@warlock.js/ai-openai 5.25.0 → 5.27.0

package/esm/utils/wrap-openai-error.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"wrap-openai-error.mjs","names":[],"sources":["../../../../../../../ai-openai/src/utils/wrap-openai-error.ts"],"sourcesContent":["import {\n  AIError,\n  ContentFilterError,\n  ContextLengthExceededError,\n  InvalidRequestError,\n  ProviderAuthError,\n  ProviderError,\n  ProviderRateLimitError,\n  ProviderTimeoutError,\n  QuotaExceededError,\n} from \"@warlock.js/ai\";\nimport OpenAI from \"openai\";\n\n/**\n * Raw-error fields the wrapper reads off an OpenAI SDK error.\n *\n * `APIError` exposes `status`, `code`, `message`, `type`, `headers` —\n * we duck-type because wrapped retries, proxied errors, and custom\n * error subclasses sometimes lose the `instanceof` relationship.\n */\ntype OpenAIErrorShape = {\n  status?: number;\n  code?: string | null;\n  message?: string;\n  type?: string | null;\n  headers?: Record<string, string> | undefined;\n  name?: string;\n};\n\n/**\n * Wrap any thrown value caught inside the OpenAI adapter into the\n * appropriate `@warlock.js/ai` `AIError` subclass.\n *\n * **Dispatch strategy.** Prefers `APIError.code` when present (stable\n * machine identifier across SDK versions), falls back to `status` when\n * `code` is missing (common with proxied deployments that strip the\n * field). Name-based detection (`APIConnectionTimeoutError`) catches\n * transport-layer errors that never produced an HTTP response.\n *\n * `AIError` instances are returned unchanged — callers can pass the\n * error through `try/catch/throw wrap(e)` pipelines without accidental\n * double-wrapping.\n *\n * @example\n * try {\n *   return await this.client.chat.completions.create(...);\n * } catch (thrown) {\n *   throw wrapOpenAIError(thrown);\n * }\n */\nexport function wrapOpenAIError(thrown: unknown): AIError {\n  if (thrown instanceof AIError) {\n    return thrown;\n  }\n\n  const shape = toShape(thrown);\n  const context = buildContext(thrown, shape);\n  const message = shape.message ?? (thrown instanceof Error ? thrown.message : String(thrown));\n\n  if (isTimeout(thrown, shape)) {\n    return new ProviderTimeoutError(message, { cause: thrown, context });\n  }\n\n  if (shape.status === 401 || shape.code === \"invalid_api_key\") {\n    return new ProviderAuthError(message, { cause: thrown, context });\n  }\n\n  if (shape.code === \"insufficient_quota\") {\n    return new QuotaExceededError(message, { cause: thrown, context });\n  }\n\n  if (shape.status === 429 || shape.code === \"rate_limit_exceeded\") {\n    return new ProviderRateLimitError(message, {\n      cause: thrown,\n      context,\n      retryAfter: parseRetryAfter(shape.headers),\n    });\n  }\n\n  if (shape.code === \"context_length_exceeded\") {\n    return new ContextLengthExceededError(message, { cause: thrown, context });\n  }\n\n  if (shape.code === \"content_filter\") {\n    return new ContentFilterError(message, {\n      cause: thrown,\n      context,\n      reason: message,\n    });\n  }\n\n  if (typeof shape.status === \"number\" && shape.status >= 400 && shape.status < 500) {\n    return new InvalidRequestError(message, { cause: thrown, context });\n  }\n\n  return new ProviderError(message, { cause: thrown, context });\n}\n\n/**\n * Read the raw error shape without depending on `instanceof APIError`\n * — some consumers wrap the SDK, and proxies sometimes strip the\n * prototype chain. Duck-typing on the visible fields is resilient to\n * both.\n */\nfunction toShape(thrown: unknown): OpenAIErrorShape {\n  if (thrown instanceof OpenAI.APIError) {\n    return {\n      status: thrown.status,\n      code: thrown.code,\n      message: thrown.message,\n      type: thrown.type,\n      headers: thrown.headers as Record<string, string> | undefined,\n      name: thrown.name,\n    };\n  }\n\n  if (typeof thrown === \"object\" && thrown !== null) {\n    const raw = thrown as Record<string, unknown>;\n\n    return {\n      status: typeof raw.status === \"number\" ? raw.status : undefined,\n      code: typeof raw.code === \"string\" ? raw.code : undefined,\n      message: typeof raw.message === \"string\" ? raw.message : undefined,\n      type: typeof raw.type === \"string\" ? raw.type : undefined,\n      headers:\n        typeof raw.headers === \"object\" && raw.headers !== null\n          ? (raw.headers as Record<string, string>)\n          : undefined,\n      name: typeof raw.name === \"string\" ? raw.name : undefined,\n    };\n  }\n\n  return {};\n}\n\n/**\n * Decide whether the thrown value represents a timeout. OpenAI's SDK\n * throws `APIConnectionTimeoutError` for transport-level timeouts, and\n * Node surfaces `ETIMEDOUT` / `ECONNABORTED` on the lower socket\n * layer. Either signal counts.\n */\nfunction isTimeout(thrown: unknown, shape: OpenAIErrorShape): boolean {\n  if (thrown instanceof OpenAI.APIConnectionTimeoutError) {\n    return true;\n  }\n\n  if (shape.name === \"APIConnectionTimeoutError\") {\n    return true;\n  }\n\n  if (shape.code === \"ETIMEDOUT\" || shape.code === \"ECONNABORTED\") {\n    return true;\n  }\n\n  return false;\n}\n\n/**\n * Attach the raw diagnostic fields to `error.context` so consumers\n * have everything the provider surfaced without each subclass having\n * to redeclare them. Never includes `cause` — that lives on\n * `error.cause`.\n */\nfunction buildContext(\n  thrown: unknown,\n  shape: OpenAIErrorShape,\n): Record<string, unknown> {\n  const context: Record<string, unknown> = {};\n\n  if (shape.status !== undefined) {\n    context.status = shape.status;\n  }\n\n  if (shape.code) {\n    context.code = shape.code;\n  }\n\n  if (shape.type) {\n    context.type = shape.type;\n  }\n\n  const requestId = readRequestId(thrown);\n\n  if (requestId) {\n    context.requestId = requestId;\n  }\n\n  return context;\n}\n\n/**\n * OpenAI puts the request id on `APIError.request_id`. Extract\n * defensively — both camel and snake keys exist across SDK versions.\n */\nfunction readRequestId(thrown: unknown): string | undefined {\n  if (typeof thrown !== \"object\" || thrown === null) {\n    return undefined;\n  }\n\n  const raw = thrown as Record<string, unknown>;\n\n  if (typeof raw.request_id === \"string\") {\n    return raw.request_id;\n  }\n\n  if (typeof raw.requestId === \"string\") {\n    return raw.requestId;\n  }\n\n  return undefined;\n}\n\n/**\n * Parse the `Retry-After` response header (seconds per HTTP spec)\n * into milliseconds so consumers can feed it straight to `setTimeout`.\n * Returns `undefined` when missing or unparseable.\n */\nfunction parseRetryAfter(headers: Record<string, string> | undefined): number | undefined {\n  if (!headers) {\n    return undefined;\n  }\n\n  const raw = headers[\"retry-after\"] ?? headers[\"Retry-After\"];\n\n  if (!raw) {\n    return undefined;\n  }\n\n  const seconds = Number(raw);\n\n  if (!Number.isFinite(seconds) || seconds < 0) {\n    return undefined;\n  }\n\n  return Math.round(seconds * 1000);\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;AAkDA,SAAgB,gBAAgB,QAA0B;CACxD,IAAI,kBAAkB,SACpB,OAAO;CAGT,MAAM,QAAQ,QAAQ,MAAM;CAC5B,MAAM,UAAU,aAAa,QAAQ,KAAK;CAC1C,MAAM,UAAU,MAAM,YAAY,kBAAkB,QAAQ,OAAO,UAAU,OAAO,MAAM;CAE1F,IAAI,UAAU,QAAQ,KAAK,GACzB,OAAO,IAAI,qBAAqB,SAAS;EAAE,OAAO;EAAQ;CAAQ,CAAC;CAGrE,IAAI,MAAM,WAAW,OAAO,MAAM,SAAS,mBACzC,OAAO,IAAI,kBAAkB,SAAS;EAAE,OAAO;EAAQ;CAAQ,CAAC;CAGlE,IAAI,MAAM,SAAS,sBACjB,OAAO,IAAI,mBAAmB,SAAS;EAAE,OAAO;EAAQ;CAAQ,CAAC;CAGnE,IAAI,MAAM,WAAW,OAAO,MAAM,SAAS,uBACzC,OAAO,IAAI,uBAAuB,SAAS;EACzC,OAAO;EACP;EACA,YAAY,gBAAgB,MAAM,OAAO;CAC3C,CAAC;CAGH,IAAI,MAAM,SAAS,2BACjB,OAAO,IAAI,2BAA2B,SAAS;EAAE,OAAO;EAAQ;CAAQ,CAAC;CAG3E,IAAI,MAAM,SAAS,kBACjB,OAAO,IAAI,mBAAmB,SAAS;EACrC,OAAO;EACP;EACA,QAAQ;CACV,CAAC;CAGH,IAAI,OAAO,MAAM,WAAW,YAAY,MAAM,UAAU,OAAO,MAAM,SAAS,KAC5E,OAAO,IAAI,oBAAoB,SAAS;EAAE,OAAO;EAAQ;CAAQ,CAAC;CAGpE,OAAO,IAAI,cAAc,SAAS;EAAE,OAAO;EAAQ;CAAQ,CAAC;AAC9D;;;;;;;AAQA,SAAS,QAAQ,QAAmC;CAClD,IAAI,kBAAkB,OAAO,UAC3B,OAAO;EACL,QAAQ,OAAO;EACf,MAAM,OAAO;EACb,SAAS,OAAO;EAChB,MAAM,OAAO;EACb,SAAS,OAAO;EAChB,MAAM,OAAO;CACf;CAGF,IAAI,OAAO,WAAW,YAAY,WAAW,MAAM;EACjD,MAAM,MAAM;EAEZ,OAAO;GACL,QAAQ,OAAO,IAAI,WAAW,WAAW,IAAI,SAAS;GACtD,MAAM,OAAO,IAAI,SAAS,WAAW,IAAI,OAAO;GAChD,SAAS,OAAO,IAAI,YAAY,WAAW,IAAI,UAAU;GACzD,MAAM,OAAO,IAAI,SAAS,WAAW,IAAI,OAAO;GAChD,SACE,OAAO,IAAI,YAAY,YAAY,IAAI,YAAY,OAC9C,IAAI,UACL;GACN,MAAM,OAAO,IAAI,SAAS,WAAW,IAAI,OAAO;EAClD;CACF;CAEA,OAAO,CAAC;AACV;;;;;;;AAQA,SAAS,UAAU,QAAiB,OAAkC;CACpE,IAAI,kBAAkB,OAAO,2BAC3B,OAAO;CAGT,IAAI,MAAM,SAAS,6BACjB,OAAO;CAGT,IAAI,MAAM,SAAS,eAAe,MAAM,SAAS,gBAC/C,OAAO;CAGT,OAAO;AACT;;;;;;;AAQA,SAAS,aACP,QACA,OACyB;CACzB,MAAM,UAAmC,CAAC;CAE1C,IAAI,MAAM,WAAW,QACnB,QAAQ,SAAS,MAAM;CAGzB,IAAI,MAAM,MACR,QAAQ,OAAO,MAAM;CAGvB,IAAI,MAAM,MACR,QAAQ,OAAO,MAAM;CAGvB,MAAM,YAAY,cAAc,MAAM;CAEtC,IAAI,WACF,QAAQ,YAAY;CAGtB,OAAO;AACT;;;;;AAMA,SAAS,cAAc,QAAqC;CAC1D,IAAI,OAAO,WAAW,YAAY,WAAW,MAC3C;CAGF,MAAM,MAAM;CAEZ,IAAI,OAAO,IAAI,eAAe,UAC5B,OAAO,IAAI;CAGb,IAAI,OAAO,IAAI,cAAc,UAC3B,OAAO,IAAI;AAIf;;;;;;AAOA,SAAS,gBAAgB,SAAiE;CACxF,IAAI,CAAC,SACH;CAGF,MAAM,MAAM,QAAQ,kBAAkB,QAAQ;CAE9C,IAAI,CAAC,KACH;CAGF,MAAM,UAAU,OAAO,GAAG;CAE1B,IAAI,CAAC,OAAO,SAAS,OAAO,KAAK,UAAU,GACzC;CAGF,OAAO,KAAK,MAAM,UAAU,GAAI;AAClC"}
+{"version":3,"file":"wrap-openai-error.mjs","names":[],"sources":["../../../../../../../ai-openai/src/utils/wrap-openai-error.ts"],"sourcesContent":["import {\n  AIError,\n  ContentFilterError,\n  ContextLengthExceededError,\n  InvalidRequestError,\n  ProviderAuthError,\n  ProviderError,\n  ProviderRateLimitError,\n  ProviderTimeoutError,\n  QuotaExceededError,\n} from \"@warlock.js/ai\";\nimport OpenAI from \"openai\";\n\n/**\n * Raw-error fields the wrapper reads off an OpenAI SDK error.\n *\n * `APIError` exposes `status`, `code`, `message`, `type`, `headers` —\n * we duck-type because wrapped retries, proxied errors, and custom\n * error subclasses sometimes lose the `instanceof` relationship.\n */\ntype OpenAIErrorShape = {\n  status?: number;\n  code?: string | null;\n  message?: string;\n  type?: string | null;\n  /** The request parameter the provider rejected, e.g. `max_tokens`. */\n  param?: string | null;\n  headers?: Record<string, string> | undefined;\n  name?: string;\n};\n\n/**\n * Wrap any thrown value caught inside the OpenAI adapter into the\n * appropriate `@warlock.js/ai` `AIError` subclass.\n *\n * **Dispatch strategy.** Prefers `APIError.code` when present (stable\n * machine identifier across SDK versions), falls back to `status` when\n * `code` is missing (common with proxied deployments that strip the\n * field). Name-based detection (`APIConnectionTimeoutError`) catches\n * transport-layer errors that never produced an HTTP response.\n *\n * `AIError` instances are returned unchanged — callers can pass the\n * error through `try/catch/throw wrap(e)` pipelines without accidental\n * double-wrapping.\n *\n * @example\n * try {\n *   return await this.client.chat.completions.create(...);\n * } catch (thrown) {\n *   throw wrapOpenAIError(thrown);\n * }\n */\nexport function wrapOpenAIError(thrown: unknown): AIError {\n  if (thrown instanceof AIError) {\n    return thrown;\n  }\n\n  const shape = toShape(thrown);\n  const context = buildContext(thrown, shape);\n  const message = shape.message ?? (thrown instanceof Error ? thrown.message : String(thrown));\n\n  if (isTimeout(thrown, shape)) {\n    return new ProviderTimeoutError(message, { cause: thrown, context });\n  }\n\n  if (shape.status === 401 || shape.code === \"invalid_api_key\") {\n    return new ProviderAuthError(message, { cause: thrown, context });\n  }\n\n  if (shape.code === \"insufficient_quota\") {\n    return new QuotaExceededError(message, { cause: thrown, context });\n  }\n\n  if (shape.status === 429 || shape.code === \"rate_limit_exceeded\") {\n    return new ProviderRateLimitError(message, {\n      cause: thrown,\n      context,\n      retryAfter: parseRetryAfter(shape.headers),\n    });\n  }\n\n  if (shape.code === \"context_length_exceeded\") {\n    return new ContextLengthExceededError(message, { cause: thrown, context });\n  }\n\n  if (shape.code === \"content_filter\") {\n    return new ContentFilterError(message, {\n      cause: thrown,\n      context,\n      reason: message,\n    });\n  }\n\n  if (typeof shape.status === \"number\" && shape.status >= 400 && shape.status < 500) {\n    return new InvalidRequestError(message, { cause: thrown, context });\n  }\n\n  return new ProviderError(message, { cause: thrown, context });\n}\n\n/**\n * Read the raw error shape without depending on `instanceof APIError`\n * — some consumers wrap the SDK, and proxies sometimes strip the\n * prototype chain. Duck-typing on the visible fields is resilient to\n * both.\n */\nfunction toShape(thrown: unknown): OpenAIErrorShape {\n  if (thrown instanceof OpenAI.APIError) {\n    return {\n      status: thrown.status,\n      code: thrown.code,\n      message: thrown.message,\n      type: thrown.type,\n      param: thrown.param,\n      headers: thrown.headers as Record<string, string> | undefined,\n      name: thrown.name,\n    };\n  }\n\n  if (typeof thrown === \"object\" && thrown !== null) {\n    const raw = thrown as Record<string, unknown>;\n\n    return {\n      status: typeof raw.status === \"number\" ? raw.status : undefined,\n      code: typeof raw.code === \"string\" ? raw.code : undefined,\n      message: typeof raw.message === \"string\" ? raw.message : undefined,\n      type: typeof raw.type === \"string\" ? raw.type : undefined,\n      param: typeof raw.param === \"string\" ? raw.param : undefined,\n      headers:\n        typeof raw.headers === \"object\" && raw.headers !== null\n          ? (raw.headers as Record<string, string>)\n          : undefined,\n      name: typeof raw.name === \"string\" ? raw.name : undefined,\n    };\n  }\n\n  return {};\n}\n\n/**\n * Decide whether the thrown value represents a timeout. OpenAI's SDK\n * throws `APIConnectionTimeoutError` for transport-level timeouts, and\n * Node surfaces `ETIMEDOUT` / `ECONNABORTED` on the lower socket\n * layer. Either signal counts.\n */\nfunction isTimeout(thrown: unknown, shape: OpenAIErrorShape): boolean {\n  if (thrown instanceof OpenAI.APIConnectionTimeoutError) {\n    return true;\n  }\n\n  if (shape.name === \"APIConnectionTimeoutError\") {\n    return true;\n  }\n\n  if (shape.code === \"ETIMEDOUT\" || shape.code === \"ECONNABORTED\") {\n    return true;\n  }\n\n  return false;\n}\n\n/**\n * Attach the raw diagnostic fields to `error.context` so consumers\n * have everything the provider surfaced without each subclass having\n * to redeclare them. Never includes `cause` — that lives on\n * `error.cause`.\n */\nfunction buildContext(\n  thrown: unknown,\n  shape: OpenAIErrorShape,\n): Record<string, unknown> {\n  const context: Record<string, unknown> = {};\n\n  if (shape.status !== undefined) {\n    context.status = shape.status;\n  }\n\n  if (shape.code) {\n    context.code = shape.code;\n  }\n\n  if (shape.type) {\n    context.type = shape.type;\n  }\n\n  if (shape.param) {\n    context.param = shape.param;\n  }\n\n  const requestId = readRequestId(thrown);\n\n  if (requestId) {\n    context.requestId = requestId;\n  }\n\n  return context;\n}\n\n/**\n * The request id OpenAI returned. Current SDKs expose it as\n * `APIError.requestID`; older ones and hand-rolled errors use\n * `request_id` / `requestId`. Extract defensively.\n */\nfunction readRequestId(thrown: unknown): string | undefined {\n  if (typeof thrown !== \"object\" || thrown === null) {\n    return undefined;\n  }\n\n  const raw = thrown as Record<string, unknown>;\n\n  if (typeof raw.request_id === \"string\") {\n    return raw.request_id;\n  }\n\n  if (typeof raw.requestId === \"string\") {\n    return raw.requestId;\n  }\n\n  if (typeof raw.requestID === \"string\") {\n    return raw.requestID;\n  }\n\n  return undefined;\n}\n\n/**\n * Parse the `Retry-After` response header (seconds per HTTP spec)\n * into milliseconds so consumers can feed it straight to `setTimeout`.\n * Returns `undefined` when missing or unparseable.\n */\nfunction parseRetryAfter(headers: Record<string, string> | undefined): number | undefined {\n  if (!headers) {\n    return undefined;\n  }\n\n  const raw = headers[\"retry-after\"] ?? headers[\"Retry-After\"];\n\n  if (!raw) {\n    return undefined;\n  }\n\n  const seconds = Number(raw);\n\n  if (!Number.isFinite(seconds) || seconds < 0) {\n    return undefined;\n  }\n\n  return Math.round(seconds * 1000);\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;AAoDA,SAAgB,gBAAgB,QAA0B;CACxD,IAAI,kBAAkB,SACpB,OAAO;CAGT,MAAM,QAAQ,QAAQ,MAAM;CAC5B,MAAM,UAAU,aAAa,QAAQ,KAAK;CAC1C,MAAM,UAAU,MAAM,YAAY,kBAAkB,QAAQ,OAAO,UAAU,OAAO,MAAM;CAE1F,IAAI,UAAU,QAAQ,KAAK,GACzB,OAAO,IAAI,qBAAqB,SAAS;EAAE,OAAO;EAAQ;CAAQ,CAAC;CAGrE,IAAI,MAAM,WAAW,OAAO,MAAM,SAAS,mBACzC,OAAO,IAAI,kBAAkB,SAAS;EAAE,OAAO;EAAQ;CAAQ,CAAC;CAGlE,IAAI,MAAM,SAAS,sBACjB,OAAO,IAAI,mBAAmB,SAAS;EAAE,OAAO;EAAQ;CAAQ,CAAC;CAGnE,IAAI,MAAM,WAAW,OAAO,MAAM,SAAS,uBACzC,OAAO,IAAI,uBAAuB,SAAS;EACzC,OAAO;EACP;EACA,YAAY,gBAAgB,MAAM,OAAO;CAC3C,CAAC;CAGH,IAAI,MAAM,SAAS,2BACjB,OAAO,IAAI,2BAA2B,SAAS;EAAE,OAAO;EAAQ;CAAQ,CAAC;CAG3E,IAAI,MAAM,SAAS,kBACjB,OAAO,IAAI,mBAAmB,SAAS;EACrC,OAAO;EACP;EACA,QAAQ;CACV,CAAC;CAGH,IAAI,OAAO,MAAM,WAAW,YAAY,MAAM,UAAU,OAAO,MAAM,SAAS,KAC5E,OAAO,IAAI,oBAAoB,SAAS;EAAE,OAAO;EAAQ;CAAQ,CAAC;CAGpE,OAAO,IAAI,cAAc,SAAS;EAAE,OAAO;EAAQ;CAAQ,CAAC;AAC9D;;;;;;;AAQA,SAAS,QAAQ,QAAmC;CAClD,IAAI,kBAAkB,OAAO,UAC3B,OAAO;EACL,QAAQ,OAAO;EACf,MAAM,OAAO;EACb,SAAS,OAAO;EAChB,MAAM,OAAO;EACb,OAAO,OAAO;EACd,SAAS,OAAO;EAChB,MAAM,OAAO;CACf;CAGF,IAAI,OAAO,WAAW,YAAY,WAAW,MAAM;EACjD,MAAM,MAAM;EAEZ,OAAO;GACL,QAAQ,OAAO,IAAI,WAAW,WAAW,IAAI,SAAS;GACtD,MAAM,OAAO,IAAI,SAAS,WAAW,IAAI,OAAO;GAChD,SAAS,OAAO,IAAI,YAAY,WAAW,IAAI,UAAU;GACzD,MAAM,OAAO,IAAI,SAAS,WAAW,IAAI,OAAO;GAChD,OAAO,OAAO,IAAI,UAAU,WAAW,IAAI,QAAQ;GACnD,SACE,OAAO,IAAI,YAAY,YAAY,IAAI,YAAY,OAC9C,IAAI,UACL;GACN,MAAM,OAAO,IAAI,SAAS,WAAW,IAAI,OAAO;EAClD;CACF;CAEA,OAAO,CAAC;AACV;;;;;;;AAQA,SAAS,UAAU,QAAiB,OAAkC;CACpE,IAAI,kBAAkB,OAAO,2BAC3B,OAAO;CAGT,IAAI,MAAM,SAAS,6BACjB,OAAO;CAGT,IAAI,MAAM,SAAS,eAAe,MAAM,SAAS,gBAC/C,OAAO;CAGT,OAAO;AACT;;;;;;;AAQA,SAAS,aACP,QACA,OACyB;CACzB,MAAM,UAAmC,CAAC;CAE1C,IAAI,MAAM,WAAW,QACnB,QAAQ,SAAS,MAAM;CAGzB,IAAI,MAAM,MACR,QAAQ,OAAO,MAAM;CAGvB,IAAI,MAAM,MACR,QAAQ,OAAO,MAAM;CAGvB,IAAI,MAAM,OACR,QAAQ,QAAQ,MAAM;CAGxB,MAAM,YAAY,cAAc,MAAM;CAEtC,IAAI,WACF,QAAQ,YAAY;CAGtB,OAAO;AACT;;;;;;AAOA,SAAS,cAAc,QAAqC;CAC1D,IAAI,OAAO,WAAW,YAAY,WAAW,MAC3C;CAGF,MAAM,MAAM;CAEZ,IAAI,OAAO,IAAI,eAAe,UAC5B,OAAO,IAAI;CAGb,IAAI,OAAO,IAAI,cAAc,UAC3B,OAAO,IAAI;CAGb,IAAI,OAAO,IAAI,cAAc,UAC3B,OAAO,IAAI;AAIf;;;;;;AAOA,SAAS,gBAAgB,SAAiE;CACxF,IAAI,CAAC,SACH;CAGF,MAAM,MAAM,QAAQ,kBAAkB,QAAQ;CAE9C,IAAI,CAAC,KACH;CAGF,MAAM,UAAU,OAAO,GAAG;CAE1B,IAAI,CAAC,OAAO,SAAS,OAAO,KAAK,UAAU,GACzC;CAGF,OAAO,KAAK,MAAM,UAAU,GAAI;AAClC"}

package/llms-full.txt

@@ -8,7 +8,7 @@
 
 ---
 name: setup-openai
-description: 'Wire @warlock.js/ai-openai — new OpenAISDK({apiKey, baseURL?, provider?, pricing?}) for OpenAI / Azure / OpenRouter, .model({name, vision?, reasoning?, structuredOutput?, responseFormat?, pdf?, audio?}) for ModelContract, .embedder({name, dimensions?}) for embeddings, .image({name, pricing?}) for gpt-image-*/dall-e-* image generation (via ai.image). PDF input maps to OpenAI file parts (opt-in pdf:true), audio input to input_audio (opt-in audio:true). Triggers: `OpenAISDK`, `.model`, `.embedder`, `.image`, `.embed`, `.embedMany`, `baseURL`, `pricing`, `responseSchema`, `responseFormat`, `reasoning_effort`, `reasoningTokens`, `cachedTokens`, `gpt-image`, `dall-e`, `ai.image`, `pdf input`, `input_audio`, o-series / gpt-5 reasoning, prompt caching; "wire openai into a warlock agent", "configure gpt-4o", "use o3 / gpt-5 reasoning effort", "route through openrouter or azure openai", "openai embeddings with warlock", "generate images with gpt-image / dall-e", "send a pdf / audio to gpt-4o"; typical import `import { OpenAISDK } from "@warlock.js/ai-openai"`. Skip: the ai.image verb surface — `@warlock.js/ai/generate-images/SKILL.md`; agent wiring — `@warlock.js/ai/run-ai-agent/SKILL.md`; adapter comparison — `@warlock.js/ai/pick-ai-provider/SKILL.md`; competing adapters `@warlock.js/ai-anthropic`, `@warlock.js/ai-bedrock`, `@warlock.js/ai-google`, `@warlock.js/ai-ollama`; raw `openai` SDK, Vercel `@ai-sdk/openai`.'
+description: 'Wire @warlock.js/ai-openai — new OpenAISDK({apiKey, baseURL?, provider?, pricing?}) for OpenAI / Azure / OpenRouter, .model({name, vision?, reasoning?, structuredOutput?, responseFormat?, pdf?, audio?}) for ModelContract, .embedder({name, dimensions?}) for embeddings, .image({name, pricing?}) for gpt-image-*/dall-e-* image generation (via ai.image). PDF input maps to OpenAI file parts (opt-in pdf:true), audio input to input_audio (opt-in audio:true). Triggers: `OpenAISDK`, `.model`, `.embedder`, `.image`, `.embed`, `.embedMany`, `baseURL`, `pricing`, `responseSchema`, `responseFormat`, `reasoning_effort`, `reasoningTokens`, `cachedTokens`, `gpt-image`, `dall-e`, `ai.image`, `pdf input`, `input_audio`, o-series / gpt-5 reasoning, `api: "responses"`, Responses API, reasoning with tools, prompt caching; "wire openai into a warlock agent", "configure gpt-4o", "use o3 / gpt-5 reasoning effort", "route through openrouter or azure openai", "openai embeddings with warlock", "generate images with gpt-image / dall-e", "send a pdf / audio to gpt-4o"; typical import `import { OpenAISDK } from "@warlock.js/ai-openai"`. Skip: the ai.image verb surface — `@warlock.js/ai/generate-images/SKILL.md`; agent wiring — `@warlock.js/ai/run-ai-agent/SKILL.md`; adapter comparison — `@warlock.js/ai/pick-ai-provider/SKILL.md`; competing adapters `@warlock.js/ai-anthropic`, `@warlock.js/ai-bedrock`, `@warlock.js/ai-google`, `@warlock.js/ai-ollama`; raw `openai` SDK, Vercel `@ai-sdk/openai`.'
 ---
 
 # `@warlock.js/ai-openai`
@@ -53,7 +53,7 @@ Returns a `ModelContract` you pass straight into `ai.agent({ model })`.
 | `structuredOutput` | `true`, unless `responseFormat` is set to `"json_object"` or `"text"` (loose modes) — then `false`. |
 | `vision` | Inferred from model name. `true` for `gpt-4o*`, `gpt-4-turbo*`, `gpt-4.1*`, `o1*`, `o3*`, `chatgpt-4o*`; `false` otherwise. |
 | `reasoning` | Inferred from model name. `true` for the o-series (`o1*`, `o3*`, `o4*`) and the `gpt-5*` family; `false` otherwise. Drives whether `reasoning_effort` is forwarded. |
-| `promptCaching` | Always `true`. OpenAI caches long prompt prefixes automatically and reports hits via `usage.cachedTokens` — there are no caller-supplied write breakpoints. |
+| `promptCaching` | `true` for the direct OpenAI provider. It supports automatic cache hits plus explicit content-part write breakpoints; OpenAI-compatible wrapper providers leave the OpenAI-only fields off. |
 | `pdf` | `false` by default (opt-in `.model({ pdf: true })`). OpenAI accepts PDF `file` parts only on specific models (the `gpt-4o` family) — honest off until you set it. |
 | `audio` | `false` by default (opt-in `.model({ audio: true })`). Only the `gpt-4o-audio-preview` family accepts `input_audio`. |
 
@@ -157,7 +157,32 @@ await model.complete(messages, { tools, reasoning: { effort: "high" } }); // you
 await model.complete(messages, { tools, reasoning: { effort: "none" } }); // same effect as the default, explicit
 ```
 
-Calling **without** `tools` is unaffected — `reasoning_effort` is still omitted by default in that case (provider default reasoning, nothing to unblock). Trade-off of the `"none"` default: you lose reasoning on that call. For tool-heavy agent work (function calls + good replies rather than deep analysis) this is the right call — it's the difference between empty replies and working ones. When you need reasoning **and** tools together, use the Responses API (planned).
+The gpt-6 models (`gpt-6-astra` / `gpt-6-sol` / `gpt-6-luna`) are hard-400 cases, verified live on `gpt-6-sol`:
+
+```
+400 Function tools with reasoning_effort are not supported for gpt-6-sol in /v1/chat/completions. To use function tools, use /v1/responses or set reasoning_effort to 'none'.
+```
+
+Tools work with reasoning off (the adapter's default when tools are present and no reasoning is set). For reasoning **and** tools together, use the Responses API below.
+
+Calling **without** `tools` is unaffected — `reasoning_effort` is still omitted by default in that case (provider default reasoning, nothing to unblock). Trade-off of the `"none"` default: you lose reasoning on that call. For tool-heavy agent work (function calls + good replies rather than deep analysis) this is the right call — it's the difference between empty replies and working ones. When you need reasoning **and** tools together, use the Responses API (next section).
+
+### Reasoning + tools: the Responses API (`api: "responses"`)
+
+Opt a direct-OpenAI model into `/v1/responses`, where reasoning models accept function tools with reasoning on:
+
+```ts
+const model = openai.model({ name: "gpt-6-sol", api: "responses" });
+await model.complete(messages, { tools, reasoning: { effort: "high" } }); // reasoning stays on, tools work
+```
+
+- Chat Completions stays the default; nothing changes unless you pass `api: "responses"`.
+- Only the direct `openai` provider accepts it. `@warlock.js/ai-deepseek`, `ai-groq`, `ai-xai` and `ai-mistral` (which reuse this adapter with their own `baseURL`) reject `api: "responses"` with `InvalidRequestError`.
+- With tools attached, reasoning uses the provider default instead of being forced to `"none"`: expect more reasoning tokens and latency than the Chat path. Pass `reasoning: { effort }` to choose.
+- The system prompt is sent as `instructions`; requests use `store: false` (no server-side conversation state).
+- Reasoning is carried across a tool loop without server storage: each response asks for encrypted reasoning, which rides on the first tool call of the turn (`providerMetadata.openaiResponses`) and is sent back on the next request when the model is the same. It is persisted with your conversation history.
+- `complete()` and `stream()` both work; streamed chunks follow the same order as the Chat path (deltas, then tool calls, then `done`). Usage reports `reasoningTokens` and `cachedTokens` the same way.
+- `pdf` and `audio` inputs are not available on this path.
 
 ## Token usage — what's reported
 
@@ -171,7 +196,21 @@ Calling **without** `tools` is unaffected — `reasoning_effort` is still omitte
 
 Both `cachedTokens` and `reasoningTokens` are omitted (not set to `0`) when the provider reports zero, so non-reasoning / uncached calls keep the lean `{ input, output, total }` shape. Reported identically on `complete()` and the streaming `done` event.
 
-`cacheControl` write breakpoints (`ModelCallOptions.cacheControl`) are a **no-op** — OpenAI has no caller-driven cache marker on the Chat Completions wire. The read-side `cachedTokens` accounting above works without any caller action.
+### Explicit prompt caching (direct OpenAI only)
+
+Chat Completions exposes OpenAI-specific cache controls on current SDK types. Configure a stable cache bucket and (when needed) its maximum retention policy per model:
+
+```ts
+const model = openai.model({
+  name: "gpt-5.6",
+  promptCacheKey: "tenant:acme",
+  promptCacheRetention: "24h", // or "in_memory"
+});
+```
+
+`promptCacheKey` improves matching for related requests. `promptCacheRetention` is OpenAI's maximum policy (`"in_memory" | "24h"`). `cacheControl.breakpoints` maps to `prompt_cache_breakpoint: { mode: "explicit" }` on the final user content parts (up to OpenAI's four-request limit); use it when you need to pin a reusable prefix rather than rely only on OpenAI's implicit breakpoint.
+
+These fields are sent only when the SDK provider label is exactly `"openai"`. `DeepSeekSDK`, `GroqSDK`, `XaiSDK`, and `MistralSDK` label their wrapped endpoint differently, so their requests omit every OpenAI-only cache field even though they reuse this transport. Read-side `usage.cachedTokens` accounting remains available whenever the upstream reports it.
 
 ## Token counting
 

package/llms.txt

@@ -6,4 +6,4 @@
 
 ## Skills
 
-- [setup-openai](@warlock.js/ai-openai/setup-openai/SKILL.md): Wire @warlock.js/ai-openai — new OpenAISDK({apiKey, baseURL?, provider?, pricing?}) for OpenAI / Azure / OpenRouter, .model({name, vision?, reasoning?, structuredOutput?, responseFormat?, pdf?, audio?}) for ModelContract, .embedder({name, dimensions?}) for embeddings, .image({name, pricing?}) for gpt-image-*/dall-e-* image generation (via ai.image). PDF input maps to OpenAI file parts (opt-in pdf:true), audio input to input_audio (opt-in audio:true). Triggers: `OpenAISDK`, `.model`, `.embedder`, `.image`, `.embed`, `.embedMany`, `baseURL`, `pricing`, `responseSchema`, `responseFormat`, `reasoning_effort`, `reasoningTokens`, `cachedTokens`, `gpt-image`, `dall-e`, `ai.image`, `pdf input`, `input_audio`, o-series / gpt-5 reasoning, prompt caching; "wire openai into a warlock agent", "configure gpt-4o", "use o3 / gpt-5 reasoning effort", "route through openrouter or azure openai", "openai embeddings with warlock", "generate images with gpt-image / dall-e", "send a pdf / audio to gpt-4o"; typical import `import { OpenAISDK } from "@warlock.js/ai-openai"`. Skip: the ai.image verb surface — `@warlock.js/ai/generate-images/SKILL.md`; agent wiring — `@warlock.js/ai/run-ai-agent/SKILL.md`; adapter comparison — `@warlock.js/ai/pick-ai-provider/SKILL.md`; competing adapters `@warlock.js/ai-anthropic`, `@warlock.js/ai-bedrock`, `@warlock.js/ai-google`, `@warlock.js/ai-ollama`; raw `openai` SDK, Vercel `@ai-sdk/openai`.
+- [setup-openai](@warlock.js/ai-openai/setup-openai/SKILL.md): Wire @warlock.js/ai-openai — new OpenAISDK({apiKey, baseURL?, provider?, pricing?}) for OpenAI / Azure / OpenRouter, .model({name, vision?, reasoning?, structuredOutput?, responseFormat?, pdf?, audio?}) for ModelContract, .embedder({name, dimensions?}) for embeddings, .image({name, pricing?}) for gpt-image-*/dall-e-* image generation (via ai.image). PDF input maps to OpenAI file parts (opt-in pdf:true), audio input to input_audio (opt-in audio:true). Triggers: `OpenAISDK`, `.model`, `.embedder`, `.image`, `.embed`, `.embedMany`, `baseURL`, `pricing`, `responseSchema`, `responseFormat`, `reasoning_effort`, `reasoningTokens`, `cachedTokens`, `gpt-image`, `dall-e`, `ai.image`, `pdf input`, `input_audio`, o-series / gpt-5 reasoning, `api: "responses"`, Responses API, reasoning with tools, prompt caching; "wire openai into a warlock agent", "configure gpt-4o", "use o3 / gpt-5 reasoning effort", "route through openrouter or azure openai", "openai embeddings with warlock", "generate images with gpt-image / dall-e", "send a pdf / audio to gpt-4o"; typical import `import { OpenAISDK } from "@warlock.js/ai-openai"`. Skip: the ai.image verb surface — `@warlock.js/ai/generate-images/SKILL.md`; agent wiring — `@warlock.js/ai/run-ai-agent/SKILL.md`; adapter comparison — `@warlock.js/ai/pick-ai-provider/SKILL.md`; competing adapters `@warlock.js/ai-anthropic`, `@warlock.js/ai-bedrock`, `@warlock.js/ai-google`, `@warlock.js/ai-ollama`; raw `openai` SDK, Vercel `@ai-sdk/openai`.

package/package.json

@@ -13,13 +13,13 @@
     "url": "https://github.com/warlockjs/ai-openai"
   },
   "dependencies": {
-    "openai": "^7.4.0",
-    "@warlock.js/logger": "5.25.0"
+    "openai": "^7.23.0",
+    "@warlock.js/logger": "5.27.0"
   },
   "peerDependencies": {
-    "@warlock.js/ai": "5.25.0"
+    "@warlock.js/ai": "5.27.0"
   },
-  "version": "5.25.0",
+  "version": "5.27.0",
   "main": "./cjs/index.cjs",
   "module": "./esm/index.mjs",
   "types": "./esm/index.d.mts",

package/skills/setup-openai/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: setup-openai
-description: 'Wire @warlock.js/ai-openai — new OpenAISDK({apiKey, baseURL?, provider?, pricing?}) for OpenAI / Azure / OpenRouter, .model({name, vision?, reasoning?, structuredOutput?, responseFormat?, pdf?, audio?}) for ModelContract, .embedder({name, dimensions?}) for embeddings, .image({name, pricing?}) for gpt-image-*/dall-e-* image generation (via ai.image). PDF input maps to OpenAI file parts (opt-in pdf:true), audio input to input_audio (opt-in audio:true). Triggers: `OpenAISDK`, `.model`, `.embedder`, `.image`, `.embed`, `.embedMany`, `baseURL`, `pricing`, `responseSchema`, `responseFormat`, `reasoning_effort`, `reasoningTokens`, `cachedTokens`, `gpt-image`, `dall-e`, `ai.image`, `pdf input`, `input_audio`, o-series / gpt-5 reasoning, prompt caching; "wire openai into a warlock agent", "configure gpt-4o", "use o3 / gpt-5 reasoning effort", "route through openrouter or azure openai", "openai embeddings with warlock", "generate images with gpt-image / dall-e", "send a pdf / audio to gpt-4o"; typical import `import { OpenAISDK } from "@warlock.js/ai-openai"`. Skip: the ai.image verb surface — `@warlock.js/ai/generate-images/SKILL.md`; agent wiring — `@warlock.js/ai/run-ai-agent/SKILL.md`; adapter comparison — `@warlock.js/ai/pick-ai-provider/SKILL.md`; competing adapters `@warlock.js/ai-anthropic`, `@warlock.js/ai-bedrock`, `@warlock.js/ai-google`, `@warlock.js/ai-ollama`; raw `openai` SDK, Vercel `@ai-sdk/openai`.'
+description: 'Wire @warlock.js/ai-openai — new OpenAISDK({apiKey, baseURL?, provider?, pricing?}) for OpenAI / Azure / OpenRouter, .model({name, vision?, reasoning?, structuredOutput?, responseFormat?, pdf?, audio?}) for ModelContract, .embedder({name, dimensions?}) for embeddings, .image({name, pricing?}) for gpt-image-*/dall-e-* image generation (via ai.image). PDF input maps to OpenAI file parts (opt-in pdf:true), audio input to input_audio (opt-in audio:true). Triggers: `OpenAISDK`, `.model`, `.embedder`, `.image`, `.embed`, `.embedMany`, `baseURL`, `pricing`, `responseSchema`, `responseFormat`, `reasoning_effort`, `reasoningTokens`, `cachedTokens`, `gpt-image`, `dall-e`, `ai.image`, `pdf input`, `input_audio`, o-series / gpt-5 reasoning, `api: "responses"`, Responses API, reasoning with tools, prompt caching; "wire openai into a warlock agent", "configure gpt-4o", "use o3 / gpt-5 reasoning effort", "route through openrouter or azure openai", "openai embeddings with warlock", "generate images with gpt-image / dall-e", "send a pdf / audio to gpt-4o"; typical import `import { OpenAISDK } from "@warlock.js/ai-openai"`. Skip: the ai.image verb surface — `@warlock.js/ai/generate-images/SKILL.md`; agent wiring — `@warlock.js/ai/run-ai-agent/SKILL.md`; adapter comparison — `@warlock.js/ai/pick-ai-provider/SKILL.md`; competing adapters `@warlock.js/ai-anthropic`, `@warlock.js/ai-bedrock`, `@warlock.js/ai-google`, `@warlock.js/ai-ollama`; raw `openai` SDK, Vercel `@ai-sdk/openai`.'
 ---
 
 # `@warlock.js/ai-openai`
@@ -45,7 +45,7 @@ Returns a `ModelContract` you pass straight into `ai.agent({ model })`.
 | `structuredOutput` | `true`, unless `responseFormat` is set to `"json_object"` or `"text"` (loose modes) — then `false`. |
 | `vision` | Inferred from model name. `true` for `gpt-4o*`, `gpt-4-turbo*`, `gpt-4.1*`, `o1*`, `o3*`, `chatgpt-4o*`; `false` otherwise. |
 | `reasoning` | Inferred from model name. `true` for the o-series (`o1*`, `o3*`, `o4*`) and the `gpt-5*` family; `false` otherwise. Drives whether `reasoning_effort` is forwarded. |
-| `promptCaching` | Always `true`. OpenAI caches long prompt prefixes automatically and reports hits via `usage.cachedTokens` — there are no caller-supplied write breakpoints. |
+| `promptCaching` | `true` for the direct OpenAI provider. It supports automatic cache hits plus explicit content-part write breakpoints; OpenAI-compatible wrapper providers leave the OpenAI-only fields off. |
 | `pdf` | `false` by default (opt-in `.model({ pdf: true })`). OpenAI accepts PDF `file` parts only on specific models (the `gpt-4o` family) — honest off until you set it. |
 | `audio` | `false` by default (opt-in `.model({ audio: true })`). Only the `gpt-4o-audio-preview` family accepts `input_audio`. |
 
@@ -149,7 +149,32 @@ await model.complete(messages, { tools, reasoning: { effort: "high" } }); // you
 await model.complete(messages, { tools, reasoning: { effort: "none" } }); // same effect as the default, explicit
 ```
 
-Calling **without** `tools` is unaffected — `reasoning_effort` is still omitted by default in that case (provider default reasoning, nothing to unblock). Trade-off of the `"none"` default: you lose reasoning on that call. For tool-heavy agent work (function calls + good replies rather than deep analysis) this is the right call — it's the difference between empty replies and working ones. When you need reasoning **and** tools together, use the Responses API (planned).
+The gpt-6 models (`gpt-6-astra` / `gpt-6-sol` / `gpt-6-luna`) are hard-400 cases, verified live on `gpt-6-sol`:
+
+```
+400 Function tools with reasoning_effort are not supported for gpt-6-sol in /v1/chat/completions. To use function tools, use /v1/responses or set reasoning_effort to 'none'.
+```
+
+Tools work with reasoning off (the adapter's default when tools are present and no reasoning is set). For reasoning **and** tools together, use the Responses API below.
+
+Calling **without** `tools` is unaffected — `reasoning_effort` is still omitted by default in that case (provider default reasoning, nothing to unblock). Trade-off of the `"none"` default: you lose reasoning on that call. For tool-heavy agent work (function calls + good replies rather than deep analysis) this is the right call — it's the difference between empty replies and working ones. When you need reasoning **and** tools together, use the Responses API (next section).
+
+### Reasoning + tools: the Responses API (`api: "responses"`)
+
+Opt a direct-OpenAI model into `/v1/responses`, where reasoning models accept function tools with reasoning on:
+
+```ts
+const model = openai.model({ name: "gpt-6-sol", api: "responses" });
+await model.complete(messages, { tools, reasoning: { effort: "high" } }); // reasoning stays on, tools work
+```
+
+- Chat Completions stays the default; nothing changes unless you pass `api: "responses"`.
+- Only the direct `openai` provider accepts it. `@warlock.js/ai-deepseek`, `ai-groq`, `ai-xai` and `ai-mistral` (which reuse this adapter with their own `baseURL`) reject `api: "responses"` with `InvalidRequestError`.
+- With tools attached, reasoning uses the provider default instead of being forced to `"none"`: expect more reasoning tokens and latency than the Chat path. Pass `reasoning: { effort }` to choose.
+- The system prompt is sent as `instructions`; requests use `store: false` (no server-side conversation state).
+- Reasoning is carried across a tool loop without server storage: each response asks for encrypted reasoning, which rides on the first tool call of the turn (`providerMetadata.openaiResponses`) and is sent back on the next request when the model is the same. It is persisted with your conversation history.
+- `complete()` and `stream()` both work; streamed chunks follow the same order as the Chat path (deltas, then tool calls, then `done`). Usage reports `reasoningTokens` and `cachedTokens` the same way.
+- `pdf` and `audio` inputs are not available on this path.
 
 ## Token usage — what's reported
 
@@ -163,7 +188,21 @@ Calling **without** `tools` is unaffected — `reasoning_effort` is still omitte
 
 Both `cachedTokens` and `reasoningTokens` are omitted (not set to `0`) when the provider reports zero, so non-reasoning / uncached calls keep the lean `{ input, output, total }` shape. Reported identically on `complete()` and the streaming `done` event.
 
-`cacheControl` write breakpoints (`ModelCallOptions.cacheControl`) are a **no-op** — OpenAI has no caller-driven cache marker on the Chat Completions wire. The read-side `cachedTokens` accounting above works without any caller action.
+### Explicit prompt caching (direct OpenAI only)
+
+Chat Completions exposes OpenAI-specific cache controls on current SDK types. Configure a stable cache bucket and (when needed) its maximum retention policy per model:
+
+```ts
+const model = openai.model({
+  name: "gpt-5.6",
+  promptCacheKey: "tenant:acme",
+  promptCacheRetention: "24h", // or "in_memory"
+});
+```
+
+`promptCacheKey` improves matching for related requests. `promptCacheRetention` is OpenAI's maximum policy (`"in_memory" | "24h"`). `cacheControl.breakpoints` maps to `prompt_cache_breakpoint: { mode: "explicit" }` on the final user content parts (up to OpenAI's four-request limit); use it when you need to pin a reusable prefix rather than rely only on OpenAI's implicit breakpoint.
+
+These fields are sent only when the SDK provider label is exactly `"openai"`. `DeepSeekSDK`, `GroqSDK`, `XaiSDK`, and `MistralSDK` label their wrapped endpoint differently, so their requests omit every OpenAI-only cache field even though they reuse this transport. Read-side `usage.cachedTokens` accounting remains available whenever the upstream reports it.
 
 ## Token counting