@mikode13/harness 1.0.1 → 2.0.0

package/dist/llm/infrastructure/openAILLMClient.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"openAILLMClient.js","sourceRoot":"","sources":["../../../src/llm/infrastructure/openAILLMClient.ts"],"names":[],"mappings":"AAAA,OAAO,MAAM,EAAE,EAAE,QAAQ,EAAE,MAAM,QAAQ,CAAC;AAO1C,OAAO,EACN,uBAAuB,EACvB,gBAAgB,EAChB,kBAAkB,EAClB,eAAe,GACf,MAAM,8BAA8B,CAAC;AACtC,OAAO,EACN,mBAAmB,EACnB,oBAAoB,EACpB,uBAAuB,EACvB,eAAe,EACf,WAAW,GACX,MAAM,uCAAuC,CAAC;AAC/C,OAAO,EAAE,OAAO,EAAE,MAAM,gCAAgC,CAAC;AAGzD,OAAO,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAC;AAItD,+EAA+E;AAC/E,MAAM,CAAC,MAAM,YAAY,GAAG;IAC3B,aAAa;IACb,aAAa;IACb,cAAc;IACd,eAAe;CACN,CAAC;AAGX,kFAAkF;AAClF,MAAM,qBAAqB,GAAG,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,CAAC,CAAC;AAExD,gGAAgG;AAChG,mDAAmD;AACnD,MAAM,mBAAmB,GAAG;IAC3B,oBAAoB;IACpB,0BAA0B;IAC1B,8BAA8B;CAC9B,CAAC;AAEF,gFAAgF;AAChF,MAAM,2BAA2B,GAAG,CAAC,cAAc,EAAE,qBAAqB,CAAC,CAAC;AAE5E;;;;GAIG;AACH,SAAS,aAAa,CAAC,OAAgB;IACtC,MAAM,IAAI,GAAG,OAAO,CAAC,OAAO;SAC1B,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,IAAI,KAAK,MAAM,CAAC;SACpC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC;SACtB,IAAI,CAAC,IAAI,CAAC,CAAC;IAEb,OAAO,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,OAAO,CAAC,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;AAC5D,CAAC;AAED,SAAS,YAAY,CAAC,IAAwB,EAAE,MAAe;IAC9D,QAAQ,IAAI,CAAC,IAAI,EAAE,CAAC;QACnB,KAAK,SAAS;YACb,4DAA4D;YAC5D,OAAO,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CACrC,OAAO,CAAC,IAAI,KAAK,aAAa,CAAC,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,MAAe,EAAE,IAAI,EAAE,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,EAAE,CACrF,CAAC;QACH,KAAK,WAAW,CAAC,CAAC,CAAC;YAClB,+EAA+E;YAC/E,MAAM,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAClE,OAAO,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,WAAW,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;QAClD,CAAC;QACD;YACC,kFAAkF;YAClF,WAAW,CACV,GAAG,EAAE;gBACJ,MAAM,CAAC,IAAI,CAAC,IAAI,EAAE,wDAAwD,CAAC,CAAC;YAC7E,CAAC,EACD,mBAAmB,EACnB,sDAAsD,CACtD,CAAC;YACF,OAAO,EAAE,CAAC;IACZ,CAAC;AACF,CAAC;AAED,SAAS,YAAY,CAAC,QAAkB;IACvC,IAAI,QAAQ,CAAC,MAAM,KAAK,YAAY,EAAE,CAAC;QACtC,OAAO,QAAQ,CAAC,kBAAkB,EAAE,MAAM,KAAK,gBAAgB,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,WAAW,CAAC;IAC3F,CAAC;IACD,IAAI,QAAQ,CAAC,MAAM,KAAK,WAAW,EAAE,CAAC;QACrC,MAAM,KAAK,GACV,QAAQ,CAAC,KAAK,EAAE,OAAO;YACvB,mCAAmC,QAAQ,CAAC,MAAM,IAAI,SAAS,IAAI,CAAC;QACrE,uFAAuF;QACvF,IAAI,QAAQ,CAAC,KAAK,IAAI,2BAA2B,CAAC,QAAQ,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC;YACjF,MAAM,IAAI,gBAAgB,CAAC,4BAA4B,EAAE,EAAE,KAAK,EAAE,CAAC,CAAC;QACrE,CAAC;QACD,MAAM,IAAI,kBAAkB,CAAC,sCAAsC,EAAE,EAAE,KAAK,EAAE,CAAC,CAAC;IACjF,CAAC;IAED,MAAM,OAAO,GAAG,QAAQ,CAAC,MAAM,CAAC,IAAI,CACnC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,IAAI,KAAK,SAAS,IAAI,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC,OAAO,CAAC,IAAI,KAAK,SAAS,CAAC,CAC3F,CAAC;IACF,OAAO,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,WAAW,CAAC;AAC1C,CAAC;AAED,6EAA6E;AAC7E,SAAS,QAAQ,CAAC,KAAgC,EAAE,MAAe;IAClE,IAAI,CAAC,KAAK,EAAE,CAAC;QACZ,+DAA+D;QAC/D,WAAW,CACV,GAAG,EAAE;YACJ,MAAM,CAAC,IAAI,CAAC,sEAAsE,CAAC,CAAC;QACrF,CAAC,EACD,mBAAmB,EACnB,iDAAiD,CACjD,CAAC;QACF,OAAO,SAAS,CAAC;IAClB,CAAC;IAED,MAAM,EAAE,aAAa,EAAE,kBAAkB,EAAE,GAAG,KAAK,CAAC,oBAAoB,CAAC;IACzE,OAAO;QACN,WAAW,EAAE,KAAK,CAAC,YAAY,GAAG,aAAa,GAAG,kBAAkB;QACpE,eAAe,EAAE,aAAa;QAC9B,kBAAkB,EAAE,kBAAkB;QACtC,YAAY,EAAE,KAAK,CAAC,aAAa;KACjC,CAAC;AACH,CAAC;AAED,SAAS,aAAa,CAAC,QAAkB,EAAE,MAAe;IACzD,IAAI,KAAyB,CAAC;IAE9B,IAAI,CAAC;QACJ,sFAAsF;QACtF,KAAK,GAAG,QAAQ,CAAC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO;YACN,OAAO,EAAE;gBACR,IAAI,EAAE,WAAW;gBACjB,OAAO,EAAE,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,CAAC,YAAY,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;aACpE;YACD,KAAK;YACL,UAAU,EAAE,YAAY,CAAC,QAAQ,CAAC;SAClC,CAAC;IACH,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QAChB,wFAAwF;QACxF,0DAA0D;QAC1D,MAAM,eAAe,CAAC,KAAK,EAAE,KAAK,EAAE,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;IACtD,CAAC;AACF,CAAC;AAED,SAAS,qBAAqB,CAAC,KAAc;IAC5C,IAAI,KAAK,YAAY,QAAQ,EAAE,CAAC;QAC/B,IAAI,KAAK,CAAC,IAAI,KAAK,yBAAyB,EAAE,CAAC;YAC9C,OAAO,IAAI,eAAe,CAAC,6CAA6C,EAAE;gBACzE,KAAK,EAAE,eAAe,CAAC,KAAK,CAAC;aAC7B,CAAC,CAAC;QACJ,CAAC;QACD,IACC,mBAAmB,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,KAAK,KAAK,CAAC,IAAI,CAAC;YACrD,qBAAqB,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM,CAAC,EAC5D,CAAC;YACF,OAAO,IAAI,kBAAkB,CAAC,6BAA6B,EAAE;gBAC5D,KAAK,EAAE,eAAe,CAAC,KAAK,CAAC;aAC7B,CAAC,CAAC;QACJ,CAAC;IACF,CAAC;IAED,OAAO,uBAAuB,CAAC,KAAK,EAAE,2BAA2B,CAAC,CAAC;AACpE,CAAC;AAED;;;;GAIG;AACH,MAAM,OAAO,eAAe;IACV,MAAM,CAAS;IACf,KAAK,CAAc;IACnB,YAAY,CAAS;IACrB,MAAM,CAAU;IAEjC,0FAA0F;IAC1F,YAAY,EACX,KAAK,EACL,YAAY,EACZ,MAAM,GAKN;QACA,IAAI,CAAC,OAAO,CAAC,YAAY,EAAE,KAAK,CAAC,EAAE,CAAC;YACnC,MAAM,IAAI,uBAAuB,CAChC,IAAI,KAAK,8CAA8C,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAChF,CAAC;QACH,CAAC;QAED,IAAI,CAAC;YACJ,mFAAmF;YACnF,8FAA8F;YAC9F,IAAI,CAAC,MAAM,GAAG,IAAI,MAAM,EAAE,CAAC;QAC5B,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,2EAA2E;YAC3E,MAAM,IAAI,kBAAkB,CAAC,uCAAuC,EAAE;gBACrE,KAAK,EAAE,eAAe,CAAC,KAAK,CAAC;aAC7B,CAAC,CAAC;QACJ,CAAC;QAED,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;QACnB,IAAI,CAAC,YAAY,GAAG,YAAY,CAAC;QACjC,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;IACtB,CAAC;IAED,KAAK,CAAC,IAAI,CAAC,OAAkB,EAAE,MAAmB;QACjD,IAAI,QAAkB,CAAC;QAEvB,IAAI,CAAC;YACJ,QAAQ,GAAG,MAAM,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAC5C;gBACC,KAAK,EAAE,IAAI,CAAC,KAAK;gBACjB,YAAY,EAAE,IAAI,CAAC,YAAY;gBAC/B,KAAK,EAAE,OAAO,CAAC,OAAO,CAAC,aAAa,CAAC;gBACrC,KAAK,EAAE,KAAK;gBACZ,SAAS,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE;aAC9B,EACD,EAAE,MAAM,EAAE,CACV,CAAC;QACH,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,mFAAmF;YACnF,MAAM,CAAC,cAAc,EAAE,CAAC;YACxB,MAAM,qBAAqB,CAAC,KAAK,CAAC,CAAC;QACpC,CAAC;QAED,OAAO,WAAW,CACjB,GAAG,EAAE,CAAC,aAAa,CAAC,QAAQ,EAAE,IAAI,CAAC,MAAM,CAAC,EAC1C,oBAAoB,EACpB,gCAAgC,CAChC,CAAC;IACH,CAAC;CACD","sourcesContent":["import OpenAI, { APIError } from 'openai';\nimport type {\n\tResponse,\n\tResponseInputItem,\n\tResponseOutputItem,\n\tResponseUsage,\n} from 'openai/resources/responses/responses';\nimport {\n\tInvalidAgentConfigError,\n\tRecoverableError,\n\tUnrecoverableError,\n\twithSpentTokens,\n} from '../../agent/domain/errors.ts';\nimport {\n\tclassifyHostFailure,\n\tclassifyLocalFailure,\n\tclassifyProviderFailure,\n\tdescribeFailure,\n\ttreatErrors,\n} from '../../agent/domain/providerFailure.ts';\nimport { isOneOf } from '../../shared/domain/isOneOf.ts';\nimport type { ILogger } from '../../shared/domain/logger.ts';\nimport type { Tokens } from '../../shared/domain/tokens.ts';\nimport { MaxContextError } from '../domain/errors.ts';\nimport type { LLMClient, LLMResponse, StopReason } from '../domain/llm.ts';\nimport type { Message, MessagePart } from '../domain/message.ts';\n\n// The SDK types `model` as a plain string, so this list is maintained by hand.\nexport const openAIModels = [\n\t'gpt-6-astra',\n\t'gpt-5.6-sol',\n\t'gpt-5.6-luna',\n\t'gpt-5.6-terra',\n] as const;\nexport type OpenAIModel = (typeof openAIModels)[number];\n\n// The request itself is wrong or not allowed, so sending it again cannot succeed.\nconst unrecoverableStatuses = [400, 401, 403, 404, 422];\n\n// Out of credit or over a spend limit. These arrive as a 429 like a rate limit, but the account\n// stays blocked however often the request is sent.\nconst exhaustedQuotaCodes = [\n\t'insufficient_quota',\n\t'credit_balance_exhausted',\n\t'project_spend_limit_exceeded',\n];\n\n// The codes OpenAI gives a failed response that a new attempt can succeed past.\nconst transientResponseErrorCodes = ['server_error', 'rate_limit_exceeded'];\n\n/**\n * Only the text crosses back. OpenAI replays reasoning only from the encrypted item it\n * returned, which a `Message` does not keep, so a reasoning part stays in the conversation\n * for narration and is left out of the request. A message with no text is left out whole.\n */\nfunction toOpenAIInput(message: Message): ResponseInputItem[] {\n\tconst text = message.content\n\t\t.filter(part => part.type === 'text')\n\t\t.map(part => part.text)\n\t\t.join('\\n');\n\n\treturn text ? [{ role: message.role, content: text }] : [];\n}\n\nfunction describeItem(item: ResponseOutputItem, logger: ILogger): MessagePart[] {\n\tswitch (item.type) {\n\t\tcase 'message':\n\t\t\t// A refusal is not an answer: it only sets the stop reason.\n\t\t\treturn item.content.flatMap(content =>\n\t\t\t\tcontent.type === 'output_text' ? [{ type: 'text' as const, text: content.text }] : [],\n\t\t\t);\n\t\tcase 'reasoning': {\n\t\t\t// The summary is the readable part; the encrypted content is opaque by design.\n\t\t\tconst text = item.summary.map(summary => summary.text).join('\\n');\n\t\t\treturn text ? [{ type: 'reasoning', text }] : [];\n\t\t}\n\t\tdefault:\n\t\t\t// No tools are sent yet, so any other item is output this client does not expect.\n\t\t\ttreatErrors(\n\t\t\t\t() => {\n\t\t\t\t\tlogger.warn(item, 'OpenAI returned an output item the client does not map');\n\t\t\t\t},\n\t\t\t\tclassifyHostFailure,\n\t\t\t\t'OpenAI client logger failed while mapping a response',\n\t\t\t);\n\t\t\treturn [];\n\t}\n}\n\nfunction toStopReason(response: Response): StopReason {\n\tif (response.status === 'incomplete') {\n\t\treturn response.incomplete_details?.reason === 'content_filter' ? 'refused' : 'truncated';\n\t}\n\tif (response.status !== 'completed') {\n\t\tconst cause =\n\t\t\tresponse.error?.message ??\n\t\t\t`The response ended with status \"${response.status ?? 'unknown'}\".`;\n\t\t// The same failure is retried when it arrives as an SDK error, so it must be here too.\n\t\tif (response.error && transientResponseErrorCodes.includes(response.error.code)) {\n\t\t\tthrow new RecoverableError('The OpenAI response failed', { cause });\n\t\t}\n\t\tthrow new UnrecoverableError('The OpenAI response did not complete', { cause });\n\t}\n\n\tconst refused = response.output.some(\n\t\titem => item.type === 'message' && item.content.some(content => content.type === 'refusal'),\n\t);\n\treturn refused ? 'refused' : 'completed';\n}\n\n/** OpenAI counts both cache reads and cache writes inside `input_tokens`. */\nfunction toTokens(usage: ResponseUsage | undefined, logger: ILogger): Tokens | undefined {\n\tif (!usage) {\n\t\t// Zeros would report a call that may have been billed as free.\n\t\ttreatErrors(\n\t\t\t() => {\n\t\t\t\tlogger.warn('OpenAI returned a response without usage; it cannot be accounted for');\n\t\t\t},\n\t\t\tclassifyHostFailure,\n\t\t\t'OpenAI client logger failed while reading usage',\n\t\t);\n\t\treturn undefined;\n\t}\n\n\tconst { cached_tokens, cache_write_tokens } = usage.input_tokens_details;\n\treturn {\n\t\tinputTokens: usage.input_tokens - cached_tokens - cache_write_tokens,\n\t\treadCacheTokens: cached_tokens,\n\t\twrittenCacheTokens: cache_write_tokens,\n\t\toutputTokens: usage.output_tokens,\n\t};\n}\n\nfunction toLLMResponse(response: Response, logger: ILogger): LLMResponse {\n\tlet usage: Tokens | undefined;\n\n\ttry {\n\t\t// Inside the try: it warns when the usage is missing, and that warning can throw too.\n\t\tusage = toTokens(response.usage, logger);\n\t\treturn {\n\t\t\tmessage: {\n\t\t\t\trole: 'assistant',\n\t\t\t\tcontent: response.output.flatMap(item => describeItem(item, logger)),\n\t\t\t},\n\t\t\tusage,\n\t\t\tstopReason: toStopReason(response),\n\t\t};\n\t} catch (error) {\n\t\t// The call was billed whether the response failed or a logger did while mapping it, and\n\t\t// an answer without usage leaves the run's count unknown.\n\t\tthrow withSpentTokens(error, usage, !response.usage);\n\t}\n}\n\nfunction classifyOpenAIFailure(error: unknown): Error {\n\tif (error instanceof APIError) {\n\t\tif (error.code === 'context_length_exceeded') {\n\t\t\treturn new MaxContextError('The context no longer fits the OpenAI model', {\n\t\t\t\tcause: describeFailure(error),\n\t\t\t});\n\t\t}\n\t\tif (\n\t\t\texhaustedQuotaCodes.some(code => code === error.code) ||\n\t\t\tunrecoverableStatuses.some(status => status === error.status)\n\t\t) {\n\t\t\treturn new UnrecoverableError('OpenAI rejected the request', {\n\t\t\t\tcause: describeFailure(error),\n\t\t\t});\n\t\t}\n\t}\n\n\treturn classifyProviderFailure(error, 'The OpenAI request failed');\n}\n\n/**\n * A stateless `LLMClient` on the OpenAI Responses API. With `store: false` OpenAI keeps\n * nothing between calls, so every call carries the whole context: the conversation is\n * MiKode's, not the provider's.\n */\nexport class OpenAILLMClient implements LLMClient {\n\tprivate readonly client: OpenAI;\n\tprivate readonly model: OpenAIModel;\n\tprivate readonly systemPrompt: string;\n\tprivate readonly logger: ILogger;\n\n\t// `model` is untyped on purpose: this adapter is the one that knows what OpenAI supports.\n\tconstructor({\n\t\tmodel,\n\t\tsystemPrompt,\n\t\tlogger,\n\t}: {\n\t\tmodel: string;\n\t\tsystemPrompt: string;\n\t\tlogger: ILogger;\n\t}) {\n\t\tif (!isOneOf(openAIModels, model)) {\n\t\t\tthrow new InvalidAgentConfigError(\n\t\t\t\t`\"${model}\" is not an OpenAI model; expected one of: ${openAIModels.join(', ')}`,\n\t\t\t);\n\t\t}\n\n\t\ttry {\n\t\t\t// The SDK's default retries stay on: they back off and honour `Retry-After`, which\n\t\t\t// `RetryingAgent` does not. See \"The OpenAI SDK keeps its transport retries\" in decisions.md.\n\t\t\tthis.client = new OpenAI();\n\t\t} catch (error) {\n\t\t\t// A missing API key, for example: building the client again cannot fix it.\n\t\t\tthrow new UnrecoverableError('The OpenAI client could not be set up', {\n\t\t\t\tcause: describeFailure(error),\n\t\t\t});\n\t\t}\n\n\t\tthis.model = model;\n\t\tthis.systemPrompt = systemPrompt;\n\t\tthis.logger = logger;\n\t}\n\n\tasync send(context: Message[], signal: AbortSignal): Promise<LLMResponse> {\n\t\tlet response: Response;\n\n\t\ttry {\n\t\t\tresponse = await this.client.responses.create(\n\t\t\t\t{\n\t\t\t\t\tmodel: this.model,\n\t\t\t\t\tinstructions: this.systemPrompt,\n\t\t\t\t\tinput: context.flatMap(toOpenAIInput),\n\t\t\t\t\tstore: false,\n\t\t\t\t\treasoning: { summary: 'auto' },\n\t\t\t\t},\n\t\t\t\t{ signal },\n\t\t\t);\n\t\t} catch (error) {\n\t\t\t// The SDK's abort error is named 'Error', so it would pass for a provider failure.\n\t\t\tsignal.throwIfAborted();\n\t\t\tthrow classifyOpenAIFailure(error);\n\t\t}\n\n\t\treturn treatErrors(\n\t\t\t() => toLLMResponse(response, this.logger),\n\t\t\tclassifyLocalFailure,\n\t\t\t'OpenAI response mapping failed',\n\t\t);\n\t}\n}\n"]}

package/dist/orchestration/domain/model/orchestratorAgent.d.ts

@@ -17,7 +17,8 @@ export declare class OrchestratorAgent implements Agent {
         maxAttempts?: number;
         logger: ILogger;
     });
-    run(prompt: string, signal: AbortSignal, callback: Callback): Promise<AgentResponse | undefined>;
+    run(prompt: string, signal: AbortSignal, callback: Callback): Promise<AgentResponse>;
+    private runRounds;
     private getReviewerDecision;
     private warn;
 }

package/dist/orchestration/domain/model/orchestratorAgent.js

@@ -1,4 +1,5 @@
-import { RecoverableError, UnrecoverableError } from "../../../agent/domain/errors.js";
+import { addTokens } from "../../../shared/domain/tokens.js";
+import { RecoverableError, UnrecoverableError, withSpentTokens, } from "../../../agent/domain/errors.js";
 import { classifyHostFailure, treatErrors } from "../../../agent/domain/providerFailure.js";
 const getPlannerPrompt = (userPrompt, previousFailureReason) => {
     const feedback = previousFailureReason
@@ -43,17 +44,22 @@ Executor response (context only; verify it independently):
 ${executorResult}
 ---${retryNotice}`;
 };
+// Every role's response counts, even an empty one that starts another round: it was billed.
 function addToTotals(totals, response) {
-    totals.duration += response.duration;
-    totals.inputTokens += response.inputTokens;
-    totals.outputTokens += response.outputTokens;
+    if (!response.tokens)
+        totals.unreported = true;
+    totals.tokens = addTokens(totals.tokens, response.tokens);
+}
+/** Unknown rather than partial: a sum missing a billed call must not look complete. */
+function runTokens(totals) {
+    return totals.unreported ? undefined : totals.tokens;
 }
 function stripCodeFence(text) {
     const match = /^```(?:json)?\s*([\s\S]*?)\s*```$/.exec(text.trim());
     return match?.[1] ?? text;
 }
 function parseReviewerDecision(response, reviewerDecisionValidator) {
-    if (!response?.response) {
+    if (!response.response) {
         throw new RecoverableError("There's no decision from the reviewer, something went wrong", {
             cause: 'Reviewer decision is missing.',
         });
@@ -94,13 +100,30 @@ export class OrchestratorAgent {
         this.logger = logger;
     }
     async run(prompt, signal, callback) {
+        const start = Date.now();
         // Per invocation, not per instance: the CLI keeps one orchestrator for a whole session.
-        const totals = { duration: 0, inputTokens: 0, outputTokens: 0 };
+        const totals = { tokens: undefined, unreported: false };
+        try {
+            await this.runRounds(prompt, signal, callback, totals);
+        }
+        catch (error) {
+            // A failing role carries its own tokens; the roles before it are in the totals.
+            throw withSpentTokens(error, totals.tokens, totals.unreported);
+        }
+        return {
+            response: 'All job has finished',
+            tokens: runTokens(totals),
+            // The whole run's wall clock, so it matches what the consumer waited.
+            duration: (Date.now() - start) / 1000,
+        };
+    }
+    async runRounds(prompt, signal, callback, totals) {
         let lastFailureReason;
         for (let attempt = 1; attempt <= this.maxAttempts; attempt++) {
             const isLastAttempt = attempt === this.maxAttempts;
             const plannerResponse = await this.plannerAgent.run(getPlannerPrompt(prompt, lastFailureReason), signal, callback);
-            if (!plannerResponse?.response) {
+            addToTotals(totals, plannerResponse);
+            if (!plannerResponse.response) {
                 lastFailureReason = 'The planner produced no response.';
                 if (isLastAttempt) {
                     throw new UnrecoverableError('Max attempts exhausted', { cause: lastFailureReason });
@@ -108,9 +131,9 @@ export class OrchestratorAgent {
                 this.warn(`Attempt ${String(attempt)}/${String(this.maxAttempts)}: the planner produced no response; starting another round`);
                 continue;
             }
-            addToTotals(totals, plannerResponse);
             const executorResponse = await this.executorAgent.run(getExecutorPrompt(prompt, plannerResponse.response), signal, callback);
-            if (!executorResponse?.response) {
+            addToTotals(totals, executorResponse);
+            if (!executorResponse.response) {
                 lastFailureReason = 'The executor produced no response.';
                 if (isLastAttempt) {
                     throw new UnrecoverableError('Max attempts exhausted', { cause: lastFailureReason });
@@ -118,11 +141,9 @@ export class OrchestratorAgent {
                 this.warn(`Attempt ${String(attempt)}/${String(this.maxAttempts)}: the executor produced no response; starting another round`);
                 continue;
             }
-            addToTotals(totals, executorResponse);
             const reviewerDecision = await this.getReviewerDecision(prompt, plannerResponse.response, executorResponse.response, signal, callback, totals);
-            if (reviewerDecision.decision === 'approved') {
-                return { response: 'All job has finished', ...totals };
-            }
+            if (reviewerDecision.decision === 'approved')
+                return;
             lastFailureReason = reviewerDecision.feedback;
             if (isLastAttempt) {
                 throw new UnrecoverableError('Max attempts exhausted', { cause: lastFailureReason });
@@ -140,9 +161,7 @@ export class OrchestratorAgent {
         let parseFailureReason;
         for (let attempt = 1; attempt <= this.maxAttempts; attempt++) {
             const reviewerResponse = await this.reviewerAgent.run(getReviewerPrompt(userPrompt, plannerPrompt, executorResult, parseFailureReason), signal, callback);
-            if (reviewerResponse) {
-                addToTotals(totals, reviewerResponse);
-            }
+            addToTotals(totals, reviewerResponse);
             try {
                 return parseReviewerDecision(reviewerResponse, this.reviewerDecisionValidator);
             }

package/dist/orchestration/domain/model/orchestratorAgent.js.map

@@ -1 +1 @@
-{"version":3,"file":"orchestratorAgent.js","sourceRoot":"","sources":["../../../../src/orchestration/domain/model/orchestratorAgent.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,gBAAgB,EAAE,kBAAkB,EAAE,MAAM,iCAAiC,CAAC;AAIvF,OAAO,EAAE,mBAAmB,EAAE,WAAW,EAAE,MAAM,0CAA0C,CAAC;AAE5F,MAAM,gBAAgB,GAAG,CAAC,UAAkB,EAAE,qBAA8B,EAAE,EAAE;IAC/E,MAAM,QAAQ,GAAG,qBAAqB;QACrC,CAAC,CAAC,iDAAiD,qBAAqB,OAAO;QAC/E,CAAC,CAAC,EAAE,CAAC;IAEN,OAAO;;;;EAIN,UAAU;KACP,QAAQ,EAAE,CAAC;AAChB,CAAC,CAAC;AAEF,MAAM,iBAAiB,GAAG,CAAC,UAAkB,EAAE,aAAqB,EAAE,EAAE,CACvE;;;;EAIC,UAAU;;;;;EAKV,aAAa;IACX,CAAC;AAEL,MAAM,iBAAiB,GAAG,CACzB,UAAkB,EAClB,aAAqB,EACrB,cAAsB,EACtB,kBAA2B,EAC1B,EAAE;IACH,MAAM,WAAW,GAAG,kBAAkB;QACrC,CAAC,CAAC,iDAAiD,kBAAkB,6GAA6G;QAClL,CAAC,CAAC,EAAE,CAAC;IAEN,OAAO;;;;EAIN,UAAU;;;;;EAKV,aAAa;;;;;EAKb,cAAc;KACX,WAAW,EAAE,CAAC;AACnB,CAAC,CAAC;AAQF,SAAS,WAAW,CAAC,MAAiB,EAAE,QAAuB;IAC9D,MAAM,CAAC,QAAQ,IAAI,QAAQ,CAAC,QAAQ,CAAC;IACrC,MAAM,CAAC,WAAW,IAAI,QAAQ,CAAC,WAAW,CAAC;IAC3C,MAAM,CAAC,YAAY,IAAI,QAAQ,CAAC,YAAY,CAAC;AAC9C,CAAC;AAED,SAAS,cAAc,CAAC,IAAY;IACnC,MAAM,KAAK,GAAG,mCAAmC,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC;IACpE,OAAO,KAAK,EAAE,CAAC,CAAC,CAAC,IAAI,IAAI,CAAC;AAC3B,CAAC;AAED,SAAS,qBAAqB,CAC7B,QAAmC,EACnC,yBAAsD;IAEtD,IAAI,CAAC,QAAQ,EAAE,QAAQ,EAAE,CAAC;QACzB,MAAM,IAAI,gBAAgB,CAAC,6DAA6D,EAAE;YACzF,KAAK,EAAE,+BAA+B;SACtC,CAAC,CAAC;IACJ,CAAC;IAED,IAAI,cAAuB,CAAC;IAC5B,IAAI,CAAC;QACJ,cAAc,GAAG,IAAI,CAAC,KAAK,CAAC,cAAc,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC,CAAC;IAChE,CAAC;IAAC,MAAM,CAAC;QACR,MAAM,IAAI,gBAAgB,CAAC,uCAAuC,EAAE;YACnE,KAAK,EAAE,uCAAuC;SAC9C,CAAC,CAAC;IACJ,CAAC;IAED,MAAM,QAAQ,GAAG,yBAAyB,CAAC,QAAQ,CAAC,cAAc,CAAC,CAAC;IACpE,IAAI,CAAC,QAAQ,EAAE,CAAC;QACf,MAAM,IAAI,gBAAgB,CAAC,uCAAuC,EAAE;YACnE,KAAK,EAAE,sDAAsD;SAC7D,CAAC,CAAC;IACJ,CAAC;IAED,OAAO,QAAQ,CAAC;AACjB,CAAC;AAED,MAAM,OAAO,iBAAiB;IACrB,YAAY,CAAQ;IACpB,aAAa,CAAQ;IACrB,aAAa,CAAQ;IACrB,yBAAyB,CAA8B;IACvD,WAAW,CAAS;IACpB,MAAM,CAAU;IAExB,YAAY,EACX,YAAY,EACZ,aAAa,EACb,aAAa,EACb,yBAAyB,EACzB,WAAW,GAAG,CAAC,EACf,MAAM,GAQN;QACA,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,WAAW,CAAC,IAAI,WAAW,GAAG,CAAC,EAAE,CAAC;YACvD,MAAM,IAAI,UAAU,CAAC,wCAAwC,CAAC,CAAC;QAChE,CAAC;QAED,IAAI,CAAC,YAAY,GAAG,YAAY,CAAC;QACjC,IAAI,CAAC,aAAa,GAAG,aAAa,CAAC;QACnC,IAAI,CAAC,aAAa,GAAG,aAAa,CAAC;QACnC,IAAI,CAAC,yBAAyB,GAAG,yBAAyB,CAAC;QAC3D,IAAI,CAAC,WAAW,GAAG,WAAW,CAAC;QAC/B,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;IACtB,CAAC;IAED,KAAK,CAAC,GAAG,CACR,MAAc,EACd,MAAmB,EACnB,QAAkB;QAElB,wFAAwF;QACxF,MAAM,MAAM,GAAc,EAAE,QAAQ,EAAE,CAAC,EAAE,WAAW,EAAE,CAAC,EAAE,YAAY,EAAE,CAAC,EAAE,CAAC;QAC3E,IAAI,iBAAqC,CAAC;QAE1C,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,IAAI,CAAC,WAAW,EAAE,OAAO,EAAE,EAAE,CAAC;YAC9D,MAAM,aAAa,GAAG,OAAO,KAAK,IAAI,CAAC,WAAW,CAAC;YAEnD,MAAM,eAAe,GAAG,MAAM,IAAI,CAAC,YAAY,CAAC,GAAG,CAClD,gBAAgB,CAAC,MAAM,EAAE,iBAAiB,CAAC,EAC3C,MAAM,EACN,QAAQ,CACR,CAAC;YAEF,IAAI,CAAC,eAAe,EAAE,QAAQ,EAAE,CAAC;gBAChC,iBAAiB,GAAG,mCAAmC,CAAC;gBACxD,IAAI,aAAa,EAAE,CAAC;oBACnB,MAAM,IAAI,kBAAkB,CAAC,wBAAwB,EAAE,EAAE,KAAK,EAAE,iBAAiB,EAAE,CAAC,CAAC;gBACtF,CAAC;gBACD,IAAI,CAAC,IAAI,CACR,WAAW,MAAM,CAAC,OAAO,CAAC,IAAI,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,4DAA4D,CAClH,CAAC;gBACF,SAAS;YACV,CAAC;YAED,WAAW,CAAC,MAAM,EAAE,eAAe,CAAC,CAAC;YAErC,MAAM,gBAAgB,GAAG,MAAM,IAAI,CAAC,aAAa,CAAC,GAAG,CACpD,iBAAiB,CAAC,MAAM,EAAE,eAAe,CAAC,QAAQ,CAAC,EACnD,MAAM,EACN,QAAQ,CACR,CAAC;YAEF,IAAI,CAAC,gBAAgB,EAAE,QAAQ,EAAE,CAAC;gBACjC,iBAAiB,GAAG,oCAAoC,CAAC;gBACzD,IAAI,aAAa,EAAE,CAAC;oBACnB,MAAM,IAAI,kBAAkB,CAAC,wBAAwB,EAAE,EAAE,KAAK,EAAE,iBAAiB,EAAE,CAAC,CAAC;gBACtF,CAAC;gBAED,IAAI,CAAC,IAAI,CACR,WAAW,MAAM,CAAC,OAAO,CAAC,IAAI,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,6DAA6D,CACnH,CAAC;gBACF,SAAS;YACV,CAAC;YAED,WAAW,CAAC,MAAM,EAAE,gBAAgB,CAAC,CAAC;YAEtC,MAAM,gBAAgB,GAAG,MAAM,IAAI,CAAC,mBAAmB,CACtD,MAAM,EACN,eAAe,CAAC,QAAQ,EACxB,gBAAgB,CAAC,QAAQ,EACzB,MAAM,EACN,QAAQ,EACR,MAAM,CACN,CAAC;YAEF,IAAI,gBAAgB,CAAC,QAAQ,KAAK,UAAU,EAAE,CAAC;gBAC9C,OAAO,EAAE,QAAQ,EAAE,sBAAsB,EAAE,GAAG,MAAM,EAAE,CAAC;YACxD,CAAC;YAED,iBAAiB,GAAG,gBAAgB,CAAC,QAAQ,CAAC;YAC9C,IAAI,aAAa,EAAE,CAAC;gBACnB,MAAM,IAAI,kBAAkB,CAAC,wBAAwB,EAAE,EAAE,KAAK,EAAE,iBAAiB,EAAE,CAAC,CAAC;YACtF,CAAC;YAED,IAAI,CAAC,IAAI,CACR,WAAW,MAAM,CAAC,OAAO,CAAC,IAAI,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,uEAAuE,EAC7H,iBAAiB,CACjB,CAAC;QACH,CAAC;QAED,MAAM,IAAI,kBAAkB,CAAC,wBAAwB,EAAE;YACtD,KAAK,EAAE,iBAAiB,IAAI,kBAAkB;SAC9C,CAAC,CAAC;IACJ,CAAC;IAED,6EAA6E;IAC7E,6EAA6E;IAC7E,6CAA6C;IACrC,KAAK,CAAC,mBAAmB,CAChC,UAAkB,EAClB,aAAqB,EACrB,cAAsB,EACtB,MAAmB,EACnB,QAAkB,EAClB,MAAiB;QAEjB,IAAI,kBAAsC,CAAC;QAE3C,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,IAAI,CAAC,WAAW,EAAE,OAAO,EAAE,EAAE,CAAC;YAC9D,MAAM,gBAAgB,GAAG,MAAM,IAAI,CAAC,aAAa,CAAC,GAAG,CACpD,iBAAiB,CAAC,UAAU,EAAE,aAAa,EAAE,cAAc,EAAE,kBAAkB,CAAC,EAChF,MAAM,EACN,QAAQ,CACR,CAAC;YAEF,IAAI,gBAAgB,EAAE,CAAC;gBACtB,WAAW,CAAC,MAAM,EAAE,gBAAgB,CAAC,CAAC;YACvC,CAAC;YAED,IAAI,CAAC;gBACJ,OAAO,qBAAqB,CAAC,gBAAgB,EAAE,IAAI,CAAC,yBAAyB,CAAC,CAAC;YAChF,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBAChB,IAAI,CAAC,CAAC,KAAK,YAAY,gBAAgB,CAAC;oBAAE,MAAM,KAAK,CAAC;gBAEtD,kBAAkB,GAAG,KAAK,CAAC,KAAK,CAAC;gBACjC,IAAI,OAAO,KAAK,IAAI,CAAC,WAAW,EAAE,CAAC;oBAClC,MAAM,IAAI,kBAAkB,CAAC,wBAAwB,EAAE,EAAE,KAAK,EAAE,kBAAkB,EAAE,CAAC,CAAC;gBACvF,CAAC;gBAED,IAAI,CAAC,IAAI,CACR,qBAAqB,MAAM,CAAC,OAAO,CAAC,IAAI,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,0CAA0C,EAC1G,kBAAkB,CAClB,CAAC;YACH,CAAC;QACF,CAAC;QAED,MAAM,IAAI,kBAAkB,CAAC,wBAAwB,EAAE;YACtD,KAAK,EAAE,kBAAkB,IAAI,kBAAkB;SAC/C,CAAC,CAAC;IACJ,CAAC;IAEO,IAAI,CAAC,GAAG,IAAe;QAC9B,WAAW,CACV,GAAG,EAAE;YACJ,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC,CAAC;QAC3B,CAAC,EACD,mBAAmB,EACnB,oDAAoD,CACpD,CAAC;IACH,CAAC;CACD","sourcesContent":["import type { Agent, AgentResponse, Callback } from '../../../agent/domain/agent.ts';\nimport { RecoverableError, UnrecoverableError } from '../../../agent/domain/errors.ts';\nimport type { ReviewerDecision } from './reviewerDecision.ts';\nimport type { Validator } from '../interface/validator.ts';\nimport type { ILogger } from '../../../shared/domain/logger.ts';\nimport { classifyHostFailure, treatErrors } from '../../../agent/domain/providerFailure.ts';\n\nconst getPlannerPrompt = (userPrompt: string, previousFailureReason?: string) => {\n\tconst feedback = previousFailureReason\n\t\t? `\\n\\nFeedback from the previous attempt:\\n---\\n${previousFailureReason}\\n---`\n\t\t: '';\n\n\treturn `You are the planner agent. Read the relevant repository code and create a concise, engineering-grade plan for the executor. Cover the requested behaviour, structural issues, affected files or symbols, API contracts and boundaries, important failure paths, and meaningful tests. Keep the plan focused on the current request and do not require an architecture redesign yet.\n\nOriginal user request:\n---\n${userPrompt}\n---${feedback}`;\n};\n\nconst getExecutorPrompt = (userPrompt: string, plannerPrompt: string) =>\n\t`You are the executor agent. Read the relevant repository code and implement the original user request according to the planner's current plan. Preserve contracts and boundaries, handle important failure paths, keep the change focused, and add or update meaningful tests. Inspect and change the code; do not merely describe what should be done.\n\nOriginal user request:\n---\n${userPrompt}\n---\n\nCurrent implementation plan:\n---\n${plannerPrompt}\n---`;\n\nconst getReviewerPrompt = (\n\tuserPrompt: string,\n\tplannerPrompt: string,\n\texecutorResult: string,\n\tparseFailureReason?: string,\n) => {\n\tconst retryNotice = parseFailureReason\n\t\t? `\\n\\nYour previous response could not be used: ${parseFailureReason} Respond with JSON only, matching the schema exactly, with no surrounding text and no markdown code fences.`\n\t\t: '';\n\n\treturn `You are the reviewer agent. Independently inspect the repository, the current diff, and relevant surrounding code; do not rely on the executor's narrative. Evaluate the original request first; plan compliance is secondary and provides supporting context. Check correctness and logic errors, important failure paths, unused or artificial abstractions, races or shared state, API contract breaks, boundary violations, regressions, scope, and test quality. If it is correct, respond with JSON only: {\"decision\":\"approved\"}. If it is not correct, respond with JSON only: {\"decision\":\"rejected\",\"feedback\":\"list concrete, prioritized findings and the required direction for each\"}. The feedback must contain actionable engineering findings, not a general summary.\n\nOriginal user request:\n---\n${userPrompt}\n---\n\nPlanner's current plan:\n---\n${plannerPrompt}\n---\n\nExecutor response (context only; verify it independently):\n---\n${executorResult}\n---${retryNotice}`;\n};\n\ninterface RunTotals {\n\tduration: number;\n\tinputTokens: number;\n\toutputTokens: number;\n}\n\nfunction addToTotals(totals: RunTotals, response: AgentResponse): void {\n\ttotals.duration += response.duration;\n\ttotals.inputTokens += response.inputTokens;\n\ttotals.outputTokens += response.outputTokens;\n}\n\nfunction stripCodeFence(text: string): string {\n\tconst match = /^```(?:json)?\\s*([\\s\\S]*?)\\s*```$/.exec(text.trim());\n\treturn match?.[1] ?? text;\n}\n\nfunction parseReviewerDecision(\n\tresponse: AgentResponse | undefined,\n\treviewerDecisionValidator: Validator<ReviewerDecision>,\n): ReviewerDecision {\n\tif (!response?.response) {\n\t\tthrow new RecoverableError(\"There's no decision from the reviewer, something went wrong\", {\n\t\t\tcause: 'Reviewer decision is missing.',\n\t\t});\n\t}\n\n\tlet parsedResponse: unknown;\n\ttry {\n\t\tparsedResponse = JSON.parse(stripCodeFence(response.response));\n\t} catch {\n\t\tthrow new RecoverableError('Reviewer returned an invalid decision', {\n\t\t\tcause: 'Reviewer response must be valid JSON.',\n\t\t});\n\t}\n\n\tconst decision = reviewerDecisionValidator.validate(parsedResponse);\n\tif (!decision) {\n\t\tthrow new RecoverableError('Reviewer returned an invalid decision', {\n\t\t\tcause: 'Reviewer response did not match the decision schema.',\n\t\t});\n\t}\n\n\treturn decision;\n}\n\nexport class OrchestratorAgent implements Agent {\n\tprivate plannerAgent: Agent;\n\tprivate executorAgent: Agent;\n\tprivate reviewerAgent: Agent;\n\tprivate reviewerDecisionValidator: Validator<ReviewerDecision>;\n\tprivate maxAttempts: number;\n\tprivate logger: ILogger;\n\n\tconstructor({\n\t\tplannerAgent,\n\t\texecutorAgent,\n\t\treviewerAgent,\n\t\treviewerDecisionValidator,\n\t\tmaxAttempts = 3,\n\t\tlogger,\n\t}: {\n\t\tplannerAgent: Agent;\n\t\texecutorAgent: Agent;\n\t\treviewerAgent: Agent;\n\t\treviewerDecisionValidator: Validator<ReviewerDecision>;\n\t\tmaxAttempts?: number;\n\t\tlogger: ILogger;\n\t}) {\n\t\tif (!Number.isInteger(maxAttempts) || maxAttempts < 1) {\n\t\t\tthrow new RangeError('maxAttempts must be a positive integer');\n\t\t}\n\n\t\tthis.plannerAgent = plannerAgent;\n\t\tthis.executorAgent = executorAgent;\n\t\tthis.reviewerAgent = reviewerAgent;\n\t\tthis.reviewerDecisionValidator = reviewerDecisionValidator;\n\t\tthis.maxAttempts = maxAttempts;\n\t\tthis.logger = logger;\n\t}\n\n\tasync run(\n\t\tprompt: string,\n\t\tsignal: AbortSignal,\n\t\tcallback: Callback,\n\t): Promise<AgentResponse | undefined> {\n\t\t// Per invocation, not per instance: the CLI keeps one orchestrator for a whole session.\n\t\tconst totals: RunTotals = { duration: 0, inputTokens: 0, outputTokens: 0 };\n\t\tlet lastFailureReason: string | undefined;\n\n\t\tfor (let attempt = 1; attempt <= this.maxAttempts; attempt++) {\n\t\t\tconst isLastAttempt = attempt === this.maxAttempts;\n\n\t\t\tconst plannerResponse = await this.plannerAgent.run(\n\t\t\t\tgetPlannerPrompt(prompt, lastFailureReason),\n\t\t\t\tsignal,\n\t\t\t\tcallback,\n\t\t\t);\n\n\t\t\tif (!plannerResponse?.response) {\n\t\t\t\tlastFailureReason = 'The planner produced no response.';\n\t\t\t\tif (isLastAttempt) {\n\t\t\t\t\tthrow new UnrecoverableError('Max attempts exhausted', { cause: lastFailureReason });\n\t\t\t\t}\n\t\t\t\tthis.warn(\n\t\t\t\t\t`Attempt ${String(attempt)}/${String(this.maxAttempts)}: the planner produced no response; starting another round`,\n\t\t\t\t);\n\t\t\t\tcontinue;\n\t\t\t}\n\n\t\t\taddToTotals(totals, plannerResponse);\n\n\t\t\tconst executorResponse = await this.executorAgent.run(\n\t\t\t\tgetExecutorPrompt(prompt, plannerResponse.response),\n\t\t\t\tsignal,\n\t\t\t\tcallback,\n\t\t\t);\n\n\t\t\tif (!executorResponse?.response) {\n\t\t\t\tlastFailureReason = 'The executor produced no response.';\n\t\t\t\tif (isLastAttempt) {\n\t\t\t\t\tthrow new UnrecoverableError('Max attempts exhausted', { cause: lastFailureReason });\n\t\t\t\t}\n\n\t\t\t\tthis.warn(\n\t\t\t\t\t`Attempt ${String(attempt)}/${String(this.maxAttempts)}: the executor produced no response; starting another round`,\n\t\t\t\t);\n\t\t\t\tcontinue;\n\t\t\t}\n\n\t\t\taddToTotals(totals, executorResponse);\n\n\t\t\tconst reviewerDecision = await this.getReviewerDecision(\n\t\t\t\tprompt,\n\t\t\t\tplannerResponse.response,\n\t\t\t\texecutorResponse.response,\n\t\t\t\tsignal,\n\t\t\t\tcallback,\n\t\t\t\ttotals,\n\t\t\t);\n\n\t\t\tif (reviewerDecision.decision === 'approved') {\n\t\t\t\treturn { response: 'All job has finished', ...totals };\n\t\t\t}\n\n\t\t\tlastFailureReason = reviewerDecision.feedback;\n\t\t\tif (isLastAttempt) {\n\t\t\t\tthrow new UnrecoverableError('Max attempts exhausted', { cause: lastFailureReason });\n\t\t\t}\n\n\t\t\tthis.warn(\n\t\t\t\t`Attempt ${String(attempt)}/${String(this.maxAttempts)}: the reviewer rejected the round; starting another with its feedback`,\n\t\t\t\tlastFailureReason,\n\t\t\t);\n\t\t}\n\n\t\tthrow new UnrecoverableError('Max attempts exhausted', {\n\t\t\tcause: lastFailureReason ?? 'Unknown failure.',\n\t\t});\n\t}\n\n\t// Retries only the reviewer call on a malformed decision, instead of redoing\n\t// planning/execution — a bad or unparsable reviewer response is not evidence\n\t// the plan or the implementation were wrong.\n\tprivate async getReviewerDecision(\n\t\tuserPrompt: string,\n\t\tplannerPrompt: string,\n\t\texecutorResult: string,\n\t\tsignal: AbortSignal,\n\t\tcallback: Callback,\n\t\ttotals: RunTotals,\n\t): Promise<ReviewerDecision> {\n\t\tlet parseFailureReason: string | undefined;\n\n\t\tfor (let attempt = 1; attempt <= this.maxAttempts; attempt++) {\n\t\t\tconst reviewerResponse = await this.reviewerAgent.run(\n\t\t\t\tgetReviewerPrompt(userPrompt, plannerPrompt, executorResult, parseFailureReason),\n\t\t\t\tsignal,\n\t\t\t\tcallback,\n\t\t\t);\n\n\t\t\tif (reviewerResponse) {\n\t\t\t\taddToTotals(totals, reviewerResponse);\n\t\t\t}\n\n\t\t\ttry {\n\t\t\t\treturn parseReviewerDecision(reviewerResponse, this.reviewerDecisionValidator);\n\t\t\t} catch (error) {\n\t\t\t\tif (!(error instanceof RecoverableError)) throw error;\n\n\t\t\t\tparseFailureReason = error.cause;\n\t\t\t\tif (attempt === this.maxAttempts) {\n\t\t\t\t\tthrow new UnrecoverableError('Max attempts exhausted', { cause: parseFailureReason });\n\t\t\t\t}\n\n\t\t\t\tthis.warn(\n\t\t\t\t\t`Reviewer decision ${String(attempt)}/${String(this.maxAttempts)} was unusable; asking the reviewer again`,\n\t\t\t\t\tparseFailureReason,\n\t\t\t\t);\n\t\t\t}\n\t\t}\n\n\t\tthrow new UnrecoverableError('Max attempts exhausted', {\n\t\t\tcause: parseFailureReason ?? 'Unknown failure.',\n\t\t});\n\t}\n\n\tprivate warn(...args: unknown[]): void {\n\t\ttreatErrors(\n\t\t\t() => {\n\t\t\t\tthis.logger.warn(...args);\n\t\t\t},\n\t\t\tclassifyHostFailure,\n\t\t\t'Orchestrator logger failed while reporting a retry',\n\t\t);\n\t}\n}\n"]}
+{"version":3,"file":"orchestratorAgent.js","sourceRoot":"","sources":["../../../../src/orchestration/domain/model/orchestratorAgent.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,SAAS,EAAe,MAAM,kCAAkC,CAAC;AAC1E,OAAO,EACN,gBAAgB,EAChB,kBAAkB,EAClB,eAAe,GACf,MAAM,iCAAiC,CAAC;AAIzC,OAAO,EAAE,mBAAmB,EAAE,WAAW,EAAE,MAAM,0CAA0C,CAAC;AAE5F,MAAM,gBAAgB,GAAG,CAAC,UAAkB,EAAE,qBAA8B,EAAE,EAAE;IAC/E,MAAM,QAAQ,GAAG,qBAAqB;QACrC,CAAC,CAAC,iDAAiD,qBAAqB,OAAO;QAC/E,CAAC,CAAC,EAAE,CAAC;IAEN,OAAO;;;;EAIN,UAAU;KACP,QAAQ,EAAE,CAAC;AAChB,CAAC,CAAC;AAEF,MAAM,iBAAiB,GAAG,CAAC,UAAkB,EAAE,aAAqB,EAAE,EAAE,CACvE;;;;EAIC,UAAU;;;;;EAKV,aAAa;IACX,CAAC;AAEL,MAAM,iBAAiB,GAAG,CACzB,UAAkB,EAClB,aAAqB,EACrB,cAAsB,EACtB,kBAA2B,EAC1B,EAAE;IACH,MAAM,WAAW,GAAG,kBAAkB;QACrC,CAAC,CAAC,iDAAiD,kBAAkB,6GAA6G;QAClL,CAAC,CAAC,EAAE,CAAC;IAEN,OAAO;;;;EAIN,UAAU;;;;;EAKV,aAAa;;;;;EAKb,cAAc;KACX,WAAW,EAAE,CAAC;AACnB,CAAC,CAAC;AAQF,4FAA4F;AAC5F,SAAS,WAAW,CAAC,MAAiB,EAAE,QAAuB;IAC9D,IAAI,CAAC,QAAQ,CAAC,MAAM;QAAE,MAAM,CAAC,UAAU,GAAG,IAAI,CAAC;IAC/C,MAAM,CAAC,MAAM,GAAG,SAAS,CAAC,MAAM,CAAC,MAAM,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC;AAC3D,CAAC;AAED,uFAAuF;AACvF,SAAS,SAAS,CAAC,MAAiB;IACnC,OAAO,MAAM,CAAC,UAAU,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC;AACtD,CAAC;AAED,SAAS,cAAc,CAAC,IAAY;IACnC,MAAM,KAAK,GAAG,mCAAmC,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC;IACpE,OAAO,KAAK,EAAE,CAAC,CAAC,CAAC,IAAI,IAAI,CAAC;AAC3B,CAAC;AAED,SAAS,qBAAqB,CAC7B,QAAuB,EACvB,yBAAsD;IAEtD,IAAI,CAAC,QAAQ,CAAC,QAAQ,EAAE,CAAC;QACxB,MAAM,IAAI,gBAAgB,CAAC,6DAA6D,EAAE;YACzF,KAAK,EAAE,+BAA+B;SACtC,CAAC,CAAC;IACJ,CAAC;IAED,IAAI,cAAuB,CAAC;IAC5B,IAAI,CAAC;QACJ,cAAc,GAAG,IAAI,CAAC,KAAK,CAAC,cAAc,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC,CAAC;IAChE,CAAC;IAAC,MAAM,CAAC;QACR,MAAM,IAAI,gBAAgB,CAAC,uCAAuC,EAAE;YACnE,KAAK,EAAE,uCAAuC;SAC9C,CAAC,CAAC;IACJ,CAAC;IAED,MAAM,QAAQ,GAAG,yBAAyB,CAAC,QAAQ,CAAC,cAAc,CAAC,CAAC;IACpE,IAAI,CAAC,QAAQ,EAAE,CAAC;QACf,MAAM,IAAI,gBAAgB,CAAC,uCAAuC,EAAE;YACnE,KAAK,EAAE,sDAAsD;SAC7D,CAAC,CAAC;IACJ,CAAC;IAED,OAAO,QAAQ,CAAC;AACjB,CAAC;AAED,MAAM,OAAO,iBAAiB;IACrB,YAAY,CAAQ;IACpB,aAAa,CAAQ;IACrB,aAAa,CAAQ;IACrB,yBAAyB,CAA8B;IACvD,WAAW,CAAS;IACpB,MAAM,CAAU;IAExB,YAAY,EACX,YAAY,EACZ,aAAa,EACb,aAAa,EACb,yBAAyB,EACzB,WAAW,GAAG,CAAC,EACf,MAAM,GAQN;QACA,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,WAAW,CAAC,IAAI,WAAW,GAAG,CAAC,EAAE,CAAC;YACvD,MAAM,IAAI,UAAU,CAAC,wCAAwC,CAAC,CAAC;QAChE,CAAC;QAED,IAAI,CAAC,YAAY,GAAG,YAAY,CAAC;QACjC,IAAI,CAAC,aAAa,GAAG,aAAa,CAAC;QACnC,IAAI,CAAC,aAAa,GAAG,aAAa,CAAC;QACnC,IAAI,CAAC,yBAAyB,GAAG,yBAAyB,CAAC;QAC3D,IAAI,CAAC,WAAW,GAAG,WAAW,CAAC;QAC/B,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;IACtB,CAAC;IAED,KAAK,CAAC,GAAG,CAAC,MAAc,EAAE,MAAmB,EAAE,QAAkB;QAChE,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QACzB,wFAAwF;QACxF,MAAM,MAAM,GAAc,EAAE,MAAM,EAAE,SAAS,EAAE,UAAU,EAAE,KAAK,EAAE,CAAC;QAEnE,IAAI,CAAC;YACJ,MAAM,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC;QACxD,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,gFAAgF;YAChF,MAAM,eAAe,CAAC,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,UAAU,CAAC,CAAC;QAChE,CAAC;QAED,OAAO;YACN,QAAQ,EAAE,sBAAsB;YAChC,MAAM,EAAE,SAAS,CAAC,MAAM,CAAC;YACzB,sEAAsE;YACtE,QAAQ,EAAE,CAAC,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC,GAAG,IAAI;SACrC,CAAC;IACH,CAAC;IAEO,KAAK,CAAC,SAAS,CACtB,MAAc,EACd,MAAmB,EACnB,QAAkB,EAClB,MAAiB;QAEjB,IAAI,iBAAqC,CAAC;QAE1C,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,IAAI,CAAC,WAAW,EAAE,OAAO,EAAE,EAAE,CAAC;YAC9D,MAAM,aAAa,GAAG,OAAO,KAAK,IAAI,CAAC,WAAW,CAAC;YAEnD,MAAM,eAAe,GAAG,MAAM,IAAI,CAAC,YAAY,CAAC,GAAG,CAClD,gBAAgB,CAAC,MAAM,EAAE,iBAAiB,CAAC,EAC3C,MAAM,EACN,QAAQ,CACR,CAAC;YACF,WAAW,CAAC,MAAM,EAAE,eAAe,CAAC,CAAC;YAErC,IAAI,CAAC,eAAe,CAAC,QAAQ,EAAE,CAAC;gBAC/B,iBAAiB,GAAG,mCAAmC,CAAC;gBACxD,IAAI,aAAa,EAAE,CAAC;oBACnB,MAAM,IAAI,kBAAkB,CAAC,wBAAwB,EAAE,EAAE,KAAK,EAAE,iBAAiB,EAAE,CAAC,CAAC;gBACtF,CAAC;gBACD,IAAI,CAAC,IAAI,CACR,WAAW,MAAM,CAAC,OAAO,CAAC,IAAI,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,4DAA4D,CAClH,CAAC;gBACF,SAAS;YACV,CAAC;YAED,MAAM,gBAAgB,GAAG,MAAM,IAAI,CAAC,aAAa,CAAC,GAAG,CACpD,iBAAiB,CAAC,MAAM,EAAE,eAAe,CAAC,QAAQ,CAAC,EACnD,MAAM,EACN,QAAQ,CACR,CAAC;YACF,WAAW,CAAC,MAAM,EAAE,gBAAgB,CAAC,CAAC;YAEtC,IAAI,CAAC,gBAAgB,CAAC,QAAQ,EAAE,CAAC;gBAChC,iBAAiB,GAAG,oCAAoC,CAAC;gBACzD,IAAI,aAAa,EAAE,CAAC;oBACnB,MAAM,IAAI,kBAAkB,CAAC,wBAAwB,EAAE,EAAE,KAAK,EAAE,iBAAiB,EAAE,CAAC,CAAC;gBACtF,CAAC;gBAED,IAAI,CAAC,IAAI,CACR,WAAW,MAAM,CAAC,OAAO,CAAC,IAAI,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,6DAA6D,CACnH,CAAC;gBACF,SAAS;YACV,CAAC;YAED,MAAM,gBAAgB,GAAG,MAAM,IAAI,CAAC,mBAAmB,CACtD,MAAM,EACN,eAAe,CAAC,QAAQ,EACxB,gBAAgB,CAAC,QAAQ,EACzB,MAAM,EACN,QAAQ,EACR,MAAM,CACN,CAAC;YAEF,IAAI,gBAAgB,CAAC,QAAQ,KAAK,UAAU;gBAAE,OAAO;YAErD,iBAAiB,GAAG,gBAAgB,CAAC,QAAQ,CAAC;YAC9C,IAAI,aAAa,EAAE,CAAC;gBACnB,MAAM,IAAI,kBAAkB,CAAC,wBAAwB,EAAE,EAAE,KAAK,EAAE,iBAAiB,EAAE,CAAC,CAAC;YACtF,CAAC;YAED,IAAI,CAAC,IAAI,CACR,WAAW,MAAM,CAAC,OAAO,CAAC,IAAI,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,uEAAuE,EAC7H,iBAAiB,CACjB,CAAC;QACH,CAAC;QAED,MAAM,IAAI,kBAAkB,CAAC,wBAAwB,EAAE;YACtD,KAAK,EAAE,iBAAiB,IAAI,kBAAkB;SAC9C,CAAC,CAAC;IACJ,CAAC;IAED,6EAA6E;IAC7E,6EAA6E;IAC7E,6CAA6C;IACrC,KAAK,CAAC,mBAAmB,CAChC,UAAkB,EAClB,aAAqB,EACrB,cAAsB,EACtB,MAAmB,EACnB,QAAkB,EAClB,MAAiB;QAEjB,IAAI,kBAAsC,CAAC;QAE3C,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,IAAI,CAAC,WAAW,EAAE,OAAO,EAAE,EAAE,CAAC;YAC9D,MAAM,gBAAgB,GAAG,MAAM,IAAI,CAAC,aAAa,CAAC,GAAG,CACpD,iBAAiB,CAAC,UAAU,EAAE,aAAa,EAAE,cAAc,EAAE,kBAAkB,CAAC,EAChF,MAAM,EACN,QAAQ,CACR,CAAC;YAEF,WAAW,CAAC,MAAM,EAAE,gBAAgB,CAAC,CAAC;YAEtC,IAAI,CAAC;gBACJ,OAAO,qBAAqB,CAAC,gBAAgB,EAAE,IAAI,CAAC,yBAAyB,CAAC,CAAC;YAChF,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBAChB,IAAI,CAAC,CAAC,KAAK,YAAY,gBAAgB,CAAC;oBAAE,MAAM,KAAK,CAAC;gBAEtD,kBAAkB,GAAG,KAAK,CAAC,KAAK,CAAC;gBACjC,IAAI,OAAO,KAAK,IAAI,CAAC,WAAW,EAAE,CAAC;oBAClC,MAAM,IAAI,kBAAkB,CAAC,wBAAwB,EAAE,EAAE,KAAK,EAAE,kBAAkB,EAAE,CAAC,CAAC;gBACvF,CAAC;gBAED,IAAI,CAAC,IAAI,CACR,qBAAqB,MAAM,CAAC,OAAO,CAAC,IAAI,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,0CAA0C,EAC1G,kBAAkB,CAClB,CAAC;YACH,CAAC;QACF,CAAC;QAED,MAAM,IAAI,kBAAkB,CAAC,wBAAwB,EAAE;YACtD,KAAK,EAAE,kBAAkB,IAAI,kBAAkB;SAC/C,CAAC,CAAC;IACJ,CAAC;IAEO,IAAI,CAAC,GAAG,IAAe;QAC9B,WAAW,CACV,GAAG,EAAE;YACJ,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC,CAAC;QAC3B,CAAC,EACD,mBAAmB,EACnB,oDAAoD,CACpD,CAAC;IACH,CAAC;CACD","sourcesContent":["import type { Agent, AgentResponse, Callback } from '../../../agent/domain/agent.ts';\nimport { addTokens, type Tokens } from '../../../shared/domain/tokens.ts';\nimport {\n\tRecoverableError,\n\tUnrecoverableError,\n\twithSpentTokens,\n} from '../../../agent/domain/errors.ts';\nimport type { ReviewerDecision } from './reviewerDecision.ts';\nimport type { Validator } from '../interface/validator.ts';\nimport type { ILogger } from '../../../shared/domain/logger.ts';\nimport { classifyHostFailure, treatErrors } from '../../../agent/domain/providerFailure.ts';\n\nconst getPlannerPrompt = (userPrompt: string, previousFailureReason?: string) => {\n\tconst feedback = previousFailureReason\n\t\t? `\\n\\nFeedback from the previous attempt:\\n---\\n${previousFailureReason}\\n---`\n\t\t: '';\n\n\treturn `You are the planner agent. Read the relevant repository code and create a concise, engineering-grade plan for the executor. Cover the requested behaviour, structural issues, affected files or symbols, API contracts and boundaries, important failure paths, and meaningful tests. Keep the plan focused on the current request and do not require an architecture redesign yet.\n\nOriginal user request:\n---\n${userPrompt}\n---${feedback}`;\n};\n\nconst getExecutorPrompt = (userPrompt: string, plannerPrompt: string) =>\n\t`You are the executor agent. Read the relevant repository code and implement the original user request according to the planner's current plan. Preserve contracts and boundaries, handle important failure paths, keep the change focused, and add or update meaningful tests. Inspect and change the code; do not merely describe what should be done.\n\nOriginal user request:\n---\n${userPrompt}\n---\n\nCurrent implementation plan:\n---\n${plannerPrompt}\n---`;\n\nconst getReviewerPrompt = (\n\tuserPrompt: string,\n\tplannerPrompt: string,\n\texecutorResult: string,\n\tparseFailureReason?: string,\n) => {\n\tconst retryNotice = parseFailureReason\n\t\t? `\\n\\nYour previous response could not be used: ${parseFailureReason} Respond with JSON only, matching the schema exactly, with no surrounding text and no markdown code fences.`\n\t\t: '';\n\n\treturn `You are the reviewer agent. Independently inspect the repository, the current diff, and relevant surrounding code; do not rely on the executor's narrative. Evaluate the original request first; plan compliance is secondary and provides supporting context. Check correctness and logic errors, important failure paths, unused or artificial abstractions, races or shared state, API contract breaks, boundary violations, regressions, scope, and test quality. If it is correct, respond with JSON only: {\"decision\":\"approved\"}. If it is not correct, respond with JSON only: {\"decision\":\"rejected\",\"feedback\":\"list concrete, prioritized findings and the required direction for each\"}. The feedback must contain actionable engineering findings, not a general summary.\n\nOriginal user request:\n---\n${userPrompt}\n---\n\nPlanner's current plan:\n---\n${plannerPrompt}\n---\n\nExecutor response (context only; verify it independently):\n---\n${executorResult}\n---${retryNotice}`;\n};\n\ninterface RunTotals {\n\ttokens: Tokens | undefined;\n\t/** A role answered without usage, so the run's total cannot be known. */\n\tunreported: boolean;\n}\n\n// Every role's response counts, even an empty one that starts another round: it was billed.\nfunction addToTotals(totals: RunTotals, response: AgentResponse): void {\n\tif (!response.tokens) totals.unreported = true;\n\ttotals.tokens = addTokens(totals.tokens, response.tokens);\n}\n\n/** Unknown rather than partial: a sum missing a billed call must not look complete. */\nfunction runTokens(totals: RunTotals): Tokens | undefined {\n\treturn totals.unreported ? undefined : totals.tokens;\n}\n\nfunction stripCodeFence(text: string): string {\n\tconst match = /^```(?:json)?\\s*([\\s\\S]*?)\\s*```$/.exec(text.trim());\n\treturn match?.[1] ?? text;\n}\n\nfunction parseReviewerDecision(\n\tresponse: AgentResponse,\n\treviewerDecisionValidator: Validator<ReviewerDecision>,\n): ReviewerDecision {\n\tif (!response.response) {\n\t\tthrow new RecoverableError(\"There's no decision from the reviewer, something went wrong\", {\n\t\t\tcause: 'Reviewer decision is missing.',\n\t\t});\n\t}\n\n\tlet parsedResponse: unknown;\n\ttry {\n\t\tparsedResponse = JSON.parse(stripCodeFence(response.response));\n\t} catch {\n\t\tthrow new RecoverableError('Reviewer returned an invalid decision', {\n\t\t\tcause: 'Reviewer response must be valid JSON.',\n\t\t});\n\t}\n\n\tconst decision = reviewerDecisionValidator.validate(parsedResponse);\n\tif (!decision) {\n\t\tthrow new RecoverableError('Reviewer returned an invalid decision', {\n\t\t\tcause: 'Reviewer response did not match the decision schema.',\n\t\t});\n\t}\n\n\treturn decision;\n}\n\nexport class OrchestratorAgent implements Agent {\n\tprivate plannerAgent: Agent;\n\tprivate executorAgent: Agent;\n\tprivate reviewerAgent: Agent;\n\tprivate reviewerDecisionValidator: Validator<ReviewerDecision>;\n\tprivate maxAttempts: number;\n\tprivate logger: ILogger;\n\n\tconstructor({\n\t\tplannerAgent,\n\t\texecutorAgent,\n\t\treviewerAgent,\n\t\treviewerDecisionValidator,\n\t\tmaxAttempts = 3,\n\t\tlogger,\n\t}: {\n\t\tplannerAgent: Agent;\n\t\texecutorAgent: Agent;\n\t\treviewerAgent: Agent;\n\t\treviewerDecisionValidator: Validator<ReviewerDecision>;\n\t\tmaxAttempts?: number;\n\t\tlogger: ILogger;\n\t}) {\n\t\tif (!Number.isInteger(maxAttempts) || maxAttempts < 1) {\n\t\t\tthrow new RangeError('maxAttempts must be a positive integer');\n\t\t}\n\n\t\tthis.plannerAgent = plannerAgent;\n\t\tthis.executorAgent = executorAgent;\n\t\tthis.reviewerAgent = reviewerAgent;\n\t\tthis.reviewerDecisionValidator = reviewerDecisionValidator;\n\t\tthis.maxAttempts = maxAttempts;\n\t\tthis.logger = logger;\n\t}\n\n\tasync run(prompt: string, signal: AbortSignal, callback: Callback): Promise<AgentResponse> {\n\t\tconst start = Date.now();\n\t\t// Per invocation, not per instance: the CLI keeps one orchestrator for a whole session.\n\t\tconst totals: RunTotals = { tokens: undefined, unreported: false };\n\n\t\ttry {\n\t\t\tawait this.runRounds(prompt, signal, callback, totals);\n\t\t} catch (error) {\n\t\t\t// A failing role carries its own tokens; the roles before it are in the totals.\n\t\t\tthrow withSpentTokens(error, totals.tokens, totals.unreported);\n\t\t}\n\n\t\treturn {\n\t\t\tresponse: 'All job has finished',\n\t\t\ttokens: runTokens(totals),\n\t\t\t// The whole run's wall clock, so it matches what the consumer waited.\n\t\t\tduration: (Date.now() - start) / 1000,\n\t\t};\n\t}\n\n\tprivate async runRounds(\n\t\tprompt: string,\n\t\tsignal: AbortSignal,\n\t\tcallback: Callback,\n\t\ttotals: RunTotals,\n\t): Promise<void> {\n\t\tlet lastFailureReason: string | undefined;\n\n\t\tfor (let attempt = 1; attempt <= this.maxAttempts; attempt++) {\n\t\t\tconst isLastAttempt = attempt === this.maxAttempts;\n\n\t\t\tconst plannerResponse = await this.plannerAgent.run(\n\t\t\t\tgetPlannerPrompt(prompt, lastFailureReason),\n\t\t\t\tsignal,\n\t\t\t\tcallback,\n\t\t\t);\n\t\t\taddToTotals(totals, plannerResponse);\n\n\t\t\tif (!plannerResponse.response) {\n\t\t\t\tlastFailureReason = 'The planner produced no response.';\n\t\t\t\tif (isLastAttempt) {\n\t\t\t\t\tthrow new UnrecoverableError('Max attempts exhausted', { cause: lastFailureReason });\n\t\t\t\t}\n\t\t\t\tthis.warn(\n\t\t\t\t\t`Attempt ${String(attempt)}/${String(this.maxAttempts)}: the planner produced no response; starting another round`,\n\t\t\t\t);\n\t\t\t\tcontinue;\n\t\t\t}\n\n\t\t\tconst executorResponse = await this.executorAgent.run(\n\t\t\t\tgetExecutorPrompt(prompt, plannerResponse.response),\n\t\t\t\tsignal,\n\t\t\t\tcallback,\n\t\t\t);\n\t\t\taddToTotals(totals, executorResponse);\n\n\t\t\tif (!executorResponse.response) {\n\t\t\t\tlastFailureReason = 'The executor produced no response.';\n\t\t\t\tif (isLastAttempt) {\n\t\t\t\t\tthrow new UnrecoverableError('Max attempts exhausted', { cause: lastFailureReason });\n\t\t\t\t}\n\n\t\t\t\tthis.warn(\n\t\t\t\t\t`Attempt ${String(attempt)}/${String(this.maxAttempts)}: the executor produced no response; starting another round`,\n\t\t\t\t);\n\t\t\t\tcontinue;\n\t\t\t}\n\n\t\t\tconst reviewerDecision = await this.getReviewerDecision(\n\t\t\t\tprompt,\n\t\t\t\tplannerResponse.response,\n\t\t\t\texecutorResponse.response,\n\t\t\t\tsignal,\n\t\t\t\tcallback,\n\t\t\t\ttotals,\n\t\t\t);\n\n\t\t\tif (reviewerDecision.decision === 'approved') return;\n\n\t\t\tlastFailureReason = reviewerDecision.feedback;\n\t\t\tif (isLastAttempt) {\n\t\t\t\tthrow new UnrecoverableError('Max attempts exhausted', { cause: lastFailureReason });\n\t\t\t}\n\n\t\t\tthis.warn(\n\t\t\t\t`Attempt ${String(attempt)}/${String(this.maxAttempts)}: the reviewer rejected the round; starting another with its feedback`,\n\t\t\t\tlastFailureReason,\n\t\t\t);\n\t\t}\n\n\t\tthrow new UnrecoverableError('Max attempts exhausted', {\n\t\t\tcause: lastFailureReason ?? 'Unknown failure.',\n\t\t});\n\t}\n\n\t// Retries only the reviewer call on a malformed decision, instead of redoing\n\t// planning/execution — a bad or unparsable reviewer response is not evidence\n\t// the plan or the implementation were wrong.\n\tprivate async getReviewerDecision(\n\t\tuserPrompt: string,\n\t\tplannerPrompt: string,\n\t\texecutorResult: string,\n\t\tsignal: AbortSignal,\n\t\tcallback: Callback,\n\t\ttotals: RunTotals,\n\t): Promise<ReviewerDecision> {\n\t\tlet parseFailureReason: string | undefined;\n\n\t\tfor (let attempt = 1; attempt <= this.maxAttempts; attempt++) {\n\t\t\tconst reviewerResponse = await this.reviewerAgent.run(\n\t\t\t\tgetReviewerPrompt(userPrompt, plannerPrompt, executorResult, parseFailureReason),\n\t\t\t\tsignal,\n\t\t\t\tcallback,\n\t\t\t);\n\n\t\t\taddToTotals(totals, reviewerResponse);\n\n\t\t\ttry {\n\t\t\t\treturn parseReviewerDecision(reviewerResponse, this.reviewerDecisionValidator);\n\t\t\t} catch (error) {\n\t\t\t\tif (!(error instanceof RecoverableError)) throw error;\n\n\t\t\t\tparseFailureReason = error.cause;\n\t\t\t\tif (attempt === this.maxAttempts) {\n\t\t\t\t\tthrow new UnrecoverableError('Max attempts exhausted', { cause: parseFailureReason });\n\t\t\t\t}\n\n\t\t\t\tthis.warn(\n\t\t\t\t\t`Reviewer decision ${String(attempt)}/${String(this.maxAttempts)} was unusable; asking the reviewer again`,\n\t\t\t\t\tparseFailureReason,\n\t\t\t\t);\n\t\t\t}\n\t\t}\n\n\t\tthrow new UnrecoverableError('Max attempts exhausted', {\n\t\t\tcause: parseFailureReason ?? 'Unknown failure.',\n\t\t});\n\t}\n\n\tprivate warn(...args: unknown[]): void {\n\t\ttreatErrors(\n\t\t\t() => {\n\t\t\t\tthis.logger.warn(...args);\n\t\t\t},\n\t\t\tclassifyHostFailure,\n\t\t\t'Orchestrator logger failed while reporting a retry',\n\t\t);\n\t}\n}\n"]}

package/dist/retry/domain/model/retryingAgent.d.ts

@@ -4,10 +4,18 @@ export declare class RetryingAgent implements Agent {
     private inner;
     private maxAttempts;
     private logger;
-    constructor({ inner, maxAttempts, logger, }: {
+    private noteFailures;
+    constructor({ inner, maxAttempts, logger, noteFailures, }: {
         inner: Agent;
         maxAttempts?: number;
         logger: ILogger;
+        /**
+         * Appends the previous failure to the retried prompt, so a provider session that may
+         * remember the failed turn is told what happened. Off for an agent that records nothing
+         * from a failed call, such as `LLMAgent`: there the note would become a permanent user
+         * message resent on every later turn.
+         */
+        noteFailures?: boolean;
     });
-    run(prompt: string, signal: AbortSignal, callback: Callback): Promise<AgentResponse | undefined>;
+    run(prompt: string, signal: AbortSignal, callback: Callback): Promise<AgentResponse>;
 }

package/dist/retry/domain/model/retryingAgent.js

@@ -1,48 +1,78 @@
-import { RecoverableError, UnrecoverableError } from "../../../agent/domain/errors.js";
+import { RecoverableError, UnrecoverableError, withSpentTokens, } from "../../../agent/domain/errors.js";
 import { classifyHostFailure, describeFailure, treatErrors, } from "../../../agent/domain/providerFailure.js";
 import { isAbortError } from "../../../shared/domain/isAbortError.js";
+import { addTokens } from "../../../shared/domain/tokens.js";
 export class RetryingAgent {
     inner;
     maxAttempts;
     logger;
-    constructor({ inner, maxAttempts = 3, logger, }) {
+    noteFailures;
+    constructor({ inner, maxAttempts = 3, logger, noteFailures = true, }) {
         if (!Number.isInteger(maxAttempts) || maxAttempts < 1) {
             throw new RangeError('maxAttempts must be a positive integer');
         }
         this.inner = inner;
         this.maxAttempts = maxAttempts;
         this.logger = logger;
+        this.noteFailures = noteFailures;
     }
     async run(prompt, signal, callback) {
+        const start = Date.now();
         let lastPrompt = null;
-        for (let attempt = 1; attempt <= this.maxAttempts; attempt++) {
+        // Failed attempts were billed too, so they travel with whichever way the run ends.
+        let spent;
+        // An attempt was answered without usage, so `spent` is only a subtotal.
+        let unreported = false;
+        for (let attempt = 1;; attempt++) {
             try {
                 const promptToSend = lastPrompt ?? prompt;
-                return await this.inner.run(promptToSend, signal, callback);
+                const response = await this.inner.run(promptToSend, signal, callback);
+                return {
+                    ...response,
+                    // Any call answered without usage makes the whole count unknown, not partial.
+                    tokens: unreported ? undefined : response.tokens && addTokens(spent, response.tokens),
+                    // Every attempt was waited for, not only the last one.
+                    duration: (Date.now() - start) / 1000,
+                };
             }
             catch (e) {
-                if (isAbortError(e) || e instanceof UnrecoverableError)
+                if (isAbortError(e))
                     throw e;
+                if (e instanceof UnrecoverableError)
+                    throw withSpentTokens(e, spent, unreported);
                 // Only a failure the agent classified as recoverable earns another call. An
                 // unclassified one breaks the `Agent` contract, so nothing here knows whether
                 // replaying it is safe — it may already have written files or run commands.
                 if (!(e instanceof RecoverableError))
                     throw new UnrecoverableError('The agent failed without classifying the failure', {
                         cause: describeFailure(e),
+                        tokens: spent,
+                        usageUnreported: unreported,
                     });
+                spent = addTokens(spent, e.tokens);
+                unreported ||= e.usageUnreported;
                 if (attempt === this.maxAttempts)
                     throw new UnrecoverableError('Max attempts exhausted', {
                         cause: `Gave up after ${String(this.maxAttempts)} attempts. Last failure: ${e.cause}`,
+                        tokens: spent,
+                        usageUnreported: unreported,
                     });
                 // Keeps the original request: an attempt that failed before the provider registered
                 // the turn left no session that remembers it.
-                lastPrompt = `${prompt}\n\nThe previous attempt failed for the following reason: ${e.cause}`;
-                treatErrors(() => {
-                    this.logger.warn(`Attempt ${String(attempt)}/${String(this.maxAttempts)} failed; retrying`, e);
-                }, classifyHostFailure, 'RetryingAgent logger failed while reporting a retry');
+                if (this.noteFailures) {
+                    lastPrompt = `${prompt}\n\nThe previous attempt failed for the following reason: ${e.cause}`;
+                }
+                try {
+                    treatErrors(() => {
+                        this.logger.warn(`Attempt ${String(attempt)}/${String(this.maxAttempts)} failed; retrying`, e);
+                    }, classifyHostFailure, 'RetryingAgent logger failed while reporting a retry');
+                }
+                catch (logFailure) {
+                    // The attempts already spent their tokens, whatever the logger did.
+                    throw withSpentTokens(logFailure, spent, unreported);
+                }
             }
         }
-        return undefined;
     }
 }
 //# sourceMappingURL=retryingAgent.js.map

package/dist/retry/domain/model/retryingAgent.js.map

@@ -1 +1 @@
-{"version":3,"file":"retryingAgent.js","sourceRoot":"","sources":["../../../../src/retry/domain/model/retryingAgent.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,gBAAgB,EAAE,kBAAkB,EAAE,MAAM,iCAAiC,CAAC;AACvF,OAAO,EACN,mBAAmB,EACnB,eAAe,EACf,WAAW,GACX,MAAM,0CAA0C,CAAC;AAClD,OAAO,EAAE,YAAY,EAAE,MAAM,wCAAwC,CAAC;AAGtE,MAAM,OAAO,aAAa;IACjB,KAAK,CAAQ;IACb,WAAW,CAAS;IACpB,MAAM,CAAU;IAExB,YAAY,EACX,KAAK,EACL,WAAW,GAAG,CAAC,EACf,MAAM,GAKN;QACA,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,WAAW,CAAC,IAAI,WAAW,GAAG,CAAC,EAAE,CAAC;YACvD,MAAM,IAAI,UAAU,CAAC,wCAAwC,CAAC,CAAC;QAChE,CAAC;QAED,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;QACnB,IAAI,CAAC,WAAW,GAAG,WAAW,CAAC;QAC/B,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;IACtB,CAAC;IAED,KAAK,CAAC,GAAG,CACR,MAAc,EACd,MAAmB,EACnB,QAAkB;QAElB,IAAI,UAAU,GAAkB,IAAI,CAAC;QACrC,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,IAAI,CAAC,WAAW,EAAE,OAAO,EAAE,EAAE,CAAC;YAC9D,IAAI,CAAC;gBACJ,MAAM,YAAY,GAAG,UAAU,IAAI,MAAM,CAAC;gBAC1C,OAAO,MAAM,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,YAAY,EAAE,MAAM,EAAE,QAAQ,CAAC,CAAC;YAC7D,CAAC;YAAC,OAAO,CAAC,EAAE,CAAC;gBACZ,IAAI,YAAY,CAAC,CAAC,CAAC,IAAI,CAAC,YAAY,kBAAkB;oBAAE,MAAM,CAAC,CAAC;gBAEhE,4EAA4E;gBAC5E,8EAA8E;gBAC9E,4EAA4E;gBAC5E,IAAI,CAAC,CAAC,CAAC,YAAY,gBAAgB,CAAC;oBACnC,MAAM,IAAI,kBAAkB,CAAC,kDAAkD,EAAE;wBAChF,KAAK,EAAE,eAAe,CAAC,CAAC,CAAC;qBACzB,CAAC,CAAC;gBAEJ,IAAI,OAAO,KAAK,IAAI,CAAC,WAAW;oBAC/B,MAAM,IAAI,kBAAkB,CAAC,wBAAwB,EAAE;wBACtD,KAAK,EAAE,iBAAiB,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,4BAA4B,CAAC,CAAC,KAAK,EAAE;qBACrF,CAAC,CAAC;gBAEJ,oFAAoF;gBACpF,8CAA8C;gBAC9C,UAAU,GAAG,GAAG,MAAM,6DAA6D,CAAC,CAAC,KAAK,EAAE,CAAC;gBAC7F,WAAW,CACV,GAAG,EAAE;oBACJ,IAAI,CAAC,MAAM,CAAC,IAAI,CACf,WAAW,MAAM,CAAC,OAAO,CAAC,IAAI,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,mBAAmB,EACzE,CAAC,CACD,CAAC;gBACH,CAAC,EACD,mBAAmB,EACnB,qDAAqD,CACrD,CAAC;YACH,CAAC;QACF,CAAC;QACD,OAAO,SAAS,CAAC;IAClB,CAAC;CACD","sourcesContent":["import type { Agent, AgentResponse, Callback } from '../../../agent/domain/agent.ts';\nimport { RecoverableError, UnrecoverableError } from '../../../agent/domain/errors.ts';\nimport {\n\tclassifyHostFailure,\n\tdescribeFailure,\n\ttreatErrors,\n} from '../../../agent/domain/providerFailure.ts';\nimport { isAbortError } from '../../../shared/domain/isAbortError.ts';\nimport type { ILogger } from '../../../shared/domain/logger.ts';\n\nexport class RetryingAgent implements Agent {\n\tprivate inner: Agent;\n\tprivate maxAttempts: number;\n\tprivate logger: ILogger;\n\n\tconstructor({\n\t\tinner,\n\t\tmaxAttempts = 3,\n\t\tlogger,\n\t}: {\n\t\tinner: Agent;\n\t\tmaxAttempts?: number;\n\t\tlogger: ILogger;\n\t}) {\n\t\tif (!Number.isInteger(maxAttempts) || maxAttempts < 1) {\n\t\t\tthrow new RangeError('maxAttempts must be a positive integer');\n\t\t}\n\n\t\tthis.inner = inner;\n\t\tthis.maxAttempts = maxAttempts;\n\t\tthis.logger = logger;\n\t}\n\n\tasync run(\n\t\tprompt: string,\n\t\tsignal: AbortSignal,\n\t\tcallback: Callback,\n\t): Promise<AgentResponse | undefined> {\n\t\tlet lastPrompt: string | null = null;\n\t\tfor (let attempt = 1; attempt <= this.maxAttempts; attempt++) {\n\t\t\ttry {\n\t\t\t\tconst promptToSend = lastPrompt ?? prompt;\n\t\t\t\treturn await this.inner.run(promptToSend, signal, callback);\n\t\t\t} catch (e) {\n\t\t\t\tif (isAbortError(e) || e instanceof UnrecoverableError) throw e;\n\n\t\t\t\t// Only a failure the agent classified as recoverable earns another call. An\n\t\t\t\t// unclassified one breaks the `Agent` contract, so nothing here knows whether\n\t\t\t\t// replaying it is safe — it may already have written files or run commands.\n\t\t\t\tif (!(e instanceof RecoverableError))\n\t\t\t\t\tthrow new UnrecoverableError('The agent failed without classifying the failure', {\n\t\t\t\t\t\tcause: describeFailure(e),\n\t\t\t\t\t});\n\n\t\t\t\tif (attempt === this.maxAttempts)\n\t\t\t\t\tthrow new UnrecoverableError('Max attempts exhausted', {\n\t\t\t\t\t\tcause: `Gave up after ${String(this.maxAttempts)} attempts. Last failure: ${e.cause}`,\n\t\t\t\t\t});\n\n\t\t\t\t// Keeps the original request: an attempt that failed before the provider registered\n\t\t\t\t// the turn left no session that remembers it.\n\t\t\t\tlastPrompt = `${prompt}\\n\\nThe previous attempt failed for the following reason: ${e.cause}`;\n\t\t\t\ttreatErrors(\n\t\t\t\t\t() => {\n\t\t\t\t\t\tthis.logger.warn(\n\t\t\t\t\t\t\t`Attempt ${String(attempt)}/${String(this.maxAttempts)} failed; retrying`,\n\t\t\t\t\t\t\te,\n\t\t\t\t\t\t);\n\t\t\t\t\t},\n\t\t\t\t\tclassifyHostFailure,\n\t\t\t\t\t'RetryingAgent logger failed while reporting a retry',\n\t\t\t\t);\n\t\t\t}\n\t\t}\n\t\treturn undefined;\n\t}\n}\n"]}
+{"version":3,"file":"retryingAgent.js","sourceRoot":"","sources":["../../../../src/retry/domain/model/retryingAgent.ts"],"names":[],"mappings":"AACA,OAAO,EACN,gBAAgB,EAChB,kBAAkB,EAClB,eAAe,GACf,MAAM,iCAAiC,CAAC;AACzC,OAAO,EACN,mBAAmB,EACnB,eAAe,EACf,WAAW,GACX,MAAM,0CAA0C,CAAC;AAClD,OAAO,EAAE,YAAY,EAAE,MAAM,wCAAwC,CAAC;AAEtE,OAAO,EAAE,SAAS,EAAe,MAAM,kCAAkC,CAAC;AAE1E,MAAM,OAAO,aAAa;IACjB,KAAK,CAAQ;IACb,WAAW,CAAS;IACpB,MAAM,CAAU;IAChB,YAAY,CAAU;IAE9B,YAAY,EACX,KAAK,EACL,WAAW,GAAG,CAAC,EACf,MAAM,EACN,YAAY,GAAG,IAAI,GAYnB;QACA,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,WAAW,CAAC,IAAI,WAAW,GAAG,CAAC,EAAE,CAAC;YACvD,MAAM,IAAI,UAAU,CAAC,wCAAwC,CAAC,CAAC;QAChE,CAAC;QAED,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;QACnB,IAAI,CAAC,WAAW,GAAG,WAAW,CAAC;QAC/B,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;QACrB,IAAI,CAAC,YAAY,GAAG,YAAY,CAAC;IAClC,CAAC;IAED,KAAK,CAAC,GAAG,CAAC,MAAc,EAAE,MAAmB,EAAE,QAAkB;QAChE,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QACzB,IAAI,UAAU,GAAkB,IAAI,CAAC;QACrC,mFAAmF;QACnF,IAAI,KAAyB,CAAC;QAC9B,wEAAwE;QACxE,IAAI,UAAU,GAAG,KAAK,CAAC;QAEvB,KAAK,IAAI,OAAO,GAAG,CAAC,GAAI,OAAO,EAAE,EAAE,CAAC;YACnC,IAAI,CAAC;gBACJ,MAAM,YAAY,GAAG,UAAU,IAAI,MAAM,CAAC;gBAC1C,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,YAAY,EAAE,MAAM,EAAE,QAAQ,CAAC,CAAC;gBACtE,OAAO;oBACN,GAAG,QAAQ;oBACX,8EAA8E;oBAC9E,MAAM,EAAE,UAAU,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,QAAQ,CAAC,MAAM,IAAI,SAAS,CAAC,KAAK,EAAE,QAAQ,CAAC,MAAM,CAAC;oBACrF,uDAAuD;oBACvD,QAAQ,EAAE,CAAC,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC,GAAG,IAAI;iBACrC,CAAC;YACH,CAAC;YAAC,OAAO,CAAC,EAAE,CAAC;gBACZ,IAAI,YAAY,CAAC,CAAC,CAAC;oBAAE,MAAM,CAAC,CAAC;gBAC7B,IAAI,CAAC,YAAY,kBAAkB;oBAAE,MAAM,eAAe,CAAC,CAAC,EAAE,KAAK,EAAE,UAAU,CAAC,CAAC;gBAEjF,4EAA4E;gBAC5E,8EAA8E;gBAC9E,4EAA4E;gBAC5E,IAAI,CAAC,CAAC,CAAC,YAAY,gBAAgB,CAAC;oBACnC,MAAM,IAAI,kBAAkB,CAAC,kDAAkD,EAAE;wBAChF,KAAK,EAAE,eAAe,CAAC,CAAC,CAAC;wBACzB,MAAM,EAAE,KAAK;wBACb,eAAe,EAAE,UAAU;qBAC3B,CAAC,CAAC;gBAEJ,KAAK,GAAG,SAAS,CAAC,KAAK,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC;gBACnC,UAAU,KAAK,CAAC,CAAC,eAAe,CAAC;gBAEjC,IAAI,OAAO,KAAK,IAAI,CAAC,WAAW;oBAC/B,MAAM,IAAI,kBAAkB,CAAC,wBAAwB,EAAE;wBACtD,KAAK,EAAE,iBAAiB,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,4BAA4B,CAAC,CAAC,KAAK,EAAE;wBACrF,MAAM,EAAE,KAAK;wBACb,eAAe,EAAE,UAAU;qBAC3B,CAAC,CAAC;gBAEJ,oFAAoF;gBACpF,8CAA8C;gBAC9C,IAAI,IAAI,CAAC,YAAY,EAAE,CAAC;oBACvB,UAAU,GAAG,GAAG,MAAM,6DAA6D,CAAC,CAAC,KAAK,EAAE,CAAC;gBAC9F,CAAC;gBACD,IAAI,CAAC;oBACJ,WAAW,CACV,GAAG,EAAE;wBACJ,IAAI,CAAC,MAAM,CAAC,IAAI,CACf,WAAW,MAAM,CAAC,OAAO,CAAC,IAAI,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,mBAAmB,EACzE,CAAC,CACD,CAAC;oBACH,CAAC,EACD,mBAAmB,EACnB,qDAAqD,CACrD,CAAC;gBACH,CAAC;gBAAC,OAAO,UAAU,EAAE,CAAC;oBACrB,oEAAoE;oBACpE,MAAM,eAAe,CAAC,UAAU,EAAE,KAAK,EAAE,UAAU,CAAC,CAAC;gBACtD,CAAC;YACF,CAAC;QACF,CAAC;IACF,CAAC;CACD","sourcesContent":["import type { Agent, AgentResponse, Callback } from '../../../agent/domain/agent.ts';\nimport {\n\tRecoverableError,\n\tUnrecoverableError,\n\twithSpentTokens,\n} from '../../../agent/domain/errors.ts';\nimport {\n\tclassifyHostFailure,\n\tdescribeFailure,\n\ttreatErrors,\n} from '../../../agent/domain/providerFailure.ts';\nimport { isAbortError } from '../../../shared/domain/isAbortError.ts';\nimport type { ILogger } from '../../../shared/domain/logger.ts';\nimport { addTokens, type Tokens } from '../../../shared/domain/tokens.ts';\n\nexport class RetryingAgent implements Agent {\n\tprivate inner: Agent;\n\tprivate maxAttempts: number;\n\tprivate logger: ILogger;\n\tprivate noteFailures: boolean;\n\n\tconstructor({\n\t\tinner,\n\t\tmaxAttempts = 3,\n\t\tlogger,\n\t\tnoteFailures = true,\n\t}: {\n\t\tinner: Agent;\n\t\tmaxAttempts?: number;\n\t\tlogger: ILogger;\n\t\t/**\n\t\t * Appends the previous failure to the retried prompt, so a provider session that may\n\t\t * remember the failed turn is told what happened. Off for an agent that records nothing\n\t\t * from a failed call, such as `LLMAgent`: there the note would become a permanent user\n\t\t * message resent on every later turn.\n\t\t */\n\t\tnoteFailures?: boolean;\n\t}) {\n\t\tif (!Number.isInteger(maxAttempts) || maxAttempts < 1) {\n\t\t\tthrow new RangeError('maxAttempts must be a positive integer');\n\t\t}\n\n\t\tthis.inner = inner;\n\t\tthis.maxAttempts = maxAttempts;\n\t\tthis.logger = logger;\n\t\tthis.noteFailures = noteFailures;\n\t}\n\n\tasync run(prompt: string, signal: AbortSignal, callback: Callback): Promise<AgentResponse> {\n\t\tconst start = Date.now();\n\t\tlet lastPrompt: string | null = null;\n\t\t// Failed attempts were billed too, so they travel with whichever way the run ends.\n\t\tlet spent: Tokens | undefined;\n\t\t// An attempt was answered without usage, so `spent` is only a subtotal.\n\t\tlet unreported = false;\n\n\t\tfor (let attempt = 1; ; attempt++) {\n\t\t\ttry {\n\t\t\t\tconst promptToSend = lastPrompt ?? prompt;\n\t\t\t\tconst response = await this.inner.run(promptToSend, signal, callback);\n\t\t\t\treturn {\n\t\t\t\t\t...response,\n\t\t\t\t\t// Any call answered without usage makes the whole count unknown, not partial.\n\t\t\t\t\ttokens: unreported ? undefined : response.tokens && addTokens(spent, response.tokens),\n\t\t\t\t\t// Every attempt was waited for, not only the last one.\n\t\t\t\t\tduration: (Date.now() - start) / 1000,\n\t\t\t\t};\n\t\t\t} catch (e) {\n\t\t\t\tif (isAbortError(e)) throw e;\n\t\t\t\tif (e instanceof UnrecoverableError) throw withSpentTokens(e, spent, unreported);\n\n\t\t\t\t// Only a failure the agent classified as recoverable earns another call. An\n\t\t\t\t// unclassified one breaks the `Agent` contract, so nothing here knows whether\n\t\t\t\t// replaying it is safe — it may already have written files or run commands.\n\t\t\t\tif (!(e instanceof RecoverableError))\n\t\t\t\t\tthrow new UnrecoverableError('The agent failed without classifying the failure', {\n\t\t\t\t\t\tcause: describeFailure(e),\n\t\t\t\t\t\ttokens: spent,\n\t\t\t\t\t\tusageUnreported: unreported,\n\t\t\t\t\t});\n\n\t\t\t\tspent = addTokens(spent, e.tokens);\n\t\t\t\tunreported ||= e.usageUnreported;\n\n\t\t\t\tif (attempt === this.maxAttempts)\n\t\t\t\t\tthrow new UnrecoverableError('Max attempts exhausted', {\n\t\t\t\t\t\tcause: `Gave up after ${String(this.maxAttempts)} attempts. Last failure: ${e.cause}`,\n\t\t\t\t\t\ttokens: spent,\n\t\t\t\t\t\tusageUnreported: unreported,\n\t\t\t\t\t});\n\n\t\t\t\t// Keeps the original request: an attempt that failed before the provider registered\n\t\t\t\t// the turn left no session that remembers it.\n\t\t\t\tif (this.noteFailures) {\n\t\t\t\t\tlastPrompt = `${prompt}\\n\\nThe previous attempt failed for the following reason: ${e.cause}`;\n\t\t\t\t}\n\t\t\t\ttry {\n\t\t\t\t\ttreatErrors(\n\t\t\t\t\t\t() => {\n\t\t\t\t\t\t\tthis.logger.warn(\n\t\t\t\t\t\t\t\t`Attempt ${String(attempt)}/${String(this.maxAttempts)} failed; retrying`,\n\t\t\t\t\t\t\t\te,\n\t\t\t\t\t\t\t);\n\t\t\t\t\t\t},\n\t\t\t\t\t\tclassifyHostFailure,\n\t\t\t\t\t\t'RetryingAgent logger failed while reporting a retry',\n\t\t\t\t\t);\n\t\t\t\t} catch (logFailure) {\n\t\t\t\t\t// The attempts already spent their tokens, whatever the logger did.\n\t\t\t\t\tthrow withSpentTokens(logFailure, spent, unreported);\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n}\n"]}

package/dist/shared/domain/tokens.d.ts

@@ -0,0 +1,26 @@
+/**
+ * Token counts split into the categories providers bill separately. The fields never
+ * overlap: the prompt the model read is `inputTokens + readCacheTokens + writtenCacheTokens`.
+ * Each engine converts its provider's counters to this meaning, because providers disagree
+ * on whether cached tokens are part of their own input count.
+ *
+ * These are counts, not a cost. Pricing is out of scope: a run's tokens are not attributed
+ * to the model that spent them, and cache writes with different lifetimes share one field.
+ */
+export interface Tokens {
+    /** Prompt tokens neither read from nor written to the cache. */
+    inputTokens: number;
+    /** Prompt tokens read from the provider's prompt cache. */
+    readCacheTokens: number;
+    /** Prompt tokens written to the prompt cache by this call. */
+    writtenCacheTokens: number;
+    /** Every generated token, reasoning included. */
+    outputTokens: number;
+}
+/**
+ * Sums two counts where a missing side adds nothing, as for a failure that never got an
+ * answer. A call the provider answered without usage is different: it was billed, so the
+ * layers that accumulate track it apart (`usageUnreported` on an error, missing `tokens` on a
+ * response) and report the whole total as unknown, never as this partial sum.
+ */
+export declare function addTokens(a: Tokens | undefined, b: Tokens | undefined): Tokens | undefined;

package/dist/shared/domain/tokens.js

@@ -0,0 +1,19 @@
+/**
+ * Sums two counts where a missing side adds nothing, as for a failure that never got an
+ * answer. A call the provider answered without usage is different: it was billed, so the
+ * layers that accumulate track it apart (`usageUnreported` on an error, missing `tokens` on a
+ * response) and report the whole total as unknown, never as this partial sum.
+ */
+export function addTokens(a, b) {
+    if (!a)
+        return b;
+    if (!b)
+        return a;
+    return {
+        inputTokens: a.inputTokens + b.inputTokens,
+        readCacheTokens: a.readCacheTokens + b.readCacheTokens,
+        writtenCacheTokens: a.writtenCacheTokens + b.writtenCacheTokens,
+        outputTokens: a.outputTokens + b.outputTokens,
+    };
+}
+//# sourceMappingURL=tokens.js.map

package/dist/shared/domain/tokens.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"tokens.js","sourceRoot":"","sources":["../../../src/shared/domain/tokens.ts"],"names":[],"mappings":"AAoBA;;;;;GAKG;AACH,MAAM,UAAU,SAAS,CAAC,CAAqB,EAAE,CAAqB;IACrE,IAAI,CAAC,CAAC;QAAE,OAAO,CAAC,CAAC;IACjB,IAAI,CAAC,CAAC;QAAE,OAAO,CAAC,CAAC;IAEjB,OAAO;QACN,WAAW,EAAE,CAAC,CAAC,WAAW,GAAG,CAAC,CAAC,WAAW;QAC1C,eAAe,EAAE,CAAC,CAAC,eAAe,GAAG,CAAC,CAAC,eAAe;QACtD,kBAAkB,EAAE,CAAC,CAAC,kBAAkB,GAAG,CAAC,CAAC,kBAAkB;QAC/D,YAAY,EAAE,CAAC,CAAC,YAAY,GAAG,CAAC,CAAC,YAAY;KAC7C,CAAC;AACH,CAAC","sourcesContent":["/**\n * Token counts split into the categories providers bill separately. The fields never\n * overlap: the prompt the model read is `inputTokens + readCacheTokens + writtenCacheTokens`.\n * Each engine converts its provider's counters to this meaning, because providers disagree\n * on whether cached tokens are part of their own input count.\n *\n * These are counts, not a cost. Pricing is out of scope: a run's tokens are not attributed\n * to the model that spent them, and cache writes with different lifetimes share one field.\n */\nexport interface Tokens {\n\t/** Prompt tokens neither read from nor written to the cache. */\n\tinputTokens: number;\n\t/** Prompt tokens read from the provider's prompt cache. */\n\treadCacheTokens: number;\n\t/** Prompt tokens written to the prompt cache by this call. */\n\twrittenCacheTokens: number;\n\t/** Every generated token, reasoning included. */\n\toutputTokens: number;\n}\n\n/**\n * Sums two counts where a missing side adds nothing, as for a failure that never got an\n * answer. A call the provider answered without usage is different: it was billed, so the\n * layers that accumulate track it apart (`usageUnreported` on an error, missing `tokens` on a\n * response) and report the whole total as unknown, never as this partial sum.\n */\nexport function addTokens(a: Tokens | undefined, b: Tokens | undefined): Tokens | undefined {\n\tif (!a) return b;\n\tif (!b) return a;\n\n\treturn {\n\t\tinputTokens: a.inputTokens + b.inputTokens,\n\t\treadCacheTokens: a.readCacheTokens + b.readCacheTokens,\n\t\twrittenCacheTokens: a.writtenCacheTokens + b.writtenCacheTokens,\n\t\toutputTokens: a.outputTokens + b.outputTokens,\n\t};\n}\n"]}

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@mikode13/harness",
-	"version": "1.0.1",
+	"version": "2.0.0",
 	"description": "Test/agent harness utility",
 	"type": "module",
 	"license": "SEE LICENSE IN LICENSE",
@@ -41,6 +41,7 @@
 		"check": "pnpm run format:check && pnpm run lint && pnpm run typecheck"
 	},
 	"devDependencies": {
+		"@anthropic-ai/sdk": "^0.115.0",
 		"@mikode13/code-quality": "1.0.0",
 		"@mikode13/code-style": "1.0.0",
 		"@mikode13/cross-platform": "1.1.0",
@@ -56,6 +57,7 @@
 	"dependencies": {
 		"@anthropic-ai/claude-agent-sdk": "^0.3.220",
 		"@openai/codex-sdk": "^0.154.0",
+		"openai": "^7.23.0",
 		"zod": "^4.4.3"
 	}
 }