;AAEF,gBAAgB,eACd,MACA,KACA,MAC8B;CAC9B,MAAM,WAAW,SAAS,IAAI,SAAS;AACvC,YAAW,MAAM,SAAS,KAAK,OAAO,IAAI,EAAE;AAC1C,QAAM;AACN,MAAI,MAAM,SAAS,SACjB,WAAU;GAAE,OAAO,MAAM;GAAO,cAAc,MAAM;GAAc,EAAE,MAAM,MAAM,SAAS;;;AAK/F,SAAS,SAAS,OAA+D;AAC/E,KAAI,UAAU,UAAa,UAAU,KAAM,QAAO;AAClD,QAAO;;AAGT,SAAS,UACP,QAGA,MACA,MACA,UACM;AACN,KAAI,KAAK,YAAY,OAAW;CAChC,MAAM,QAAQ,OAAO;CACrB,MAAM,QAAQ,KAAK,cAAc;EAAE,cAAc,KAAK;EAAM,SAAS,KAAK;EAAS,CAAC,IAAI;CACxF,MAAM,UACJ,UAAU,OACL,MAAM,gBAAgB,MAAM,gBAAgB,KAAM,MAClD,MAAM,oBAAoB,MAAM,iBAAiB,KAAM,MACvD,MAAM,MAAM,UAAU;AAC7B,MAAK,QAAQ;EACX,cAAc,KAAK;EACnB,SAAS,KAAK;EACd,cAAc,MAAM;EACpB,kBAAkB,MAAM;EACxB,aAAa,MAAM;EACnB;EACA;EACD,CAAC"}
+{"version":3,"file":"with-cost-tracking.js","names":["ZERO_TOTALS: CostTrackingTotals"],"sources":["../../src/middleware/with-cost-tracking.ts"],"sourcesContent":["/**\n * `withCostTracking` - reports `tokensUsed` + `cost` per `provider × model`\n * from the `finish` event of every stream / one-shot generation via the\n * `onUsage` hook. For a ready-made process-local accumulator, wire\n * {@link createCostAccumulator}'s `onUsage` and read its `.totals()`; consumers\n * that prefer the framework's `CostTracker` from `@graphorin/observability`\n * pass their own delegate.\n *\n * @packageDocumentation\n */\n\nimport type { Provider, ProviderEvent, ProviderResponse } from '@graphorin/core';\n\nimport { defineProviderMiddleware } from './compose.js';\n\n/**\n * Aggregated totals for one `provider × model`, returned by\n * {@link CostAccumulator.totalFor} / {@link CostAccumulator.totals}.\n *\n * @stable\n */\nexport interface CostTrackingTotals {\n  readonly callCount: number;\n  readonly promptTokens: number;\n  readonly completionTokens: number;\n  readonly totalTokens: number;\n  /** Prompt tokens served from the provider cache (subset of promptTokens). */\n  readonly cachedReadTokens: number;\n  /** Prompt tokens written to the provider cache (subset of promptTokens). */\n  readonly cacheWriteTokens: number;\n  readonly costUsd: number;\n}\n\n/**\n * A process-local cost accumulator (PS-8). Wire {@link CostAccumulator.onUsage}\n * into {@link withCostTracking} and read the running totals - keyed by\n * `provider × model` - back via {@link CostAccumulator.totals} /\n * {@link CostAccumulator.totalFor}.\n *\n * @stable\n */\nexport interface CostAccumulator {\n  /** Pass this to {@link withCostTracking}'s `onUsage`. */\n  readonly onUsage: NonNullable<WithCostTrackingOptions['onUsage']>;\n  /** Snapshot of every tracked `provider::model` → totals. */\n  totals(): ReadonlyMap<string, CostTrackingTotals>;\n  /** Running totals for one `provider × model` (zeros when unseen). */\n  totalFor(providerName: string, modelId: string): CostTrackingTotals;\n  /** Clear all accumulated totals. */\n  reset(): void;\n}\n\nconst ZERO_TOTALS: CostTrackingTotals = Object.freeze({\n  callCount: 0,\n  promptTokens: 0,\n  completionTokens: 0,\n  totalTokens: 0,\n  cachedReadTokens: 0,\n  cacheWriteTokens: 0,\n  costUsd: 0,\n});\n\n/**\n * Create a {@link CostAccumulator} - the process-local accumulator described on\n * {@link withCostTracking}. Keys totals by `'<providerName>::<modelId>'`.\n *\n * @stable\n */\nexport function createCostAccumulator(): CostAccumulator {\n  const map = new Map<string, CostTrackingTotals>();\n  const keyOf = (providerName: string, modelId: string): string => `${providerName}::${modelId}`;\n  return {\n    onUsage: (info) => {\n      const key = keyOf(info.providerName, info.modelId);\n      const prev = map.get(key) ?? ZERO_TOTALS;\n      map.set(key, {\n        callCount: prev.callCount + 1,\n        promptTokens: prev.promptTokens + info.promptTokens,\n        completionTokens: prev.completionTokens + info.completionTokens,\n        totalTokens: prev.totalTokens + info.totalTokens,\n        cachedReadTokens: prev.cachedReadTokens + (info.cachedReadTokens ?? 0),\n        cacheWriteTokens: prev.cacheWriteTokens + (info.cacheWriteTokens ?? 0),\n        costUsd: prev.costUsd + info.costUsd,\n      });\n    },\n    totals: () => new Map(map),\n    totalFor: (providerName, modelId) => map.get(keyOf(providerName, modelId)) ?? ZERO_TOTALS,\n    reset: () => map.clear(),\n  };\n}\n\n/**\n * Options for {@link withCostTracking}.\n *\n * @stable\n */\nexport interface WithCostTrackingOptions {\n  /**\n   * Hook fired on every `finish` event with the parsed usage. The\n   * hook receives the underlying provider's name + modelId so the\n   * caller can route into a per-model accumulator.\n   */\n  readonly onUsage?: (info: {\n    readonly providerName: string;\n    readonly modelId: string;\n    readonly promptTokens: number;\n    readonly completionTokens: number;\n    readonly totalTokens: number;\n    readonly cachedReadTokens?: number;\n    readonly cacheWriteTokens?: number;\n    readonly costUsd: number;\n    readonly metadata: Readonly<Record<string, unknown>> | undefined;\n  }) => void;\n  /**\n   * Optional pricing lookup. When set, the middleware computes\n   * `costUsd` from the returned price and surfaces it on the hook.\n   *\n   * `cachedReadPerMtok` / `cacheWritePerMtok` price the prompt-cache legs\n   * (core-provider-02); when omitted, cache tokens are billed at the full\n   * input rate (never cheaper than reality, so absent price data degrades\n   * to the pre-cache behaviour).\n   */\n  readonly priceLookup?: (info: { readonly providerName: string; readonly modelId: string }) => {\n    readonly inputPerMtok?: number;\n    readonly outputPerMtok?: number;\n    readonly cachedReadPerMtok?: number;\n    readonly cacheWritePerMtok?: number;\n  } | null;\n}\n\n/**\n * @stable\n */\nexport const withCostTracking = defineProviderMiddleware<WithCostTrackingOptions>({\n  kind: 'withCostTracking',\n  factory: (opts: WithCostTrackingOptions) => {\n    return (next: Provider): Provider => ({\n      name: next.name,\n      modelId: next.modelId,\n      capabilities: next.capabilities,\n      ...(next.acceptsSensitivity !== undefined\n        ? { acceptsSensitivity: next.acceptsSensitivity }\n        : {}),\n      stream(req) {\n        return trackingStream(next, req, opts);\n      },\n      async generate(req) {\n        const result = await next.generate(req);\n        emitUsage(result, next, opts, asRecord(req.metadata));\n        return result;\n      },\n      ...(next.countTokens ? { countTokens: next.countTokens.bind(next) } : {}),\n    });\n  },\n});\n\nasync function* trackingStream(\n  next: Provider,\n  req: import('@graphorin/core').ProviderRequest,\n  opts: WithCostTrackingOptions,\n): AsyncIterable<ProviderEvent> {\n  const metadata = asRecord(req.metadata);\n  for await (const event of next.stream(req)) {\n    yield event;\n    if (event.type === 'finish') {\n      emitUsage({ usage: event.usage, finishReason: event.finishReason }, next, opts, metadata);\n    }\n  }\n}\n\nfunction asRecord(value: unknown): Readonly<Record<string, unknown>> | undefined {\n  if (value === undefined || value === null) return undefined;\n  return value as Readonly<Record<string, unknown>>;\n}\n\nfunction emitUsage(\n  result:\n    | ProviderResponse\n    | { usage: ProviderResponse['usage']; finishReason: ProviderResponse['finishReason'] },\n  next: Provider,\n  opts: WithCostTrackingOptions,\n  metadata: Readonly<Record<string, unknown>> | undefined,\n): void {\n  if (opts.onUsage === undefined) return;\n  const usage = result.usage;\n  const price = opts.priceLookup?.({ providerName: next.name, modelId: next.modelId }) ?? null;\n  // Reasoning tokens are billed at the output rate (the PS-19 contract\n  // in @graphorin/pricing) - providers that report them separately from\n  // completionTokens must not get them for free.\n  // Cache legs (core-provider-02): promptTokens INCLUDES cache reads and\n  // writes, so subtract them from the base-rate leg and bill each at its\n  // own rate; a missing cache rate falls back to the full input rate.\n  const cachedRead = usage.cachedReadTokens ?? 0;\n  const cacheWrite = usage.cacheWriteTokens ?? 0;\n  const basePromptTokens = Math.max(0, usage.promptTokens - cachedRead - cacheWrite);\n  const inputRate = price?.inputPerMtok ?? 0;\n  const costUsd =\n    price !== null\n      ? (basePromptTokens * inputRate +\n          cachedRead * (price.cachedReadPerMtok ?? inputRate) +\n          cacheWrite * (price.cacheWritePerMtok ?? inputRate) +\n          (usage.completionTokens + (usage.reasoningTokens ?? 0)) * (price.outputPerMtok ?? 0)) /\n        1_000_000\n      : (usage.cost?.amount ?? 0);\n  opts.onUsage({\n    providerName: next.name,\n    modelId: next.modelId,\n    promptTokens: usage.promptTokens,\n    completionTokens: usage.completionTokens,\n    totalTokens: usage.totalTokens,\n    ...(usage.cachedReadTokens !== undefined ? { cachedReadTokens: usage.cachedReadTokens } : {}),\n    ...(usage.cacheWriteTokens !== undefined ? { cacheWriteTokens: usage.cacheWriteTokens } : {}),\n    costUsd,\n    metadata,\n  });\n}\n"],"mappings":";;;AAoDA,MAAMA,cAAkC,OAAO,OAAO;CACpD,WAAW;CACX,cAAc;CACd,kBAAkB;CAClB,aAAa;CACb,kBAAkB;CAClB,kBAAkB;CAClB,SAAS;CACV,CAAC;;;;;;;AAQF,SAAgB,wBAAyC;CACvD,MAAM,sBAAM,IAAI,KAAiC;CACjD,MAAM,SAAS,cAAsB,YAA4B,GAAG,aAAa,IAAI;AACrF,QAAO;EACL,UAAU,SAAS;GACjB,MAAM,MAAM,MAAM,KAAK,cAAc,KAAK,QAAQ;GAClD,MAAM,OAAO,IAAI,IAAI,IAAI,IAAI;AAC7B,OAAI,IAAI,KAAK;IACX,WAAW,KAAK,YAAY;IAC5B,cAAc,KAAK,eAAe,KAAK;IACvC,kBAAkB,KAAK,mBAAmB,KAAK;IAC/C,aAAa,KAAK,cAAc,KAAK;IACrC,kBAAkB,KAAK,oBAAoB,KAAK,oBAAoB;IACpE,kBAAkB,KAAK,oBAAoB,KAAK,oBAAoB;IACpE,SAAS,KAAK,UAAU,KAAK;IAC9B,CAAC;;EAEJ,cAAc,IAAI,IAAI,IAAI;EAC1B,WAAW,cAAc,YAAY,IAAI,IAAI,MAAM,cAAc,QAAQ,CAAC,IAAI;EAC9E,aAAa,IAAI,OAAO;EACzB;;;;;AA6CH,MAAa,mBAAmB,yBAAkD;CAChF,MAAM;CACN,UAAU,SAAkC;AAC1C,UAAQ,UAA8B;GACpC,MAAM,KAAK;GACX,SAAS,KAAK;GACd,cAAc,KAAK;GACnB,GAAI,KAAK,uBAAuB,SAC5B,EAAE,oBAAoB,KAAK,oBAAoB,GAC/C,EAAE;GACN,OAAO,KAAK;AACV,WAAO,eAAe,MAAM,KAAK,KAAK;;GAExC,MAAM,SAAS,KAAK;IAClB,MAAM,SAAS,MAAM,KAAK,SAAS,IAAI;AACvC,cAAU,QAAQ,MAAM,MAAM,SAAS,IAAI,SAAS,CAAC;AACrD,WAAO;;GAET,GAAI,KAAK,cAAc,EAAE,aAAa,KAAK,YAAY,KAAK,KAAK,EAAE,GAAG,EAAE;GACzE;;CAEJ,CAAC;AAEF,gBAAgB,eACd,MACA,KACA,MAC8B;CAC9B,MAAM,WAAW,SAAS,IAAI,SAAS;AACvC,YAAW,MAAM,SAAS,KAAK,OAAO,IAAI,EAAE;AAC1C,QAAM;AACN,MAAI,MAAM,SAAS,SACjB,WAAU;GAAE,OAAO,MAAM;GAAO,cAAc,MAAM;GAAc,EAAE,MAAM,MAAM,SAAS;;;AAK/F,SAAS,SAAS,OAA+D;AAC/E,KAAI,UAAU,UAAa,UAAU,KAAM,QAAO;AAClD,QAAO;;AAGT,SAAS,UACP,QAGA,MACA,MACA,UACM;AACN,KAAI,KAAK,YAAY,OAAW;CAChC,MAAM,QAAQ,OAAO;CACrB,MAAM,QAAQ,KAAK,cAAc;EAAE,cAAc,KAAK;EAAM,SAAS,KAAK;EAAS,CAAC,IAAI;CAOxF,MAAM,aAAa,MAAM,oBAAoB;CAC7C,MAAM,aAAa,MAAM,oBAAoB;CAC7C,MAAM,mBAAmB,KAAK,IAAI,GAAG,MAAM,eAAe,aAAa,WAAW;CAClF,MAAM,YAAY,OAAO,gBAAgB;CACzC,MAAM,UACJ,UAAU,QACL,mBAAmB,YAClB,cAAc,MAAM,qBAAqB,aACzC,cAAc,MAAM,qBAAqB,cACxC,MAAM,oBAAoB,MAAM,mBAAmB,OAAO,MAAM,iBAAiB,MACpF,MACC,MAAM,MAAM,UAAU;AAC7B,MAAK,QAAQ;EACX,cAAc,KAAK;EACnB,SAAS,KAAK;EACd,cAAc,MAAM;EACpB,kBAAkB,MAAM;EACxB,aAAa,MAAM;EACnB,GAAI,MAAM,qBAAqB,SAAY,EAAE,kBAAkB,MAAM,kBAAkB,GAAG,EAAE;EAC5F,GAAI,MAAM,qBAAqB,SAAY,EAAE,kBAAkB,MAAM,kBAAkB,GAAG,EAAE;EAC5F;EACA;EACD,CAAC"}

package/dist/middleware/with-fallback.js

@@ -68,8 +68,10 @@ function defaultShouldFallback(err) {
 	if (err === null || typeof err !== "object") return false;
 	if (isAbortError(err)) return false;
 	const e = err;
-	if (e.kind === "unauthorized" || e.kind === "invalid-request") return false;
-	if (e.kind === "transient" || e.kind === "rate-limit" || e.kind === "capacity") return true;
+	const kinds = [e.kind, e.errorKind];
+	if (kinds.includes("unauthorized") || kinds.includes("invalid-request") || kinds.includes("context-length") || kinds.includes("content-filter")) return false;
+	if (kinds.includes("transient") || kinds.includes("rate-limit") || kinds.includes("rate-limit-exceeded") || kinds.includes("capacity")) return true;
+	if (typeof e.status === "number" && e.status === 429) return true;
 	if (typeof e.status === "number" && e.status >= 500) return true;
 	if (typeof e.status === "number" && e.status === 0) return true;
 	return false;