@caupulican/pi-adaptative 0.81.21 → 0.81.23

package/dist/core/failure-corpus.d.ts

@@ -40,6 +40,16 @@ export interface FailureCorpusFs {
         size: number;
     };
 }
+export declare function createToolValidationFailureCorpusRecord(args: {
+    provider?: string;
+    modelId?: string;
+    tool: string;
+    failureModes: readonly string[];
+    shape: readonly ToolValidationFailureShapeEntry[];
+    errorKeywords?: readonly string[];
+    ts: string;
+}): ToolValidationFailureCorpusRecord;
+export declare function writeFailureCorpusRecord(filePath: string, record: FailureCorpusRecord, fs?: FailureCorpusFs): void;
 export declare class FailureCorpusRecorder {
     private total;
     private unknown;
@@ -69,7 +79,6 @@ export declare class FailureCorpusRecorder {
     }): void;
     stats(): FailureCorpusStats;
     private appendRecord;
-    private rotateIfNeeded;
 }
 export declare function redactSecrets(message: string): string;
 //# sourceMappingURL=failure-corpus.d.ts.map

package/dist/core/failure-corpus.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"failure-corpus.d.ts","sourceRoot":"","sources":["../../src/core/failure-corpus.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,2BAA2B,CAAC;AAEjE,MAAM,WAAW,2BAA2B;IAC3C,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,eAAe,CAAC,QAAQ,CAAC,CAAC;IAClC,SAAS,EAAE,OAAO,CAAC;IACnB,OAAO,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,+BAA+B;IAC/C,IAAI,EAAE,MAAM,CAAC;IACb,YAAY,EAAE,MAAM,CAAC;IACrB,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,CAAC,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,iCAAiC;IACjD,IAAI,EAAE,iBAAiB,CAAC;IACxB,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE,MAAM,CAAC;IACb,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,KAAK,EAAE,+BAA+B,EAAE,CAAC;IACzC,aAAa,EAAE,MAAM,EAAE,CAAC;CACxB;AAED,MAAM,MAAM,mBAAmB,GAAG,2BAA2B,GAAG,iCAAiC,CAAC;AAElG,MAAM,WAAW,kBAAkB;IAClC,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,eAAe;IAC/B,UAAU,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC;IAClC,SAAS,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE;QAAE,SAAS,EAAE,IAAI,CAAA;KAAE,GAAG,IAAI,CAAC;IAC5D,cAAc,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,QAAQ,EAAE,OAAO,GAAG,IAAI,CAAC;IACpE,YAAY,CAAC,IAAI,EAAE,MAAM,EAAE,QAAQ,EAAE,OAAO,GAAG,MAAM,CAAC;IACtD,aAAa,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,QAAQ,EAAE,OAAO,GAAG,IAAI,CAAC;IACnE,QAAQ,CAAC,IAAI,EAAE,MAAM,GAAG;QAAE,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC;CACzC;AAOD,qBAAa,qBAAqB;IACjC,OAAO,CAAC,KAAK,CAAK;IAClB,OAAO,CAAC,OAAO,CAAK;IACpB,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAS;IAClC,OAAO,CAAC,QAAQ,CAAC,EAAE,CAAkB;IACrC,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAa;IACjC,OAAO,CAAC,QAAQ,CAAC,KAAK,CAA4B;IAElD,YAAY,IAAI,EAAE;QAAE,QAAQ,EAAE,MAAM,CAAC;QAAC,EAAE,CAAC,EAAE,eAAe,CAAC;QAAC,GAAG,CAAC,EAAE,MAAM,IAAI,CAAC;QAAC,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,IAAI,CAAA;KAAE,EAKhH;IAED,MAAM,CAAC,IAAI,EAAE;QAAE,QAAQ,CAAC,EAAE,MAAM,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,eAAe,CAAA;KAAE,GAAG,IAAI,CAWxG;IAED,oBAAoB,CAAC,IAAI,EAAE;QAC1B,QAAQ,CAAC,EAAE,MAAM,CAAC;QAClB,OAAO,CAAC,EAAE,MAAM,CAAC;QACjB,IAAI,EAAE,MAAM,CAAC;QACb,YAAY,EAAE,SAAS,MAAM,EAAE,CAAC;QAChC,KAAK,EAAE,SAAS,+BAA+B,EAAE,CAAC;QAClD,aAAa,CAAC,EAAE,SAAS,MAAM,EAAE,CAAC;KAClC,GAAG,IAAI,CAWP;IAED,KAAK,IAAI,kBAAkB,CAE1B;IAED,OAAO,CAAC,YAAY;IAUpB,OAAO,CAAC,cAAc;CAStB;AAaD,wBAAgB,aAAa,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,CAKrD","sourcesContent":["import { appendFileSync, existsSync, mkdirSync, readFileSync, statSync, writeFileSync } from \"node:fs\";\nimport { dirname } from \"node:path\";\nimport type { ClassifiedError } from \"@caupulican/pi-agent-core\";\n\nexport interface ProviderFailureCorpusRecord {\n\tts: string;\n\tprovider?: string;\n\tmodelId?: string;\n\treason: ClassifiedError[\"reason\"];\n\tretryable: boolean;\n\tmessage: string;\n}\n\nexport interface ToolValidationFailureShapeEntry {\n\tpath: string;\n\texpectedType: string;\n\treceivedType: string;\n\tkeyword?: string;\n}\n\nexport interface ToolValidationFailureCorpusRecord {\n\tkind: \"tool_validation\";\n\tts: string;\n\tprovider?: string;\n\tmodelId?: string;\n\ttool: string;\n\tfailureModes: string[];\n\tshape: ToolValidationFailureShapeEntry[];\n\terrorKeywords: string[];\n}\n\nexport type FailureCorpusRecord = ProviderFailureCorpusRecord | ToolValidationFailureCorpusRecord;\n\nexport interface FailureCorpusStats {\n\ttotal: number;\n\tunknown: number;\n}\n\nexport interface FailureCorpusFs {\n\texistsSync(path: string): boolean;\n\tmkdirSync(path: string, options: { recursive: true }): void;\n\tappendFileSync(path: string, data: string, encoding: \"utf-8\"): void;\n\treadFileSync(path: string, encoding: \"utf-8\"): string;\n\twriteFileSync(path: string, data: string, encoding: \"utf-8\"): void;\n\tstatSync(path: string): { size: number };\n}\n\nconst DEFAULT_FS: FailureCorpusFs = { existsSync, mkdirSync, appendFileSync, readFileSync, writeFileSync, statSync };\nconst MAX_MESSAGE_CHARS = 500;\nconst MAX_FILE_BYTES = 512 * 1024;\nconst MAX_ROTATED_RECORDS = 1000;\n\nexport class FailureCorpusRecorder {\n\tprivate total = 0;\n\tprivate unknown = 0;\n\tprivate readonly filePath: string;\n\tprivate readonly fs: FailureCorpusFs;\n\tprivate readonly now: () => Date;\n\tprivate readonly debug: (message: string) => void;\n\n\tconstructor(args: { filePath: string; fs?: FailureCorpusFs; now?: () => Date; debug?: (message: string) => void }) {\n\t\tthis.filePath = args.filePath;\n\t\tthis.fs = args.fs ?? DEFAULT_FS;\n\t\tthis.now = args.now ?? (() => new Date());\n\t\tthis.debug = args.debug ?? (() => {});\n\t}\n\n\trecord(args: { provider?: string; modelId?: string; message: string; classified: ClassifiedError }): void {\n\t\tthis.total += 1;\n\t\tif (args.classified.reason === \"unknown\") this.unknown += 1;\n\t\tthis.appendRecord({\n\t\t\tts: this.now().toISOString(),\n\t\t\tprovider: args.provider,\n\t\t\tmodelId: args.modelId,\n\t\t\treason: args.classified.reason,\n\t\t\tretryable: args.classified.retryable,\n\t\t\tmessage: redactSecrets(args.message).slice(0, MAX_MESSAGE_CHARS),\n\t\t});\n\t}\n\n\trecordToolValidation(args: {\n\t\tprovider?: string;\n\t\tmodelId?: string;\n\t\ttool: string;\n\t\tfailureModes: readonly string[];\n\t\tshape: readonly ToolValidationFailureShapeEntry[];\n\t\terrorKeywords?: readonly string[];\n\t}): void {\n\t\tthis.appendRecord({\n\t\t\tkind: \"tool_validation\",\n\t\t\tts: this.now().toISOString(),\n\t\t\tprovider: args.provider,\n\t\t\tmodelId: args.modelId,\n\t\t\ttool: args.tool,\n\t\t\tfailureModes: [...new Set(args.failureModes)].sort(),\n\t\t\tshape: sanitizeToolValidationShape(args.shape),\n\t\t\terrorKeywords: [...new Set(args.errorKeywords ?? [])].sort(),\n\t\t});\n\t}\n\n\tstats(): FailureCorpusStats {\n\t\treturn { total: this.total, unknown: this.unknown };\n\t}\n\n\tprivate appendRecord(record: FailureCorpusRecord): void {\n\t\ttry {\n\t\t\tthis.fs.mkdirSync(dirname(this.filePath), { recursive: true });\n\t\t\tthis.fs.appendFileSync(this.filePath, `${JSON.stringify(record)}\\n`, \"utf-8\");\n\t\t\tthis.rotateIfNeeded();\n\t\t} catch (error) {\n\t\t\tthis.debug(`failure corpus write skipped: ${error instanceof Error ? error.message : String(error)}`);\n\t\t}\n\t}\n\n\tprivate rotateIfNeeded(): void {\n\t\tif (!this.fs.existsSync(this.filePath) || this.fs.statSync(this.filePath).size <= MAX_FILE_BYTES) return;\n\t\tconst lines = this.fs\n\t\t\t.readFileSync(this.filePath, \"utf-8\")\n\t\t\t.split(\"\\n\")\n\t\t\t.filter((line) => line.trim().length > 0)\n\t\t\t.slice(-MAX_ROTATED_RECORDS);\n\t\tthis.fs.writeFileSync(this.filePath, `${lines.join(\"\\n\")}\\n`, \"utf-8\");\n\t}\n}\n\nfunction sanitizeToolValidationShape(\n\tshape: readonly ToolValidationFailureShapeEntry[],\n): ToolValidationFailureShapeEntry[] {\n\treturn shape.map((entry) => ({\n\t\tpath: entry.path,\n\t\texpectedType: entry.expectedType,\n\t\treceivedType: entry.receivedType,\n\t\t...(entry.keyword ? { keyword: entry.keyword } : {}),\n\t}));\n}\n\nexport function redactSecrets(message: string): string {\n\treturn message\n\t\t.replace(/sk-(?:proj-|ant-)?[A-Za-z0-9._-]{8,}/g, \"[redacted]\")\n\t\t.replace(/Bearer\\s+[A-Za-z0-9._~+/=-]+/gi, \"Bearer [redacted]\")\n\t\t.replace(/[A-Za-z0-9+/]{40,}={0,2}/g, \"[redacted]\");\n}\n"]}
+{"version":3,"file":"failure-corpus.d.ts","sourceRoot":"","sources":["../../src/core/failure-corpus.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,2BAA2B,CAAC;AAEjE,MAAM,WAAW,2BAA2B;IAC3C,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,eAAe,CAAC,QAAQ,CAAC,CAAC;IAClC,SAAS,EAAE,OAAO,CAAC;IACnB,OAAO,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,+BAA+B;IAC/C,IAAI,EAAE,MAAM,CAAC;IACb,YAAY,EAAE,MAAM,CAAC;IACrB,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,CAAC,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,iCAAiC;IACjD,IAAI,EAAE,iBAAiB,CAAC;IACxB,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE,MAAM,CAAC;IACb,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,KAAK,EAAE,+BAA+B,EAAE,CAAC;IACzC,aAAa,EAAE,MAAM,EAAE,CAAC;CACxB;AAED,MAAM,MAAM,mBAAmB,GAAG,2BAA2B,GAAG,iCAAiC,CAAC;AAElG,MAAM,WAAW,kBAAkB;IAClC,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,eAAe;IAC/B,UAAU,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC;IAClC,SAAS,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE;QAAE,SAAS,EAAE,IAAI,CAAA;KAAE,GAAG,IAAI,CAAC;IAC5D,cAAc,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,QAAQ,EAAE,OAAO,GAAG,IAAI,CAAC;IACpE,YAAY,CAAC,IAAI,EAAE,MAAM,EAAE,QAAQ,EAAE,OAAO,GAAG,MAAM,CAAC;IACtD,aAAa,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,QAAQ,EAAE,OAAO,GAAG,IAAI,CAAC;IACnE,QAAQ,CAAC,IAAI,EAAE,MAAM,GAAG;QAAE,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC;CACzC;AAOD,wBAAgB,uCAAuC,CAAC,IAAI,EAAE;IAC7D,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE,MAAM,CAAC;IACb,YAAY,EAAE,SAAS,MAAM,EAAE,CAAC;IAChC,KAAK,EAAE,SAAS,+BAA+B,EAAE,CAAC;IAClD,aAAa,CAAC,EAAE,SAAS,MAAM,EAAE,CAAC;IAClC,EAAE,EAAE,MAAM,CAAC;CACX,GAAG,iCAAiC,CAWpC;AAED,wBAAgB,wBAAwB,CACvC,QAAQ,EAAE,MAAM,EAChB,MAAM,EAAE,mBAAmB,EAC3B,EAAE,GAAE,eAA4B,GAC9B,IAAI,CAIN;AAYD,qBAAa,qBAAqB;IACjC,OAAO,CAAC,KAAK,CAAK;IAClB,OAAO,CAAC,OAAO,CAAK;IACpB,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAS;IAClC,OAAO,CAAC,QAAQ,CAAC,EAAE,CAAkB;IACrC,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAa;IACjC,OAAO,CAAC,QAAQ,CAAC,KAAK,CAA4B;IAElD,YAAY,IAAI,EAAE;QAAE,QAAQ,EAAE,MAAM,CAAC;QAAC,EAAE,CAAC,EAAE,eAAe,CAAC;QAAC,GAAG,CAAC,EAAE,MAAM,IAAI,CAAC;QAAC,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,IAAI,CAAA;KAAE,EAKhH;IAED,MAAM,CAAC,IAAI,EAAE;QAAE,QAAQ,CAAC,EAAE,MAAM,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,eAAe,CAAA;KAAE,GAAG,IAAI,CAWxG;IAED,oBAAoB,CAAC,IAAI,EAAE;QAC1B,QAAQ,CAAC,EAAE,MAAM,CAAC;QAClB,OAAO,CAAC,EAAE,MAAM,CAAC;QACjB,IAAI,EAAE,MAAM,CAAC;QACb,YAAY,EAAE,SAAS,MAAM,EAAE,CAAC;QAChC,KAAK,EAAE,SAAS,+BAA+B,EAAE,CAAC;QAClD,aAAa,CAAC,EAAE,SAAS,MAAM,EAAE,CAAC;KAClC,GAAG,IAAI,CAEP;IAED,KAAK,IAAI,kBAAkB,CAE1B;IAED,OAAO,CAAC,YAAY;CAOpB;AAaD,wBAAgB,aAAa,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,CAKrD","sourcesContent":["import { appendFileSync, existsSync, mkdirSync, readFileSync, statSync, writeFileSync } from \"node:fs\";\nimport { dirname } from \"node:path\";\nimport type { ClassifiedError } from \"@caupulican/pi-agent-core\";\n\nexport interface ProviderFailureCorpusRecord {\n\tts: string;\n\tprovider?: string;\n\tmodelId?: string;\n\treason: ClassifiedError[\"reason\"];\n\tretryable: boolean;\n\tmessage: string;\n}\n\nexport interface ToolValidationFailureShapeEntry {\n\tpath: string;\n\texpectedType: string;\n\treceivedType: string;\n\tkeyword?: string;\n}\n\nexport interface ToolValidationFailureCorpusRecord {\n\tkind: \"tool_validation\";\n\tts: string;\n\tprovider?: string;\n\tmodelId?: string;\n\ttool: string;\n\tfailureModes: string[];\n\tshape: ToolValidationFailureShapeEntry[];\n\terrorKeywords: string[];\n}\n\nexport type FailureCorpusRecord = ProviderFailureCorpusRecord | ToolValidationFailureCorpusRecord;\n\nexport interface FailureCorpusStats {\n\ttotal: number;\n\tunknown: number;\n}\n\nexport interface FailureCorpusFs {\n\texistsSync(path: string): boolean;\n\tmkdirSync(path: string, options: { recursive: true }): void;\n\tappendFileSync(path: string, data: string, encoding: \"utf-8\"): void;\n\treadFileSync(path: string, encoding: \"utf-8\"): string;\n\twriteFileSync(path: string, data: string, encoding: \"utf-8\"): void;\n\tstatSync(path: string): { size: number };\n}\n\nconst DEFAULT_FS: FailureCorpusFs = { existsSync, mkdirSync, appendFileSync, readFileSync, writeFileSync, statSync };\nconst MAX_MESSAGE_CHARS = 500;\nconst MAX_FILE_BYTES = 512 * 1024;\nconst MAX_ROTATED_RECORDS = 1000;\n\nexport function createToolValidationFailureCorpusRecord(args: {\n\tprovider?: string;\n\tmodelId?: string;\n\ttool: string;\n\tfailureModes: readonly string[];\n\tshape: readonly ToolValidationFailureShapeEntry[];\n\terrorKeywords?: readonly string[];\n\tts: string;\n}): ToolValidationFailureCorpusRecord {\n\treturn {\n\t\tkind: \"tool_validation\",\n\t\tts: args.ts,\n\t\tprovider: args.provider,\n\t\tmodelId: args.modelId,\n\t\ttool: args.tool,\n\t\tfailureModes: [...new Set(args.failureModes)].sort(),\n\t\tshape: sanitizeToolValidationShape(args.shape),\n\t\terrorKeywords: [...new Set(args.errorKeywords ?? [])].sort(),\n\t};\n}\n\nexport function writeFailureCorpusRecord(\n\tfilePath: string,\n\trecord: FailureCorpusRecord,\n\tfs: FailureCorpusFs = DEFAULT_FS,\n): void {\n\tfs.mkdirSync(dirname(filePath), { recursive: true });\n\tfs.appendFileSync(filePath, `${JSON.stringify(record)}\\n`, \"utf-8\");\n\trotateFailureCorpusIfNeeded(filePath, fs);\n}\n\nfunction rotateFailureCorpusIfNeeded(filePath: string, fs: FailureCorpusFs): void {\n\tif (!fs.existsSync(filePath) || fs.statSync(filePath).size <= MAX_FILE_BYTES) return;\n\tconst lines = fs\n\t\t.readFileSync(filePath, \"utf-8\")\n\t\t.split(\"\\n\")\n\t\t.filter((line) => line.trim().length > 0)\n\t\t.slice(-MAX_ROTATED_RECORDS);\n\tfs.writeFileSync(filePath, `${lines.join(\"\\n\")}\\n`, \"utf-8\");\n}\n\nexport class FailureCorpusRecorder {\n\tprivate total = 0;\n\tprivate unknown = 0;\n\tprivate readonly filePath: string;\n\tprivate readonly fs: FailureCorpusFs;\n\tprivate readonly now: () => Date;\n\tprivate readonly debug: (message: string) => void;\n\n\tconstructor(args: { filePath: string; fs?: FailureCorpusFs; now?: () => Date; debug?: (message: string) => void }) {\n\t\tthis.filePath = args.filePath;\n\t\tthis.fs = args.fs ?? DEFAULT_FS;\n\t\tthis.now = args.now ?? (() => new Date());\n\t\tthis.debug = args.debug ?? (() => {});\n\t}\n\n\trecord(args: { provider?: string; modelId?: string; message: string; classified: ClassifiedError }): void {\n\t\tthis.total += 1;\n\t\tif (args.classified.reason === \"unknown\") this.unknown += 1;\n\t\tthis.appendRecord({\n\t\t\tts: this.now().toISOString(),\n\t\t\tprovider: args.provider,\n\t\t\tmodelId: args.modelId,\n\t\t\treason: args.classified.reason,\n\t\t\tretryable: args.classified.retryable,\n\t\t\tmessage: redactSecrets(args.message).slice(0, MAX_MESSAGE_CHARS),\n\t\t});\n\t}\n\n\trecordToolValidation(args: {\n\t\tprovider?: string;\n\t\tmodelId?: string;\n\t\ttool: string;\n\t\tfailureModes: readonly string[];\n\t\tshape: readonly ToolValidationFailureShapeEntry[];\n\t\terrorKeywords?: readonly string[];\n\t}): void {\n\t\tthis.appendRecord(createToolValidationFailureCorpusRecord({ ...args, ts: this.now().toISOString() }));\n\t}\n\n\tstats(): FailureCorpusStats {\n\t\treturn { total: this.total, unknown: this.unknown };\n\t}\n\n\tprivate appendRecord(record: FailureCorpusRecord): void {\n\t\ttry {\n\t\t\twriteFailureCorpusRecord(this.filePath, record, this.fs);\n\t\t} catch (error) {\n\t\t\tthis.debug(`failure corpus write skipped: ${error instanceof Error ? error.message : String(error)}`);\n\t\t}\n\t}\n}\n\nfunction sanitizeToolValidationShape(\n\tshape: readonly ToolValidationFailureShapeEntry[],\n): ToolValidationFailureShapeEntry[] {\n\treturn shape.map((entry) => ({\n\t\tpath: entry.path,\n\t\texpectedType: entry.expectedType,\n\t\treceivedType: entry.receivedType,\n\t\t...(entry.keyword ? { keyword: entry.keyword } : {}),\n\t}));\n}\n\nexport function redactSecrets(message: string): string {\n\treturn message\n\t\t.replace(/sk-(?:proj-|ant-)?[A-Za-z0-9._-]{8,}/g, \"[REDACTED]\")\n\t\t.replace(/Bearer\\s+[A-Za-z0-9._~+/=-]+/gi, \"Bearer [REDACTED]\")\n\t\t.replace(/[A-Za-z0-9+/]{40,}={0,2}/g, \"[REDACTED]\");\n}\n"]}

package/dist/core/failure-corpus.js

@@ -4,6 +4,33 @@ const DEFAULT_FS = { existsSync, mkdirSync, appendFileSync, readFileSync, writeF
 const MAX_MESSAGE_CHARS = 500;
 const MAX_FILE_BYTES = 512 * 1024;
 const MAX_ROTATED_RECORDS = 1000;
+export function createToolValidationFailureCorpusRecord(args) {
+    return {
+        kind: "tool_validation",
+        ts: args.ts,
+        provider: args.provider,
+        modelId: args.modelId,
+        tool: args.tool,
+        failureModes: [...new Set(args.failureModes)].sort(),
+        shape: sanitizeToolValidationShape(args.shape),
+        errorKeywords: [...new Set(args.errorKeywords ?? [])].sort(),
+    };
+}
+export function writeFailureCorpusRecord(filePath, record, fs = DEFAULT_FS) {
+    fs.mkdirSync(dirname(filePath), { recursive: true });
+    fs.appendFileSync(filePath, `${JSON.stringify(record)}\n`, "utf-8");
+    rotateFailureCorpusIfNeeded(filePath, fs);
+}
+function rotateFailureCorpusIfNeeded(filePath, fs) {
+    if (!fs.existsSync(filePath) || fs.statSync(filePath).size <= MAX_FILE_BYTES)
+        return;
+    const lines = fs
+        .readFileSync(filePath, "utf-8")
+        .split("\n")
+        .filter((line) => line.trim().length > 0)
+        .slice(-MAX_ROTATED_RECORDS);
+    fs.writeFileSync(filePath, `${lines.join("\n")}\n`, "utf-8");
+}
 export class FailureCorpusRecorder {
     total = 0;
     unknown = 0;
@@ -31,40 +58,19 @@ export class FailureCorpusRecorder {
         });
     }
     recordToolValidation(args) {
-        this.appendRecord({
-            kind: "tool_validation",
-            ts: this.now().toISOString(),
-            provider: args.provider,
-            modelId: args.modelId,
-            tool: args.tool,
-            failureModes: [...new Set(args.failureModes)].sort(),
-            shape: sanitizeToolValidationShape(args.shape),
-            errorKeywords: [...new Set(args.errorKeywords ?? [])].sort(),
-        });
+        this.appendRecord(createToolValidationFailureCorpusRecord({ ...args, ts: this.now().toISOString() }));
     }
     stats() {
         return { total: this.total, unknown: this.unknown };
     }
     appendRecord(record) {
         try {
-            this.fs.mkdirSync(dirname(this.filePath), { recursive: true });
-            this.fs.appendFileSync(this.filePath, `${JSON.stringify(record)}\n`, "utf-8");
-            this.rotateIfNeeded();
+            writeFailureCorpusRecord(this.filePath, record, this.fs);
         }
         catch (error) {
             this.debug(`failure corpus write skipped: ${error instanceof Error ? error.message : String(error)}`);
         }
     }
-    rotateIfNeeded() {
-        if (!this.fs.existsSync(this.filePath) || this.fs.statSync(this.filePath).size <= MAX_FILE_BYTES)
-            return;
-        const lines = this.fs
-            .readFileSync(this.filePath, "utf-8")
-            .split("\n")
-            .filter((line) => line.trim().length > 0)
-            .slice(-MAX_ROTATED_RECORDS);
-        this.fs.writeFileSync(this.filePath, `${lines.join("\n")}\n`, "utf-8");
-    }
 }
 function sanitizeToolValidationShape(shape) {
     return shape.map((entry) => ({
@@ -76,8 +82,8 @@ function sanitizeToolValidationShape(shape) {
 }
 export function redactSecrets(message) {
     return message
-        .replace(/sk-(?:proj-|ant-)?[A-Za-z0-9._-]{8,}/g, "[redacted]")
-        .replace(/Bearer\s+[A-Za-z0-9._~+/=-]+/gi, "Bearer [redacted]")
-        .replace(/[A-Za-z0-9+/]{40,}={0,2}/g, "[redacted]");
+        .replace(/sk-(?:proj-|ant-)?[A-Za-z0-9._-]{8,}/g, "[REDACTED]")
+        .replace(/Bearer\s+[A-Za-z0-9._~+/=-]+/gi, "Bearer [REDACTED]")
+        .replace(/[A-Za-z0-9+/]{40,}={0,2}/g, "[REDACTED]");
 }
 //# sourceMappingURL=failure-corpus.js.map

package/dist/core/failure-corpus.js.map

@@ -1 +1 @@
-{"version":3,"file":"failure-corpus.js","sourceRoot":"","sources":["../../src/core/failure-corpus.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,UAAU,EAAE,SAAS,EAAE,YAAY,EAAE,QAAQ,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AACvG,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AA8CpC,MAAM,UAAU,GAAoB,EAAE,UAAU,EAAE,SAAS,EAAE,cAAc,EAAE,YAAY,EAAE,aAAa,EAAE,QAAQ,EAAE,CAAC;AACrH,MAAM,iBAAiB,GAAG,GAAG,CAAC;AAC9B,MAAM,cAAc,GAAG,GAAG,GAAG,IAAI,CAAC;AAClC,MAAM,mBAAmB,GAAG,IAAI,CAAC;AAEjC,MAAM,OAAO,qBAAqB;IACzB,KAAK,GAAG,CAAC,CAAC;IACV,OAAO,GAAG,CAAC,CAAC;IACH,QAAQ,CAAS;IACjB,EAAE,CAAkB;IACpB,GAAG,CAAa;IAChB,KAAK,CAA4B;IAElD,YAAY,IAAqG,EAAE;QAClH,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC,QAAQ,CAAC;QAC9B,IAAI,CAAC,EAAE,GAAG,IAAI,CAAC,EAAE,IAAI,UAAU,CAAC;QAChC,IAAI,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC;QAC1C,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,CAAC,GAAG,EAAE,CAAC,EAAC,CAAC,CAAC,CAAC;IAAA,CACtC;IAED,MAAM,CAAC,IAA2F,EAAQ;QACzG,IAAI,CAAC,KAAK,IAAI,CAAC,CAAC;QAChB,IAAI,IAAI,CAAC,UAAU,CAAC,MAAM,KAAK,SAAS;YAAE,IAAI,CAAC,OAAO,IAAI,CAAC,CAAC;QAC5D,IAAI,CAAC,YAAY,CAAC;YACjB,EAAE,EAAE,IAAI,CAAC,GAAG,EAAE,CAAC,WAAW,EAAE;YAC5B,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,OAAO,EAAE,IAAI,CAAC,OAAO;YACrB,MAAM,EAAE,IAAI,CAAC,UAAU,CAAC,MAAM;YAC9B,SAAS,EAAE,IAAI,CAAC,UAAU,CAAC,SAAS;YACpC,OAAO,EAAE,aAAa,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,iBAAiB,CAAC;SAChE,CAAC,CAAC;IAAA,CACH;IAED,oBAAoB,CAAC,IAOpB,EAAQ;QACR,IAAI,CAAC,YAAY,CAAC;YACjB,IAAI,EAAE,iBAAiB;YACvB,EAAE,EAAE,IAAI,CAAC,GAAG,EAAE,CAAC,WAAW,EAAE;YAC5B,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,OAAO,EAAE,IAAI,CAAC,OAAO;YACrB,IAAI,EAAE,IAAI,CAAC,IAAI;YACf,YAAY,EAAE,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC,CAAC,IAAI,EAAE;YACpD,KAAK,EAAE,2BAA2B,CAAC,IAAI,CAAC,KAAK,CAAC;YAC9C,aAAa,EAAE,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,aAAa,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE;SAC5D,CAAC,CAAC;IAAA,CACH;IAED,KAAK,GAAuB;QAC3B,OAAO,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,CAAC;IAAA,CACpD;IAEO,YAAY,CAAC,MAA2B,EAAQ;QACvD,IAAI,CAAC;YACJ,IAAI,CAAC,EAAE,CAAC,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;YAC/D,IAAI,CAAC,EAAE,CAAC,cAAc,CAAC,IAAI,CAAC,QAAQ,EAAE,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;YAC9E,IAAI,CAAC,cAAc,EAAE,CAAC;QACvB,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,IAAI,CAAC,KAAK,CAAC,iCAAiC,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QACvG,CAAC;IAAA,CACD;IAEO,cAAc,GAAS;QAC9B,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,IAAI,CAAC,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,IAAI,IAAI,cAAc;YAAE,OAAO;QACzG,MAAM,KAAK,GAAG,IAAI,CAAC,EAAE;aACnB,YAAY,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,CAAC;aACpC,KAAK,CAAC,IAAI,CAAC;aACX,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CAAC;aACxC,KAAK,CAAC,CAAC,mBAAmB,CAAC,CAAC;QAC9B,IAAI,CAAC,EAAE,CAAC,aAAa,CAAC,IAAI,CAAC,QAAQ,EAAE,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;IAAA,CACvE;CACD;AAED,SAAS,2BAA2B,CACnC,KAAiD,EACb;IACpC,OAAO,KAAK,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;QAC5B,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,YAAY,EAAE,KAAK,CAAC,YAAY;QAChC,YAAY,EAAE,KAAK,CAAC,YAAY;QAChC,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACpD,CAAC,CAAC,CAAC;AAAA,CACJ;AAED,MAAM,UAAU,aAAa,CAAC,OAAe,EAAU;IACtD,OAAO,OAAO;SACZ,OAAO,CAAC,uCAAuC,EAAE,YAAY,CAAC;SAC9D,OAAO,CAAC,gCAAgC,EAAE,mBAAmB,CAAC;SAC9D,OAAO,CAAC,2BAA2B,EAAE,YAAY,CAAC,CAAC;AAAA,CACrD","sourcesContent":["import { appendFileSync, existsSync, mkdirSync, readFileSync, statSync, writeFileSync } from \"node:fs\";\nimport { dirname } from \"node:path\";\nimport type { ClassifiedError } from \"@caupulican/pi-agent-core\";\n\nexport interface ProviderFailureCorpusRecord {\n\tts: string;\n\tprovider?: string;\n\tmodelId?: string;\n\treason: ClassifiedError[\"reason\"];\n\tretryable: boolean;\n\tmessage: string;\n}\n\nexport interface ToolValidationFailureShapeEntry {\n\tpath: string;\n\texpectedType: string;\n\treceivedType: string;\n\tkeyword?: string;\n}\n\nexport interface ToolValidationFailureCorpusRecord {\n\tkind: \"tool_validation\";\n\tts: string;\n\tprovider?: string;\n\tmodelId?: string;\n\ttool: string;\n\tfailureModes: string[];\n\tshape: ToolValidationFailureShapeEntry[];\n\terrorKeywords: string[];\n}\n\nexport type FailureCorpusRecord = ProviderFailureCorpusRecord | ToolValidationFailureCorpusRecord;\n\nexport interface FailureCorpusStats {\n\ttotal: number;\n\tunknown: number;\n}\n\nexport interface FailureCorpusFs {\n\texistsSync(path: string): boolean;\n\tmkdirSync(path: string, options: { recursive: true }): void;\n\tappendFileSync(path: string, data: string, encoding: \"utf-8\"): void;\n\treadFileSync(path: string, encoding: \"utf-8\"): string;\n\twriteFileSync(path: string, data: string, encoding: \"utf-8\"): void;\n\tstatSync(path: string): { size: number };\n}\n\nconst DEFAULT_FS: FailureCorpusFs = { existsSync, mkdirSync, appendFileSync, readFileSync, writeFileSync, statSync };\nconst MAX_MESSAGE_CHARS = 500;\nconst MAX_FILE_BYTES = 512 * 1024;\nconst MAX_ROTATED_RECORDS = 1000;\n\nexport class FailureCorpusRecorder {\n\tprivate total = 0;\n\tprivate unknown = 0;\n\tprivate readonly filePath: string;\n\tprivate readonly fs: FailureCorpusFs;\n\tprivate readonly now: () => Date;\n\tprivate readonly debug: (message: string) => void;\n\n\tconstructor(args: { filePath: string; fs?: FailureCorpusFs; now?: () => Date; debug?: (message: string) => void }) {\n\t\tthis.filePath = args.filePath;\n\t\tthis.fs = args.fs ?? DEFAULT_FS;\n\t\tthis.now = args.now ?? (() => new Date());\n\t\tthis.debug = args.debug ?? (() => {});\n\t}\n\n\trecord(args: { provider?: string; modelId?: string; message: string; classified: ClassifiedError }): void {\n\t\tthis.total += 1;\n\t\tif (args.classified.reason === \"unknown\") this.unknown += 1;\n\t\tthis.appendRecord({\n\t\t\tts: this.now().toISOString(),\n\t\t\tprovider: args.provider,\n\t\t\tmodelId: args.modelId,\n\t\t\treason: args.classified.reason,\n\t\t\tretryable: args.classified.retryable,\n\t\t\tmessage: redactSecrets(args.message).slice(0, MAX_MESSAGE_CHARS),\n\t\t});\n\t}\n\n\trecordToolValidation(args: {\n\t\tprovider?: string;\n\t\tmodelId?: string;\n\t\ttool: string;\n\t\tfailureModes: readonly string[];\n\t\tshape: readonly ToolValidationFailureShapeEntry[];\n\t\terrorKeywords?: readonly string[];\n\t}): void {\n\t\tthis.appendRecord({\n\t\t\tkind: \"tool_validation\",\n\t\t\tts: this.now().toISOString(),\n\t\t\tprovider: args.provider,\n\t\t\tmodelId: args.modelId,\n\t\t\ttool: args.tool,\n\t\t\tfailureModes: [...new Set(args.failureModes)].sort(),\n\t\t\tshape: sanitizeToolValidationShape(args.shape),\n\t\t\terrorKeywords: [...new Set(args.errorKeywords ?? [])].sort(),\n\t\t});\n\t}\n\n\tstats(): FailureCorpusStats {\n\t\treturn { total: this.total, unknown: this.unknown };\n\t}\n\n\tprivate appendRecord(record: FailureCorpusRecord): void {\n\t\ttry {\n\t\t\tthis.fs.mkdirSync(dirname(this.filePath), { recursive: true });\n\t\t\tthis.fs.appendFileSync(this.filePath, `${JSON.stringify(record)}\\n`, \"utf-8\");\n\t\t\tthis.rotateIfNeeded();\n\t\t} catch (error) {\n\t\t\tthis.debug(`failure corpus write skipped: ${error instanceof Error ? error.message : String(error)}`);\n\t\t}\n\t}\n\n\tprivate rotateIfNeeded(): void {\n\t\tif (!this.fs.existsSync(this.filePath) || this.fs.statSync(this.filePath).size <= MAX_FILE_BYTES) return;\n\t\tconst lines = this.fs\n\t\t\t.readFileSync(this.filePath, \"utf-8\")\n\t\t\t.split(\"\\n\")\n\t\t\t.filter((line) => line.trim().length > 0)\n\t\t\t.slice(-MAX_ROTATED_RECORDS);\n\t\tthis.fs.writeFileSync(this.filePath, `${lines.join(\"\\n\")}\\n`, \"utf-8\");\n\t}\n}\n\nfunction sanitizeToolValidationShape(\n\tshape: readonly ToolValidationFailureShapeEntry[],\n): ToolValidationFailureShapeEntry[] {\n\treturn shape.map((entry) => ({\n\t\tpath: entry.path,\n\t\texpectedType: entry.expectedType,\n\t\treceivedType: entry.receivedType,\n\t\t...(entry.keyword ? { keyword: entry.keyword } : {}),\n\t}));\n}\n\nexport function redactSecrets(message: string): string {\n\treturn message\n\t\t.replace(/sk-(?:proj-|ant-)?[A-Za-z0-9._-]{8,}/g, \"[redacted]\")\n\t\t.replace(/Bearer\\s+[A-Za-z0-9._~+/=-]+/gi, \"Bearer [redacted]\")\n\t\t.replace(/[A-Za-z0-9+/]{40,}={0,2}/g, \"[redacted]\");\n}\n"]}
+{"version":3,"file":"failure-corpus.js","sourceRoot":"","sources":["../../src/core/failure-corpus.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,UAAU,EAAE,SAAS,EAAE,YAAY,EAAE,QAAQ,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AACvG,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AA8CpC,MAAM,UAAU,GAAoB,EAAE,UAAU,EAAE,SAAS,EAAE,cAAc,EAAE,YAAY,EAAE,aAAa,EAAE,QAAQ,EAAE,CAAC;AACrH,MAAM,iBAAiB,GAAG,GAAG,CAAC;AAC9B,MAAM,cAAc,GAAG,GAAG,GAAG,IAAI,CAAC;AAClC,MAAM,mBAAmB,GAAG,IAAI,CAAC;AAEjC,MAAM,UAAU,uCAAuC,CAAC,IAQvD,EAAqC;IACrC,OAAO;QACN,IAAI,EAAE,iBAAiB;QACvB,EAAE,EAAE,IAAI,CAAC,EAAE;QACX,QAAQ,EAAE,IAAI,CAAC,QAAQ;QACvB,OAAO,EAAE,IAAI,CAAC,OAAO;QACrB,IAAI,EAAE,IAAI,CAAC,IAAI;QACf,YAAY,EAAE,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC,CAAC,IAAI,EAAE;QACpD,KAAK,EAAE,2BAA2B,CAAC,IAAI,CAAC,KAAK,CAAC;QAC9C,aAAa,EAAE,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,aAAa,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE;KAC5D,CAAC;AAAA,CACF;AAED,MAAM,UAAU,wBAAwB,CACvC,QAAgB,EAChB,MAA2B,EAC3B,EAAE,GAAoB,UAAU,EACzB;IACP,EAAE,CAAC,SAAS,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACrD,EAAE,CAAC,cAAc,CAAC,QAAQ,EAAE,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;IACpE,2BAA2B,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC;AAAA,CAC1C;AAED,SAAS,2BAA2B,CAAC,QAAgB,EAAE,EAAmB,EAAQ;IACjF,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC,IAAI,IAAI,cAAc;QAAE,OAAO;IACrF,MAAM,KAAK,GAAG,EAAE;SACd,YAAY,CAAC,QAAQ,EAAE,OAAO,CAAC;SAC/B,KAAK,CAAC,IAAI,CAAC;SACX,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CAAC;SACxC,KAAK,CAAC,CAAC,mBAAmB,CAAC,CAAC;IAC9B,EAAE,CAAC,aAAa,CAAC,QAAQ,EAAE,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;AAAA,CAC7D;AAED,MAAM,OAAO,qBAAqB;IACzB,KAAK,GAAG,CAAC,CAAC;IACV,OAAO,GAAG,CAAC,CAAC;IACH,QAAQ,CAAS;IACjB,EAAE,CAAkB;IACpB,GAAG,CAAa;IAChB,KAAK,CAA4B;IAElD,YAAY,IAAqG,EAAE;QAClH,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC,QAAQ,CAAC;QAC9B,IAAI,CAAC,EAAE,GAAG,IAAI,CAAC,EAAE,IAAI,UAAU,CAAC;QAChC,IAAI,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC;QAC1C,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,CAAC,GAAG,EAAE,CAAC,EAAC,CAAC,CAAC,CAAC;IAAA,CACtC;IAED,MAAM,CAAC,IAA2F,EAAQ;QACzG,IAAI,CAAC,KAAK,IAAI,CAAC,CAAC;QAChB,IAAI,IAAI,CAAC,UAAU,CAAC,MAAM,KAAK,SAAS;YAAE,IAAI,CAAC,OAAO,IAAI,CAAC,CAAC;QAC5D,IAAI,CAAC,YAAY,CAAC;YACjB,EAAE,EAAE,IAAI,CAAC,GAAG,EAAE,CAAC,WAAW,EAAE;YAC5B,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,OAAO,EAAE,IAAI,CAAC,OAAO;YACrB,MAAM,EAAE,IAAI,CAAC,UAAU,CAAC,MAAM;YAC9B,SAAS,EAAE,IAAI,CAAC,UAAU,CAAC,SAAS;YACpC,OAAO,EAAE,aAAa,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,iBAAiB,CAAC;SAChE,CAAC,CAAC;IAAA,CACH;IAED,oBAAoB,CAAC,IAOpB,EAAQ;QACR,IAAI,CAAC,YAAY,CAAC,uCAAuC,CAAC,EAAE,GAAG,IAAI,EAAE,EAAE,EAAE,IAAI,CAAC,GAAG,EAAE,CAAC,WAAW,EAAE,EAAE,CAAC,CAAC,CAAC;IAAA,CACtG;IAED,KAAK,GAAuB;QAC3B,OAAO,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,CAAC;IAAA,CACpD;IAEO,YAAY,CAAC,MAA2B,EAAQ;QACvD,IAAI,CAAC;YACJ,wBAAwB,CAAC,IAAI,CAAC,QAAQ,EAAE,MAAM,EAAE,IAAI,CAAC,EAAE,CAAC,CAAC;QAC1D,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,IAAI,CAAC,KAAK,CAAC,iCAAiC,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QACvG,CAAC;IAAA,CACD;CACD;AAED,SAAS,2BAA2B,CACnC,KAAiD,EACb;IACpC,OAAO,KAAK,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;QAC5B,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,YAAY,EAAE,KAAK,CAAC,YAAY;QAChC,YAAY,EAAE,KAAK,CAAC,YAAY;QAChC,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACpD,CAAC,CAAC,CAAC;AAAA,CACJ;AAED,MAAM,UAAU,aAAa,CAAC,OAAe,EAAU;IACtD,OAAO,OAAO;SACZ,OAAO,CAAC,uCAAuC,EAAE,YAAY,CAAC;SAC9D,OAAO,CAAC,gCAAgC,EAAE,mBAAmB,CAAC;SAC9D,OAAO,CAAC,2BAA2B,EAAE,YAAY,CAAC,CAAC;AAAA,CACrD","sourcesContent":["import { appendFileSync, existsSync, mkdirSync, readFileSync, statSync, writeFileSync } from \"node:fs\";\nimport { dirname } from \"node:path\";\nimport type { ClassifiedError } from \"@caupulican/pi-agent-core\";\n\nexport interface ProviderFailureCorpusRecord {\n\tts: string;\n\tprovider?: string;\n\tmodelId?: string;\n\treason: ClassifiedError[\"reason\"];\n\tretryable: boolean;\n\tmessage: string;\n}\n\nexport interface ToolValidationFailureShapeEntry {\n\tpath: string;\n\texpectedType: string;\n\treceivedType: string;\n\tkeyword?: string;\n}\n\nexport interface ToolValidationFailureCorpusRecord {\n\tkind: \"tool_validation\";\n\tts: string;\n\tprovider?: string;\n\tmodelId?: string;\n\ttool: string;\n\tfailureModes: string[];\n\tshape: ToolValidationFailureShapeEntry[];\n\terrorKeywords: string[];\n}\n\nexport type FailureCorpusRecord = ProviderFailureCorpusRecord | ToolValidationFailureCorpusRecord;\n\nexport interface FailureCorpusStats {\n\ttotal: number;\n\tunknown: number;\n}\n\nexport interface FailureCorpusFs {\n\texistsSync(path: string): boolean;\n\tmkdirSync(path: string, options: { recursive: true }): void;\n\tappendFileSync(path: string, data: string, encoding: \"utf-8\"): void;\n\treadFileSync(path: string, encoding: \"utf-8\"): string;\n\twriteFileSync(path: string, data: string, encoding: \"utf-8\"): void;\n\tstatSync(path: string): { size: number };\n}\n\nconst DEFAULT_FS: FailureCorpusFs = { existsSync, mkdirSync, appendFileSync, readFileSync, writeFileSync, statSync };\nconst MAX_MESSAGE_CHARS = 500;\nconst MAX_FILE_BYTES = 512 * 1024;\nconst MAX_ROTATED_RECORDS = 1000;\n\nexport function createToolValidationFailureCorpusRecord(args: {\n\tprovider?: string;\n\tmodelId?: string;\n\ttool: string;\n\tfailureModes: readonly string[];\n\tshape: readonly ToolValidationFailureShapeEntry[];\n\terrorKeywords?: readonly string[];\n\tts: string;\n}): ToolValidationFailureCorpusRecord {\n\treturn {\n\t\tkind: \"tool_validation\",\n\t\tts: args.ts,\n\t\tprovider: args.provider,\n\t\tmodelId: args.modelId,\n\t\ttool: args.tool,\n\t\tfailureModes: [...new Set(args.failureModes)].sort(),\n\t\tshape: sanitizeToolValidationShape(args.shape),\n\t\terrorKeywords: [...new Set(args.errorKeywords ?? [])].sort(),\n\t};\n}\n\nexport function writeFailureCorpusRecord(\n\tfilePath: string,\n\trecord: FailureCorpusRecord,\n\tfs: FailureCorpusFs = DEFAULT_FS,\n): void {\n\tfs.mkdirSync(dirname(filePath), { recursive: true });\n\tfs.appendFileSync(filePath, `${JSON.stringify(record)}\\n`, \"utf-8\");\n\trotateFailureCorpusIfNeeded(filePath, fs);\n}\n\nfunction rotateFailureCorpusIfNeeded(filePath: string, fs: FailureCorpusFs): void {\n\tif (!fs.existsSync(filePath) || fs.statSync(filePath).size <= MAX_FILE_BYTES) return;\n\tconst lines = fs\n\t\t.readFileSync(filePath, \"utf-8\")\n\t\t.split(\"\\n\")\n\t\t.filter((line) => line.trim().length > 0)\n\t\t.slice(-MAX_ROTATED_RECORDS);\n\tfs.writeFileSync(filePath, `${lines.join(\"\\n\")}\\n`, \"utf-8\");\n}\n\nexport class FailureCorpusRecorder {\n\tprivate total = 0;\n\tprivate unknown = 0;\n\tprivate readonly filePath: string;\n\tprivate readonly fs: FailureCorpusFs;\n\tprivate readonly now: () => Date;\n\tprivate readonly debug: (message: string) => void;\n\n\tconstructor(args: { filePath: string; fs?: FailureCorpusFs; now?: () => Date; debug?: (message: string) => void }) {\n\t\tthis.filePath = args.filePath;\n\t\tthis.fs = args.fs ?? DEFAULT_FS;\n\t\tthis.now = args.now ?? (() => new Date());\n\t\tthis.debug = args.debug ?? (() => {});\n\t}\n\n\trecord(args: { provider?: string; modelId?: string; message: string; classified: ClassifiedError }): void {\n\t\tthis.total += 1;\n\t\tif (args.classified.reason === \"unknown\") this.unknown += 1;\n\t\tthis.appendRecord({\n\t\t\tts: this.now().toISOString(),\n\t\t\tprovider: args.provider,\n\t\t\tmodelId: args.modelId,\n\t\t\treason: args.classified.reason,\n\t\t\tretryable: args.classified.retryable,\n\t\t\tmessage: redactSecrets(args.message).slice(0, MAX_MESSAGE_CHARS),\n\t\t});\n\t}\n\n\trecordToolValidation(args: {\n\t\tprovider?: string;\n\t\tmodelId?: string;\n\t\ttool: string;\n\t\tfailureModes: readonly string[];\n\t\tshape: readonly ToolValidationFailureShapeEntry[];\n\t\terrorKeywords?: readonly string[];\n\t}): void {\n\t\tthis.appendRecord(createToolValidationFailureCorpusRecord({ ...args, ts: this.now().toISOString() }));\n\t}\n\n\tstats(): FailureCorpusStats {\n\t\treturn { total: this.total, unknown: this.unknown };\n\t}\n\n\tprivate appendRecord(record: FailureCorpusRecord): void {\n\t\ttry {\n\t\t\twriteFailureCorpusRecord(this.filePath, record, this.fs);\n\t\t} catch (error) {\n\t\t\tthis.debug(`failure corpus write skipped: ${error instanceof Error ? error.message : String(error)}`);\n\t\t}\n\t}\n}\n\nfunction sanitizeToolValidationShape(\n\tshape: readonly ToolValidationFailureShapeEntry[],\n): ToolValidationFailureShapeEntry[] {\n\treturn shape.map((entry) => ({\n\t\tpath: entry.path,\n\t\texpectedType: entry.expectedType,\n\t\treceivedType: entry.receivedType,\n\t\t...(entry.keyword ? { keyword: entry.keyword } : {}),\n\t}));\n}\n\nexport function redactSecrets(message: string): string {\n\treturn message\n\t\t.replace(/sk-(?:proj-|ant-)?[A-Za-z0-9._-]{8,}/g, \"[REDACTED]\")\n\t\t.replace(/Bearer\\s+[A-Za-z0-9._~+/=-]+/gi, \"Bearer [REDACTED]\")\n\t\t.replace(/[A-Za-z0-9+/]{40,}={0,2}/g, \"[REDACTED]\");\n}\n"]}

package/dist/core/model-router/intent-classifier.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"intent-classifier.d.ts","sourceRoot":"","sources":["../../../src/core/model-router/intent-classifier.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,0BAA0B,CAAC;AAE9D,MAAM,MAAM,iBAAiB,GAAG,UAAU,GAAG,QAAQ,CAAC;AAiCtD,wBAAgB,wBAAwB,CAAC,MAAM,EAAE,MAAM,GAAG,aAAa,CA0ItE;AAED,wBAAgB,yBAAyB,CAAC,MAAM,EAAE,MAAM,GAAG,iBAAiB,CAG3E","sourcesContent":["import type { RouteDecision } from \"../autonomy/contracts.ts\";\n\nexport type ModelRouterIntent = \"research\" | \"modify\";\n\nconst EXPLICIT_MODIFY_REQUEST_RE =\n\t/^(?:can you|could you|please|pls|go ahead and|let'?s|i need you to|we need to|you should)\\s+.*\\b(add|apply|build|change|commit|create|delete|edit|fix|generate|implement|install|modify|patch|refactor|remove|rename|replace|run|test|update|write|publish|release|push|deploy|tag|reset|clean|rewrite)\\b/i;\n\nconst READ_ONLY_QUESTION_RE =\n\t/^(?:(?:can you|could you|please|pls|go ahead and|let'?s|i need you to|we need to|you should)\\s+)?(?:how|what|why|when|where|which|who|explain|summarize|compare|describe|list|show|search|find|view|read|locate)\\b/i;\n\nconst RELEASE_PUBLISH_RE = /\\b(publish|release|push|deploy|tag)\\b/i;\nconst SECURITY_AUTH_RE = /\\b(auth|token|credential|credentials|secret|api[-_]key)\\b/i;\nconst DESTRUCTIVE_RE = /\\b(delete|reset|rm\\s+-rf|clean)\\b/i;\n\nconst SELF_MOD_MUTATE_RE =\n\t/\\b(modify|change|write|update|edit|delete|add|remove)\\s+.*\\b(skills|prompts|settings|tools|behavior)\\b|self[-_]modification/i;\nconst ARCHITECTURE_MUTATE_RE = /\\b(rewrite|redesign|change|modify|rearchitect)\\s+.*\\b(architecture|architect)\\b/i;\n\n// Planning floor: plans steer all downstream work, so planning never routes cheap by default.\n// Only the route judge may downgrade a planning prompt back to cheap (explicit trivial verdict).\n// Core terms are always planning; design/architecture words count only with prospective phrasing,\n// so lookups like \"show me the architecture\" stay cheap.\nconst PLANNING_CORE_RE = /\\b(plan|planning|roadmap|strategy)\\b/i;\nconst PLANNING_DESIGN_WORD_RE = /\\b(design|architect\\w*|structure|approach)\\b/i;\nconst PLANNING_PROSPECTIVE_RE =\n\t/\\b(how (?:should|would|do we|can we)|what(?:'s| is) the best|propose|draft|come up with|figure out|decide (?:on|how))\\b/i;\n\nfunction isPlanningPrompt(text: string): boolean {\n\treturn PLANNING_CORE_RE.test(text) || (PLANNING_DESIGN_WORD_RE.test(text) && PLANNING_PROSPECTIVE_RE.test(text));\n}\n\nconst REFACTOR_RE = /\\b(refactor|refactoring)\\b/i;\nconst TEST_VALIDATION_RE = /\\b(test|testing|validation|lint|vitest|jest|run)\\b/i;\nconst IMPLEMENT_RE = /\\b(implement|fix|apply|change|update|create|write|generate|modify|edit|patch|add)\\b/i;\n\nexport function classifyModelRouterRoute(prompt: string): RouteDecision {\n\tconst text = prompt.trim();\n\n\tif (text.length === 0) {\n\t\treturn {\n\t\t\ttier: \"cheap\",\n\t\t\trisk: \"read-only\",\n\t\t\tconfidence: 0.1,\n\t\t\treasonCode: \"empty_prompt\",\n\t\t\treasons: [\"Empty or whitespace prompt\"],\n\t\t};\n\t}\n\n\t// 1. Explicit read-only questions/lookups dominate (unless prefixed by explicit mutation verb).\n\t// Planning-shaped questions are the exception: a plan steers expensive downstream work, so the\n\t// floor is medium even when phrased as a question.\n\tif (READ_ONLY_QUESTION_RE.test(text) && !EXPLICIT_MODIFY_REQUEST_RE.test(text)) {\n\t\tif (isPlanningPrompt(text)) {\n\t\t\treturn {\n\t\t\t\ttier: \"medium\",\n\t\t\t\trisk: \"read-only\",\n\t\t\t\tconfidence: 0.75,\n\t\t\t\treasonCode: \"planning_min_medium\",\n\t\t\t\treasons: [\"Planning/design prompts never route cheap by default; a judge may deem them trivial\"],\n\t\t\t};\n\t\t}\n\t\treturn {\n\t\t\ttier: \"cheap\",\n\t\t\trisk: \"read-only\",\n\t\t\tconfidence: 0.9,\n\t\t\treasonCode: \"read_only_question\",\n\t\t\treasons: [\"Prompt asks a question or requests an explanation, search, or lookup\"],\n\t\t};\n\t}\n\n\t// Helper function to match patterns and return appropriate decision\n\tfunction matchKeywords(input: string): RouteDecision | null {\n\t\t// A. High-risk / approval-required/expensive signals\n\t\tif (RELEASE_PUBLISH_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"expensive\",\n\t\t\t\trisk: \"approval-required\",\n\t\t\t\tconfidence: 0.9,\n\t\t\t\treasonCode: \"release_or_publish\",\n\t\t\t\treasons: [\"Prompt mentions publishing, releasing, pushing, or deploying\"],\n\t\t\t};\n\t\t}\n\t\tif (SECURITY_AUTH_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"expensive\",\n\t\t\t\trisk: \"high-impact\",\n\t\t\t\tconfidence: 0.95,\n\t\t\t\treasonCode: \"security_or_auth\",\n\t\t\t\treasons: [\"Prompt mentions credentials, authentication, tokens, or secrets\"],\n\t\t\t};\n\t\t}\n\t\tif (DESTRUCTIVE_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"expensive\",\n\t\t\t\trisk: \"approval-required\",\n\t\t\t\tconfidence: 0.85,\n\t\t\t\treasonCode: \"destructive_or_git_history\",\n\t\t\t\treasons: [\"Prompt mentions deleting, resetting, cleaning, or destructive operations\"],\n\t\t\t};\n\t\t}\n\t\tif (SELF_MOD_MUTATE_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"expensive\",\n\t\t\t\trisk: \"approval-required\",\n\t\t\t\tconfidence: 0.9,\n\t\t\t\treasonCode: \"settings_or_self_modification\",\n\t\t\t\treasons: [\"Prompt mentions modifying skills, prompts, settings, tools, or self-modification\"],\n\t\t\t};\n\t\t}\n\t\tif (ARCHITECTURE_MUTATE_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"expensive\",\n\t\t\t\trisk: \"high-impact\",\n\t\t\t\tconfidence: 0.9,\n\t\t\t\treasonCode: \"architecture_or_ambiguous\",\n\t\t\t\treasons: [\"Prompt mentions core architecture or rewrite\"],\n\t\t\t};\n\t\t}\n\n\t\t// B. Explicit implementation/scoped-write signals route medium\n\t\tif (isPlanningPrompt(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"medium\",\n\t\t\t\trisk: \"read-only\",\n\t\t\t\tconfidence: 0.75,\n\t\t\t\treasonCode: \"planning_min_medium\",\n\t\t\t\treasons: [\"Planning/design prompts never route cheap by default; a judge may deem them trivial\"],\n\t\t\t};\n\t\t}\n\t\tif (REFACTOR_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"medium\",\n\t\t\t\trisk: \"scoped-write\",\n\t\t\t\tconfidence: 0.8,\n\t\t\t\treasonCode: \"mechanical_refactor\",\n\t\t\t\treasons: [\"Prompt mentions refactoring code structure\"],\n\t\t\t};\n\t\t}\n\t\tif (TEST_VALIDATION_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"medium\",\n\t\t\t\trisk: \"scoped-write\",\n\t\t\t\tconfidence: 0.8,\n\t\t\t\treasonCode: \"test_or_validation\",\n\t\t\t\treasons: [\"Prompt mentions testing, validation, or linting\"],\n\t\t\t};\n\t\t}\n\t\tif (IMPLEMENT_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"medium\",\n\t\t\t\trisk: \"scoped-write\",\n\t\t\t\tconfidence: 0.85,\n\t\t\t\treasonCode: \"normal_implementation\",\n\t\t\t\treasons: [\"Prompt mentions implementing, updating, creating, or modifying code\"],\n\t\t\t};\n\t\t}\n\n\t\treturn null;\n\t}\n\n\tconst match = matchKeywords(text);\n\tif (match) {\n\t\treturn match;\n\t}\n\n\t// 4. Default fallbacks\n\treturn {\n\t\ttier: \"cheap\",\n\t\trisk: \"read-only\",\n\t\tconfidence: 0.5,\n\t\treasonCode: \"default_read_only\",\n\t\treasons: [\"No explicit implementation, destructive, or release patterns detected\"],\n\t};\n}\n\nexport function classifyModelRouterIntent(prompt: string): ModelRouterIntent {\n\tconst decision = classifyModelRouterRoute(prompt);\n\treturn decision.tier === \"cheap\" ? \"research\" : \"modify\";\n}\n"]}
+{"version":3,"file":"intent-classifier.d.ts","sourceRoot":"","sources":["../../../src/core/model-router/intent-classifier.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,0BAA0B,CAAC;AAE9D,MAAM,MAAM,iBAAiB,GAAG,UAAU,GAAG,QAAQ,CAAC;AAiCtD,wBAAgB,wBAAwB,CAAC,MAAM,EAAE,MAAM,GAAG,aAAa,CA0ItE;AAED,wBAAgB,yBAAyB,CAAC,MAAM,EAAE,MAAM,GAAG,iBAAiB,CAG3E","sourcesContent":["import type { RouteDecision } from \"../autonomy/contracts.ts\";\n\nexport type ModelRouterIntent = \"research\" | \"modify\";\n\nconst EXPLICIT_MODIFY_REQUEST_RE =\n\t/^(?:can you|could you|please|pls|go ahead and|let'?s|i need you to|we need to|you should)\\s+.*\\b(add|apply|build|change|commit|create|delete|edit|fix|generate|implement|install|modify|patch|refactor|remove|rename|replace|run|test|update|write|publish|release|push|deploy|tag|reset|clean|rewrite)\\b/i;\n\nconst READ_ONLY_QUESTION_RE =\n\t/^(?:(?:can you|could you|please|pls|go ahead and|let'?s|i need you to|we need to|you should)\\s+)?(?:how|what|why|when|where|which|who|explain|summarize|compare|describe|list|show|search|find|view|read|locate)\\b/i;\n\nconst RELEASE_PUBLISH_RE = /\\b(publish|release|push|deploy|tag)\\b/i;\nconst SECURITY_AUTH_RE = /\\b(auth|token|credential|credentials|secret|api[-_]key)\\b/i;\nconst DESTRUCTIVE_RE = /\\b(delete|reset|rm\\s+-rf|clean)\\b/i;\n\nconst SELF_MOD_MUTATE_RE =\n\t/\\b(modify|change|write|update|edit|delete|add|remove)\\s+.*\\b(skills|prompts|settings|tools|behavior)\\b|self[-_]modification/i;\nconst ARCHITECTURE_MUTATE_RE = /\\b(rewrite|redesign|change|modify|rearchitect)\\s+.*\\b(architecture|architect)\\b/i;\n\n// Planning floor: plans steer all downstream work, so planning never routes cheap by default.\n// Only the route judge may downgrade a planning prompt back to cheap (explicit trivial verdict).\n// Core terms are always planning; design/architecture words count only with prospective phrasing,\n// so lookups like \"show me the architecture\" stay cheap.\nconst PLANNING_CORE_RE = /\\b(plan|planning|roadmap|strategy)\\b/i;\nconst PLANNING_DESIGN_WORD_RE = /\\b(design|architect\\w*|structure|approach)\\b/i;\nconst PLANNING_PROSPECTIVE_RE =\n\t/\\b(how (?:should|would|do we|can we)|what(?:'s| is) the (?:best|cleanest|right)|propose|draft|come up with|figure out|decide (?:on|how))\\b/i;\n\nfunction isPlanningPrompt(text: string): boolean {\n\treturn PLANNING_CORE_RE.test(text) || (PLANNING_DESIGN_WORD_RE.test(text) && PLANNING_PROSPECTIVE_RE.test(text));\n}\n\nconst REFACTOR_RE = /\\b(refactor|refactoring)\\b/i;\nconst TEST_VALIDATION_RE = /\\b(test|testing|validation|lint|vitest|jest|run)\\b/i;\nconst IMPLEMENT_RE = /\\b(implement|fix|apply|change|update|create|write|generate|modify|edit|patch|add)\\b/i;\n\nexport function classifyModelRouterRoute(prompt: string): RouteDecision {\n\tconst text = prompt.trim();\n\n\tif (text.length === 0) {\n\t\treturn {\n\t\t\ttier: \"cheap\",\n\t\t\trisk: \"read-only\",\n\t\t\tconfidence: 0.1,\n\t\t\treasonCode: \"empty_prompt\",\n\t\t\treasons: [\"Empty or whitespace prompt\"],\n\t\t};\n\t}\n\n\t// 1. Explicit read-only questions/lookups dominate (unless prefixed by explicit mutation verb).\n\t// Planning-shaped questions are the exception: a plan steers expensive downstream work, so the\n\t// floor is medium even when phrased as a question.\n\tif (READ_ONLY_QUESTION_RE.test(text) && !EXPLICIT_MODIFY_REQUEST_RE.test(text)) {\n\t\tif (isPlanningPrompt(text)) {\n\t\t\treturn {\n\t\t\t\ttier: \"medium\",\n\t\t\t\trisk: \"read-only\",\n\t\t\t\tconfidence: 0.75,\n\t\t\t\treasonCode: \"planning_min_medium\",\n\t\t\t\treasons: [\"Planning/design prompts never route cheap by default; a judge may deem them trivial\"],\n\t\t\t};\n\t\t}\n\t\treturn {\n\t\t\ttier: \"cheap\",\n\t\t\trisk: \"read-only\",\n\t\t\tconfidence: 0.9,\n\t\t\treasonCode: \"read_only_question\",\n\t\t\treasons: [\"Prompt asks a question or requests an explanation, search, or lookup\"],\n\t\t};\n\t}\n\n\t// Helper function to match patterns and return appropriate decision\n\tfunction matchKeywords(input: string): RouteDecision | null {\n\t\t// A. High-risk / approval-required/expensive signals\n\t\tif (RELEASE_PUBLISH_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"expensive\",\n\t\t\t\trisk: \"approval-required\",\n\t\t\t\tconfidence: 0.9,\n\t\t\t\treasonCode: \"release_or_publish\",\n\t\t\t\treasons: [\"Prompt mentions publishing, releasing, pushing, or deploying\"],\n\t\t\t};\n\t\t}\n\t\tif (SECURITY_AUTH_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"expensive\",\n\t\t\t\trisk: \"high-impact\",\n\t\t\t\tconfidence: 0.95,\n\t\t\t\treasonCode: \"security_or_auth\",\n\t\t\t\treasons: [\"Prompt mentions credentials, authentication, tokens, or secrets\"],\n\t\t\t};\n\t\t}\n\t\tif (DESTRUCTIVE_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"expensive\",\n\t\t\t\trisk: \"approval-required\",\n\t\t\t\tconfidence: 0.85,\n\t\t\t\treasonCode: \"destructive_or_git_history\",\n\t\t\t\treasons: [\"Prompt mentions deleting, resetting, cleaning, or destructive operations\"],\n\t\t\t};\n\t\t}\n\t\tif (SELF_MOD_MUTATE_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"expensive\",\n\t\t\t\trisk: \"approval-required\",\n\t\t\t\tconfidence: 0.9,\n\t\t\t\treasonCode: \"settings_or_self_modification\",\n\t\t\t\treasons: [\"Prompt mentions modifying skills, prompts, settings, tools, or self-modification\"],\n\t\t\t};\n\t\t}\n\t\tif (ARCHITECTURE_MUTATE_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"expensive\",\n\t\t\t\trisk: \"high-impact\",\n\t\t\t\tconfidence: 0.9,\n\t\t\t\treasonCode: \"architecture_or_ambiguous\",\n\t\t\t\treasons: [\"Prompt mentions core architecture or rewrite\"],\n\t\t\t};\n\t\t}\n\n\t\t// B. Explicit implementation/scoped-write signals route medium\n\t\tif (isPlanningPrompt(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"medium\",\n\t\t\t\trisk: \"read-only\",\n\t\t\t\tconfidence: 0.75,\n\t\t\t\treasonCode: \"planning_min_medium\",\n\t\t\t\treasons: [\"Planning/design prompts never route cheap by default; a judge may deem them trivial\"],\n\t\t\t};\n\t\t}\n\t\tif (REFACTOR_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"medium\",\n\t\t\t\trisk: \"scoped-write\",\n\t\t\t\tconfidence: 0.8,\n\t\t\t\treasonCode: \"mechanical_refactor\",\n\t\t\t\treasons: [\"Prompt mentions refactoring code structure\"],\n\t\t\t};\n\t\t}\n\t\tif (TEST_VALIDATION_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"medium\",\n\t\t\t\trisk: \"scoped-write\",\n\t\t\t\tconfidence: 0.8,\n\t\t\t\treasonCode: \"test_or_validation\",\n\t\t\t\treasons: [\"Prompt mentions testing, validation, or linting\"],\n\t\t\t};\n\t\t}\n\t\tif (IMPLEMENT_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"medium\",\n\t\t\t\trisk: \"scoped-write\",\n\t\t\t\tconfidence: 0.85,\n\t\t\t\treasonCode: \"normal_implementation\",\n\t\t\t\treasons: [\"Prompt mentions implementing, updating, creating, or modifying code\"],\n\t\t\t};\n\t\t}\n\n\t\treturn null;\n\t}\n\n\tconst match = matchKeywords(text);\n\tif (match) {\n\t\treturn match;\n\t}\n\n\t// 4. Default fallbacks\n\treturn {\n\t\ttier: \"cheap\",\n\t\trisk: \"read-only\",\n\t\tconfidence: 0.5,\n\t\treasonCode: \"default_read_only\",\n\t\treasons: [\"No explicit implementation, destructive, or release patterns detected\"],\n\t};\n}\n\nexport function classifyModelRouterIntent(prompt: string): ModelRouterIntent {\n\tconst decision = classifyModelRouterRoute(prompt);\n\treturn decision.tier === \"cheap\" ? \"research\" : \"modify\";\n}\n"]}

package/dist/core/model-router/intent-classifier.js

@@ -11,7 +11,7 @@ const ARCHITECTURE_MUTATE_RE = /\b(rewrite|redesign|change|modify|rearchitect)\s
 // so lookups like "show me the architecture" stay cheap.
 const PLANNING_CORE_RE = /\b(plan|planning|roadmap|strategy)\b/i;
 const PLANNING_DESIGN_WORD_RE = /\b(design|architect\w*|structure|approach)\b/i;
-const PLANNING_PROSPECTIVE_RE = /\b(how (?:should|would|do we|can we)|what(?:'s| is) the best|propose|draft|come up with|figure out|decide (?:on|how))\b/i;
+const PLANNING_PROSPECTIVE_RE = /\b(how (?:should|would|do we|can we)|what(?:'s| is) the (?:best|cleanest|right)|propose|draft|come up with|figure out|decide (?:on|how))\b/i;
 function isPlanningPrompt(text) {
     return PLANNING_CORE_RE.test(text) || (PLANNING_DESIGN_WORD_RE.test(text) && PLANNING_PROSPECTIVE_RE.test(text));
 }

package/dist/core/model-router/intent-classifier.js.map

@@ -1 +1 @@
-{"version":3,"file":"intent-classifier.js","sourceRoot":"","sources":["../../../src/core/model-router/intent-classifier.ts"],"names":[],"mappings":"AAIA,MAAM,0BAA0B,GAC/B,4SAA4S,CAAC;AAE9S,MAAM,qBAAqB,GAC1B,qNAAqN,CAAC;AAEvN,MAAM,kBAAkB,GAAG,wCAAwC,CAAC;AACpE,MAAM,gBAAgB,GAAG,4DAA4D,CAAC;AACtF,MAAM,cAAc,GAAG,oCAAoC,CAAC;AAE5D,MAAM,kBAAkB,GACvB,8HAA8H,CAAC;AAChI,MAAM,sBAAsB,GAAG,kFAAkF,CAAC;AAElH,8FAA8F;AAC9F,iGAAiG;AACjG,kGAAkG;AAClG,yDAAyD;AACzD,MAAM,gBAAgB,GAAG,uCAAuC,CAAC;AACjE,MAAM,uBAAuB,GAAG,+CAA+C,CAAC;AAChF,MAAM,uBAAuB,GAC5B,0HAA0H,CAAC;AAE5H,SAAS,gBAAgB,CAAC,IAAY,EAAW;IAChD,OAAO,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,uBAAuB,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,uBAAuB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;AAAA,CACjH;AAED,MAAM,WAAW,GAAG,6BAA6B,CAAC;AAClD,MAAM,kBAAkB,GAAG,qDAAqD,CAAC;AACjF,MAAM,YAAY,GAAG,sFAAsF,CAAC;AAE5G,MAAM,UAAU,wBAAwB,CAAC,MAAc,EAAiB;IACvE,MAAM,IAAI,GAAG,MAAM,CAAC,IAAI,EAAE,CAAC;IAE3B,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACvB,OAAO;YACN,IAAI,EAAE,OAAO;YACb,IAAI,EAAE,WAAW;YACjB,UAAU,EAAE,GAAG;YACf,UAAU,EAAE,cAAc;YAC1B,OAAO,EAAE,CAAC,4BAA4B,CAAC;SACvC,CAAC;IACH,CAAC;IAED,gGAAgG;IAChG,+FAA+F;IAC/F,mDAAmD;IACnD,IAAI,qBAAqB,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,0BAA0B,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;QAChF,IAAI,gBAAgB,CAAC,IAAI,CAAC,EAAE,CAAC;YAC5B,OAAO;gBACN,IAAI,EAAE,QAAQ;gBACd,IAAI,EAAE,WAAW;gBACjB,UAAU,EAAE,IAAI;gBAChB,UAAU,EAAE,qBAAqB;gBACjC,OAAO,EAAE,CAAC,qFAAqF,CAAC;aAChG,CAAC;QACH,CAAC;QACD,OAAO;YACN,IAAI,EAAE,OAAO;YACb,IAAI,EAAE,WAAW;YACjB,UAAU,EAAE,GAAG;YACf,UAAU,EAAE,oBAAoB;YAChC,OAAO,EAAE,CAAC,sEAAsE,CAAC;SACjF,CAAC;IACH,CAAC;IAED,oEAAoE;IACpE,SAAS,aAAa,CAAC,KAAa,EAAwB;QAC3D,qDAAqD;QACrD,IAAI,kBAAkB,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;YACpC,OAAO;gBACN,IAAI,EAAE,WAAW;gBACjB,IAAI,EAAE,mBAAmB;gBACzB,UAAU,EAAE,GAAG;gBACf,UAAU,EAAE,oBAAoB;gBAChC,OAAO,EAAE,CAAC,8DAA8D,CAAC;aACzE,CAAC;QACH,CAAC;QACD,IAAI,gBAAgB,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;YAClC,OAAO;gBACN,IAAI,EAAE,WAAW;gBACjB,IAAI,EAAE,aAAa;gBACnB,UAAU,EAAE,IAAI;gBAChB,UAAU,EAAE,kBAAkB;gBAC9B,OAAO,EAAE,CAAC,iEAAiE,CAAC;aAC5E,CAAC;QACH,CAAC;QACD,IAAI,cAAc,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;YAChC,OAAO;gBACN,IAAI,EAAE,WAAW;gBACjB,IAAI,EAAE,mBAAmB;gBACzB,UAAU,EAAE,IAAI;gBAChB,UAAU,EAAE,4BAA4B;gBACxC,OAAO,EAAE,CAAC,0EAA0E,CAAC;aACrF,CAAC;QACH,CAAC;QACD,IAAI,kBAAkB,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;YACpC,OAAO;gBACN,IAAI,EAAE,WAAW;gBACjB,IAAI,EAAE,mBAAmB;gBACzB,UAAU,EAAE,GAAG;gBACf,UAAU,EAAE,+BAA+B;gBAC3C,OAAO,EAAE,CAAC,kFAAkF,CAAC;aAC7F,CAAC;QACH,CAAC;QACD,IAAI,sBAAsB,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;YACxC,OAAO;gBACN,IAAI,EAAE,WAAW;gBACjB,IAAI,EAAE,aAAa;gBACnB,UAAU,EAAE,GAAG;gBACf,UAAU,EAAE,2BAA2B;gBACvC,OAAO,EAAE,CAAC,8CAA8C,CAAC;aACzD,CAAC;QACH,CAAC;QAED,+DAA+D;QAC/D,IAAI,gBAAgB,CAAC,KAAK,CAAC,EAAE,CAAC;YAC7B,OAAO;gBACN,IAAI,EAAE,QAAQ;gBACd,IAAI,EAAE,WAAW;gBACjB,UAAU,EAAE,IAAI;gBAChB,UAAU,EAAE,qBAAqB;gBACjC,OAAO,EAAE,CAAC,qFAAqF,CAAC;aAChG,CAAC;QACH,CAAC;QACD,IAAI,WAAW,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;YAC7B,OAAO;gBACN,IAAI,EAAE,QAAQ;gBACd,IAAI,EAAE,cAAc;gBACpB,UAAU,EAAE,GAAG;gBACf,UAAU,EAAE,qBAAqB;gBACjC,OAAO,EAAE,CAAC,4CAA4C,CAAC;aACvD,CAAC;QACH,CAAC;QACD,IAAI,kBAAkB,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;YACpC,OAAO;gBACN,IAAI,EAAE,QAAQ;gBACd,IAAI,EAAE,cAAc;gBACpB,UAAU,EAAE,GAAG;gBACf,UAAU,EAAE,oBAAoB;gBAChC,OAAO,EAAE,CAAC,iDAAiD,CAAC;aAC5D,CAAC;QACH,CAAC;QACD,IAAI,YAAY,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;YAC9B,OAAO;gBACN,IAAI,EAAE,QAAQ;gBACd,IAAI,EAAE,cAAc;gBACpB,UAAU,EAAE,IAAI;gBAChB,UAAU,EAAE,uBAAuB;gBACnC,OAAO,EAAE,CAAC,qEAAqE,CAAC;aAChF,CAAC;QACH,CAAC;QAED,OAAO,IAAI,CAAC;IAAA,CACZ;IAED,MAAM,KAAK,GAAG,aAAa,CAAC,IAAI,CAAC,CAAC;IAClC,IAAI,KAAK,EAAE,CAAC;QACX,OAAO,KAAK,CAAC;IACd,CAAC;IAED,uBAAuB;IACvB,OAAO;QACN,IAAI,EAAE,OAAO;QACb,IAAI,EAAE,WAAW;QACjB,UAAU,EAAE,GAAG;QACf,UAAU,EAAE,mBAAmB;QAC/B,OAAO,EAAE,CAAC,uEAAuE,CAAC;KAClF,CAAC;AAAA,CACF;AAED,MAAM,UAAU,yBAAyB,CAAC,MAAc,EAAqB;IAC5E,MAAM,QAAQ,GAAG,wBAAwB,CAAC,MAAM,CAAC,CAAC;IAClD,OAAO,QAAQ,CAAC,IAAI,KAAK,OAAO,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,QAAQ,CAAC;AAAA,CACzD","sourcesContent":["import type { RouteDecision } from \"../autonomy/contracts.ts\";\n\nexport type ModelRouterIntent = \"research\" | \"modify\";\n\nconst EXPLICIT_MODIFY_REQUEST_RE =\n\t/^(?:can you|could you|please|pls|go ahead and|let'?s|i need you to|we need to|you should)\\s+.*\\b(add|apply|build|change|commit|create|delete|edit|fix|generate|implement|install|modify|patch|refactor|remove|rename|replace|run|test|update|write|publish|release|push|deploy|tag|reset|clean|rewrite)\\b/i;\n\nconst READ_ONLY_QUESTION_RE =\n\t/^(?:(?:can you|could you|please|pls|go ahead and|let'?s|i need you to|we need to|you should)\\s+)?(?:how|what|why|when|where|which|who|explain|summarize|compare|describe|list|show|search|find|view|read|locate)\\b/i;\n\nconst RELEASE_PUBLISH_RE = /\\b(publish|release|push|deploy|tag)\\b/i;\nconst SECURITY_AUTH_RE = /\\b(auth|token|credential|credentials|secret|api[-_]key)\\b/i;\nconst DESTRUCTIVE_RE = /\\b(delete|reset|rm\\s+-rf|clean)\\b/i;\n\nconst SELF_MOD_MUTATE_RE =\n\t/\\b(modify|change|write|update|edit|delete|add|remove)\\s+.*\\b(skills|prompts|settings|tools|behavior)\\b|self[-_]modification/i;\nconst ARCHITECTURE_MUTATE_RE = /\\b(rewrite|redesign|change|modify|rearchitect)\\s+.*\\b(architecture|architect)\\b/i;\n\n// Planning floor: plans steer all downstream work, so planning never routes cheap by default.\n// Only the route judge may downgrade a planning prompt back to cheap (explicit trivial verdict).\n// Core terms are always planning; design/architecture words count only with prospective phrasing,\n// so lookups like \"show me the architecture\" stay cheap.\nconst PLANNING_CORE_RE = /\\b(plan|planning|roadmap|strategy)\\b/i;\nconst PLANNING_DESIGN_WORD_RE = /\\b(design|architect\\w*|structure|approach)\\b/i;\nconst PLANNING_PROSPECTIVE_RE =\n\t/\\b(how (?:should|would|do we|can we)|what(?:'s| is) the best|propose|draft|come up with|figure out|decide (?:on|how))\\b/i;\n\nfunction isPlanningPrompt(text: string): boolean {\n\treturn PLANNING_CORE_RE.test(text) || (PLANNING_DESIGN_WORD_RE.test(text) && PLANNING_PROSPECTIVE_RE.test(text));\n}\n\nconst REFACTOR_RE = /\\b(refactor|refactoring)\\b/i;\nconst TEST_VALIDATION_RE = /\\b(test|testing|validation|lint|vitest|jest|run)\\b/i;\nconst IMPLEMENT_RE = /\\b(implement|fix|apply|change|update|create|write|generate|modify|edit|patch|add)\\b/i;\n\nexport function classifyModelRouterRoute(prompt: string): RouteDecision {\n\tconst text = prompt.trim();\n\n\tif (text.length === 0) {\n\t\treturn {\n\t\t\ttier: \"cheap\",\n\t\t\trisk: \"read-only\",\n\t\t\tconfidence: 0.1,\n\t\t\treasonCode: \"empty_prompt\",\n\t\t\treasons: [\"Empty or whitespace prompt\"],\n\t\t};\n\t}\n\n\t// 1. Explicit read-only questions/lookups dominate (unless prefixed by explicit mutation verb).\n\t// Planning-shaped questions are the exception: a plan steers expensive downstream work, so the\n\t// floor is medium even when phrased as a question.\n\tif (READ_ONLY_QUESTION_RE.test(text) && !EXPLICIT_MODIFY_REQUEST_RE.test(text)) {\n\t\tif (isPlanningPrompt(text)) {\n\t\t\treturn {\n\t\t\t\ttier: \"medium\",\n\t\t\t\trisk: \"read-only\",\n\t\t\t\tconfidence: 0.75,\n\t\t\t\treasonCode: \"planning_min_medium\",\n\t\t\t\treasons: [\"Planning/design prompts never route cheap by default; a judge may deem them trivial\"],\n\t\t\t};\n\t\t}\n\t\treturn {\n\t\t\ttier: \"cheap\",\n\t\t\trisk: \"read-only\",\n\t\t\tconfidence: 0.9,\n\t\t\treasonCode: \"read_only_question\",\n\t\t\treasons: [\"Prompt asks a question or requests an explanation, search, or lookup\"],\n\t\t};\n\t}\n\n\t// Helper function to match patterns and return appropriate decision\n\tfunction matchKeywords(input: string): RouteDecision | null {\n\t\t// A. High-risk / approval-required/expensive signals\n\t\tif (RELEASE_PUBLISH_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"expensive\",\n\t\t\t\trisk: \"approval-required\",\n\t\t\t\tconfidence: 0.9,\n\t\t\t\treasonCode: \"release_or_publish\",\n\t\t\t\treasons: [\"Prompt mentions publishing, releasing, pushing, or deploying\"],\n\t\t\t};\n\t\t}\n\t\tif (SECURITY_AUTH_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"expensive\",\n\t\t\t\trisk: \"high-impact\",\n\t\t\t\tconfidence: 0.95,\n\t\t\t\treasonCode: \"security_or_auth\",\n\t\t\t\treasons: [\"Prompt mentions credentials, authentication, tokens, or secrets\"],\n\t\t\t};\n\t\t}\n\t\tif (DESTRUCTIVE_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"expensive\",\n\t\t\t\trisk: \"approval-required\",\n\t\t\t\tconfidence: 0.85,\n\t\t\t\treasonCode: \"destructive_or_git_history\",\n\t\t\t\treasons: [\"Prompt mentions deleting, resetting, cleaning, or destructive operations\"],\n\t\t\t};\n\t\t}\n\t\tif (SELF_MOD_MUTATE_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"expensive\",\n\t\t\t\trisk: \"approval-required\",\n\t\t\t\tconfidence: 0.9,\n\t\t\t\treasonCode: \"settings_or_self_modification\",\n\t\t\t\treasons: [\"Prompt mentions modifying skills, prompts, settings, tools, or self-modification\"],\n\t\t\t};\n\t\t}\n\t\tif (ARCHITECTURE_MUTATE_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"expensive\",\n\t\t\t\trisk: \"high-impact\",\n\t\t\t\tconfidence: 0.9,\n\t\t\t\treasonCode: \"architecture_or_ambiguous\",\n\t\t\t\treasons: [\"Prompt mentions core architecture or rewrite\"],\n\t\t\t};\n\t\t}\n\n\t\t// B. Explicit implementation/scoped-write signals route medium\n\t\tif (isPlanningPrompt(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"medium\",\n\t\t\t\trisk: \"read-only\",\n\t\t\t\tconfidence: 0.75,\n\t\t\t\treasonCode: \"planning_min_medium\",\n\t\t\t\treasons: [\"Planning/design prompts never route cheap by default; a judge may deem them trivial\"],\n\t\t\t};\n\t\t}\n\t\tif (REFACTOR_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"medium\",\n\t\t\t\trisk: \"scoped-write\",\n\t\t\t\tconfidence: 0.8,\n\t\t\t\treasonCode: \"mechanical_refactor\",\n\t\t\t\treasons: [\"Prompt mentions refactoring code structure\"],\n\t\t\t};\n\t\t}\n\t\tif (TEST_VALIDATION_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"medium\",\n\t\t\t\trisk: \"scoped-write\",\n\t\t\t\tconfidence: 0.8,\n\t\t\t\treasonCode: \"test_or_validation\",\n\t\t\t\treasons: [\"Prompt mentions testing, validation, or linting\"],\n\t\t\t};\n\t\t}\n\t\tif (IMPLEMENT_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"medium\",\n\t\t\t\trisk: \"scoped-write\",\n\t\t\t\tconfidence: 0.85,\n\t\t\t\treasonCode: \"normal_implementation\",\n\t\t\t\treasons: [\"Prompt mentions implementing, updating, creating, or modifying code\"],\n\t\t\t};\n\t\t}\n\n\t\treturn null;\n\t}\n\n\tconst match = matchKeywords(text);\n\tif (match) {\n\t\treturn match;\n\t}\n\n\t// 4. Default fallbacks\n\treturn {\n\t\ttier: \"cheap\",\n\t\trisk: \"read-only\",\n\t\tconfidence: 0.5,\n\t\treasonCode: \"default_read_only\",\n\t\treasons: [\"No explicit implementation, destructive, or release patterns detected\"],\n\t};\n}\n\nexport function classifyModelRouterIntent(prompt: string): ModelRouterIntent {\n\tconst decision = classifyModelRouterRoute(prompt);\n\treturn decision.tier === \"cheap\" ? \"research\" : \"modify\";\n}\n"]}
+{"version":3,"file":"intent-classifier.js","sourceRoot":"","sources":["../../../src/core/model-router/intent-classifier.ts"],"names":[],"mappings":"AAIA,MAAM,0BAA0B,GAC/B,4SAA4S,CAAC;AAE9S,MAAM,qBAAqB,GAC1B,qNAAqN,CAAC;AAEvN,MAAM,kBAAkB,GAAG,wCAAwC,CAAC;AACpE,MAAM,gBAAgB,GAAG,4DAA4D,CAAC;AACtF,MAAM,cAAc,GAAG,oCAAoC,CAAC;AAE5D,MAAM,kBAAkB,GACvB,8HAA8H,CAAC;AAChI,MAAM,sBAAsB,GAAG,kFAAkF,CAAC;AAElH,8FAA8F;AAC9F,iGAAiG;AACjG,kGAAkG;AAClG,yDAAyD;AACzD,MAAM,gBAAgB,GAAG,uCAAuC,CAAC;AACjE,MAAM,uBAAuB,GAAG,+CAA+C,CAAC;AAChF,MAAM,uBAAuB,GAC5B,6IAA6I,CAAC;AAE/I,SAAS,gBAAgB,CAAC,IAAY,EAAW;IAChD,OAAO,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,uBAAuB,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,uBAAuB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;AAAA,CACjH;AAED,MAAM,WAAW,GAAG,6BAA6B,CAAC;AAClD,MAAM,kBAAkB,GAAG,qDAAqD,CAAC;AACjF,MAAM,YAAY,GAAG,sFAAsF,CAAC;AAE5G,MAAM,UAAU,wBAAwB,CAAC,MAAc,EAAiB;IACvE,MAAM,IAAI,GAAG,MAAM,CAAC,IAAI,EAAE,CAAC;IAE3B,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACvB,OAAO;YACN,IAAI,EAAE,OAAO;YACb,IAAI,EAAE,WAAW;YACjB,UAAU,EAAE,GAAG;YACf,UAAU,EAAE,cAAc;YAC1B,OAAO,EAAE,CAAC,4BAA4B,CAAC;SACvC,CAAC;IACH,CAAC;IAED,gGAAgG;IAChG,+FAA+F;IAC/F,mDAAmD;IACnD,IAAI,qBAAqB,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,0BAA0B,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;QAChF,IAAI,gBAAgB,CAAC,IAAI,CAAC,EAAE,CAAC;YAC5B,OAAO;gBACN,IAAI,EAAE,QAAQ;gBACd,IAAI,EAAE,WAAW;gBACjB,UAAU,EAAE,IAAI;gBAChB,UAAU,EAAE,qBAAqB;gBACjC,OAAO,EAAE,CAAC,qFAAqF,CAAC;aAChG,CAAC;QACH,CAAC;QACD,OAAO;YACN,IAAI,EAAE,OAAO;YACb,IAAI,EAAE,WAAW;YACjB,UAAU,EAAE,GAAG;YACf,UAAU,EAAE,oBAAoB;YAChC,OAAO,EAAE,CAAC,sEAAsE,CAAC;SACjF,CAAC;IACH,CAAC;IAED,oEAAoE;IACpE,SAAS,aAAa,CAAC,KAAa,EAAwB;QAC3D,qDAAqD;QACrD,IAAI,kBAAkB,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;YACpC,OAAO;gBACN,IAAI,EAAE,WAAW;gBACjB,IAAI,EAAE,mBAAmB;gBACzB,UAAU,EAAE,GAAG;gBACf,UAAU,EAAE,oBAAoB;gBAChC,OAAO,EAAE,CAAC,8DAA8D,CAAC;aACzE,CAAC;QACH,CAAC;QACD,IAAI,gBAAgB,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;YAClC,OAAO;gBACN,IAAI,EAAE,WAAW;gBACjB,IAAI,EAAE,aAAa;gBACnB,UAAU,EAAE,IAAI;gBAChB,UAAU,EAAE,kBAAkB;gBAC9B,OAAO,EAAE,CAAC,iEAAiE,CAAC;aAC5E,CAAC;QACH,CAAC;QACD,IAAI,cAAc,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;YAChC,OAAO;gBACN,IAAI,EAAE,WAAW;gBACjB,IAAI,EAAE,mBAAmB;gBACzB,UAAU,EAAE,IAAI;gBAChB,UAAU,EAAE,4BAA4B;gBACxC,OAAO,EAAE,CAAC,0EAA0E,CAAC;aACrF,CAAC;QACH,CAAC;QACD,IAAI,kBAAkB,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;YACpC,OAAO;gBACN,IAAI,EAAE,WAAW;gBACjB,IAAI,EAAE,mBAAmB;gBACzB,UAAU,EAAE,GAAG;gBACf,UAAU,EAAE,+BAA+B;gBAC3C,OAAO,EAAE,CAAC,kFAAkF,CAAC;aAC7F,CAAC;QACH,CAAC;QACD,IAAI,sBAAsB,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;YACxC,OAAO;gBACN,IAAI,EAAE,WAAW;gBACjB,IAAI,EAAE,aAAa;gBACnB,UAAU,EAAE,GAAG;gBACf,UAAU,EAAE,2BAA2B;gBACvC,OAAO,EAAE,CAAC,8CAA8C,CAAC;aACzD,CAAC;QACH,CAAC;QAED,+DAA+D;QAC/D,IAAI,gBAAgB,CAAC,KAAK,CAAC,EAAE,CAAC;YAC7B,OAAO;gBACN,IAAI,EAAE,QAAQ;gBACd,IAAI,EAAE,WAAW;gBACjB,UAAU,EAAE,IAAI;gBAChB,UAAU,EAAE,qBAAqB;gBACjC,OAAO,EAAE,CAAC,qFAAqF,CAAC;aAChG,CAAC;QACH,CAAC;QACD,IAAI,WAAW,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;YAC7B,OAAO;gBACN,IAAI,EAAE,QAAQ;gBACd,IAAI,EAAE,cAAc;gBACpB,UAAU,EAAE,GAAG;gBACf,UAAU,EAAE,qBAAqB;gBACjC,OAAO,EAAE,CAAC,4CAA4C,CAAC;aACvD,CAAC;QACH,CAAC;QACD,IAAI,kBAAkB,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;YACpC,OAAO;gBACN,IAAI,EAAE,QAAQ;gBACd,IAAI,EAAE,cAAc;gBACpB,UAAU,EAAE,GAAG;gBACf,UAAU,EAAE,oBAAoB;gBAChC,OAAO,EAAE,CAAC,iDAAiD,CAAC;aAC5D,CAAC;QACH,CAAC;QACD,IAAI,YAAY,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;YAC9B,OAAO;gBACN,IAAI,EAAE,QAAQ;gBACd,IAAI,EAAE,cAAc;gBACpB,UAAU,EAAE,IAAI;gBAChB,UAAU,EAAE,uBAAuB;gBACnC,OAAO,EAAE,CAAC,qEAAqE,CAAC;aAChF,CAAC;QACH,CAAC;QAED,OAAO,IAAI,CAAC;IAAA,CACZ;IAED,MAAM,KAAK,GAAG,aAAa,CAAC,IAAI,CAAC,CAAC;IAClC,IAAI,KAAK,EAAE,CAAC;QACX,OAAO,KAAK,CAAC;IACd,CAAC;IAED,uBAAuB;IACvB,OAAO;QACN,IAAI,EAAE,OAAO;QACb,IAAI,EAAE,WAAW;QACjB,UAAU,EAAE,GAAG;QACf,UAAU,EAAE,mBAAmB;QAC/B,OAAO,EAAE,CAAC,uEAAuE,CAAC;KAClF,CAAC;AAAA,CACF;AAED,MAAM,UAAU,yBAAyB,CAAC,MAAc,EAAqB;IAC5E,MAAM,QAAQ,GAAG,wBAAwB,CAAC,MAAM,CAAC,CAAC;IAClD,OAAO,QAAQ,CAAC,IAAI,KAAK,OAAO,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,QAAQ,CAAC;AAAA,CACzD","sourcesContent":["import type { RouteDecision } from \"../autonomy/contracts.ts\";\n\nexport type ModelRouterIntent = \"research\" | \"modify\";\n\nconst EXPLICIT_MODIFY_REQUEST_RE =\n\t/^(?:can you|could you|please|pls|go ahead and|let'?s|i need you to|we need to|you should)\\s+.*\\b(add|apply|build|change|commit|create|delete|edit|fix|generate|implement|install|modify|patch|refactor|remove|rename|replace|run|test|update|write|publish|release|push|deploy|tag|reset|clean|rewrite)\\b/i;\n\nconst READ_ONLY_QUESTION_RE =\n\t/^(?:(?:can you|could you|please|pls|go ahead and|let'?s|i need you to|we need to|you should)\\s+)?(?:how|what|why|when|where|which|who|explain|summarize|compare|describe|list|show|search|find|view|read|locate)\\b/i;\n\nconst RELEASE_PUBLISH_RE = /\\b(publish|release|push|deploy|tag)\\b/i;\nconst SECURITY_AUTH_RE = /\\b(auth|token|credential|credentials|secret|api[-_]key)\\b/i;\nconst DESTRUCTIVE_RE = /\\b(delete|reset|rm\\s+-rf|clean)\\b/i;\n\nconst SELF_MOD_MUTATE_RE =\n\t/\\b(modify|change|write|update|edit|delete|add|remove)\\s+.*\\b(skills|prompts|settings|tools|behavior)\\b|self[-_]modification/i;\nconst ARCHITECTURE_MUTATE_RE = /\\b(rewrite|redesign|change|modify|rearchitect)\\s+.*\\b(architecture|architect)\\b/i;\n\n// Planning floor: plans steer all downstream work, so planning never routes cheap by default.\n// Only the route judge may downgrade a planning prompt back to cheap (explicit trivial verdict).\n// Core terms are always planning; design/architecture words count only with prospective phrasing,\n// so lookups like \"show me the architecture\" stay cheap.\nconst PLANNING_CORE_RE = /\\b(plan|planning|roadmap|strategy)\\b/i;\nconst PLANNING_DESIGN_WORD_RE = /\\b(design|architect\\w*|structure|approach)\\b/i;\nconst PLANNING_PROSPECTIVE_RE =\n\t/\\b(how (?:should|would|do we|can we)|what(?:'s| is) the (?:best|cleanest|right)|propose|draft|come up with|figure out|decide (?:on|how))\\b/i;\n\nfunction isPlanningPrompt(text: string): boolean {\n\treturn PLANNING_CORE_RE.test(text) || (PLANNING_DESIGN_WORD_RE.test(text) && PLANNING_PROSPECTIVE_RE.test(text));\n}\n\nconst REFACTOR_RE = /\\b(refactor|refactoring)\\b/i;\nconst TEST_VALIDATION_RE = /\\b(test|testing|validation|lint|vitest|jest|run)\\b/i;\nconst IMPLEMENT_RE = /\\b(implement|fix|apply|change|update|create|write|generate|modify|edit|patch|add)\\b/i;\n\nexport function classifyModelRouterRoute(prompt: string): RouteDecision {\n\tconst text = prompt.trim();\n\n\tif (text.length === 0) {\n\t\treturn {\n\t\t\ttier: \"cheap\",\n\t\t\trisk: \"read-only\",\n\t\t\tconfidence: 0.1,\n\t\t\treasonCode: \"empty_prompt\",\n\t\t\treasons: [\"Empty or whitespace prompt\"],\n\t\t};\n\t}\n\n\t// 1. Explicit read-only questions/lookups dominate (unless prefixed by explicit mutation verb).\n\t// Planning-shaped questions are the exception: a plan steers expensive downstream work, so the\n\t// floor is medium even when phrased as a question.\n\tif (READ_ONLY_QUESTION_RE.test(text) && !EXPLICIT_MODIFY_REQUEST_RE.test(text)) {\n\t\tif (isPlanningPrompt(text)) {\n\t\t\treturn {\n\t\t\t\ttier: \"medium\",\n\t\t\t\trisk: \"read-only\",\n\t\t\t\tconfidence: 0.75,\n\t\t\t\treasonCode: \"planning_min_medium\",\n\t\t\t\treasons: [\"Planning/design prompts never route cheap by default; a judge may deem them trivial\"],\n\t\t\t};\n\t\t}\n\t\treturn {\n\t\t\ttier: \"cheap\",\n\t\t\trisk: \"read-only\",\n\t\t\tconfidence: 0.9,\n\t\t\treasonCode: \"read_only_question\",\n\t\t\treasons: [\"Prompt asks a question or requests an explanation, search, or lookup\"],\n\t\t};\n\t}\n\n\t// Helper function to match patterns and return appropriate decision\n\tfunction matchKeywords(input: string): RouteDecision | null {\n\t\t// A. High-risk / approval-required/expensive signals\n\t\tif (RELEASE_PUBLISH_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"expensive\",\n\t\t\t\trisk: \"approval-required\",\n\t\t\t\tconfidence: 0.9,\n\t\t\t\treasonCode: \"release_or_publish\",\n\t\t\t\treasons: [\"Prompt mentions publishing, releasing, pushing, or deploying\"],\n\t\t\t};\n\t\t}\n\t\tif (SECURITY_AUTH_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"expensive\",\n\t\t\t\trisk: \"high-impact\",\n\t\t\t\tconfidence: 0.95,\n\t\t\t\treasonCode: \"security_or_auth\",\n\t\t\t\treasons: [\"Prompt mentions credentials, authentication, tokens, or secrets\"],\n\t\t\t};\n\t\t}\n\t\tif (DESTRUCTIVE_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"expensive\",\n\t\t\t\trisk: \"approval-required\",\n\t\t\t\tconfidence: 0.85,\n\t\t\t\treasonCode: \"destructive_or_git_history\",\n\t\t\t\treasons: [\"Prompt mentions deleting, resetting, cleaning, or destructive operations\"],\n\t\t\t};\n\t\t}\n\t\tif (SELF_MOD_MUTATE_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"expensive\",\n\t\t\t\trisk: \"approval-required\",\n\t\t\t\tconfidence: 0.9,\n\t\t\t\treasonCode: \"settings_or_self_modification\",\n\t\t\t\treasons: [\"Prompt mentions modifying skills, prompts, settings, tools, or self-modification\"],\n\t\t\t};\n\t\t}\n\t\tif (ARCHITECTURE_MUTATE_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"expensive\",\n\t\t\t\trisk: \"high-impact\",\n\t\t\t\tconfidence: 0.9,\n\t\t\t\treasonCode: \"architecture_or_ambiguous\",\n\t\t\t\treasons: [\"Prompt mentions core architecture or rewrite\"],\n\t\t\t};\n\t\t}\n\n\t\t// B. Explicit implementation/scoped-write signals route medium\n\t\tif (isPlanningPrompt(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"medium\",\n\t\t\t\trisk: \"read-only\",\n\t\t\t\tconfidence: 0.75,\n\t\t\t\treasonCode: \"planning_min_medium\",\n\t\t\t\treasons: [\"Planning/design prompts never route cheap by default; a judge may deem them trivial\"],\n\t\t\t};\n\t\t}\n\t\tif (REFACTOR_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"medium\",\n\t\t\t\trisk: \"scoped-write\",\n\t\t\t\tconfidence: 0.8,\n\t\t\t\treasonCode: \"mechanical_refactor\",\n\t\t\t\treasons: [\"Prompt mentions refactoring code structure\"],\n\t\t\t};\n\t\t}\n\t\tif (TEST_VALIDATION_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"medium\",\n\t\t\t\trisk: \"scoped-write\",\n\t\t\t\tconfidence: 0.8,\n\t\t\t\treasonCode: \"test_or_validation\",\n\t\t\t\treasons: [\"Prompt mentions testing, validation, or linting\"],\n\t\t\t};\n\t\t}\n\t\tif (IMPLEMENT_RE.test(input)) {\n\t\t\treturn {\n\t\t\t\ttier: \"medium\",\n\t\t\t\trisk: \"scoped-write\",\n\t\t\t\tconfidence: 0.85,\n\t\t\t\treasonCode: \"normal_implementation\",\n\t\t\t\treasons: [\"Prompt mentions implementing, updating, creating, or modifying code\"],\n\t\t\t};\n\t\t}\n\n\t\treturn null;\n\t}\n\n\tconst match = matchKeywords(text);\n\tif (match) {\n\t\treturn match;\n\t}\n\n\t// 4. Default fallbacks\n\treturn {\n\t\ttier: \"cheap\",\n\t\trisk: \"read-only\",\n\t\tconfidence: 0.5,\n\t\treasonCode: \"default_read_only\",\n\t\treasons: [\"No explicit implementation, destructive, or release patterns detected\"],\n\t};\n}\n\nexport function classifyModelRouterIntent(prompt: string): ModelRouterIntent {\n\tconst decision = classifyModelRouterRoute(prompt);\n\treturn decision.tier === \"cheap\" ? \"research\" : \"modify\";\n}\n"]}

package/dist/core/model-router/route-judge.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"route-judge.d.ts","sourceRoot":"","sources":["../../../src/core/model-router/route-judge.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,0BAA0B,CAAC;AAE9D;;;;;;;GAOG;AAEH,uGAAqG;AACrG,eAAO,MAAM,yBAAyB,QAS1B,CAAC;AAEb,eAAO,MAAM,6BAA6B,MAAM,CAAC;AACjD,eAAO,MAAM,6BAA6B,QAAS,CAAC;AAEpD,MAAM,WAAW,iBAAiB;IACjC,IAAI,EAAE,OAAO,GAAG,QAAQ,GAAG,WAAW,CAAC;IACvC,IAAI,EAAE,aAAa,CAAC,MAAM,CAAC,CAAC;IAC5B,OAAO,EAAE,OAAO,CAAC;IACjB,MAAM,EAAE,MAAM,CAAC;CACf;AAED,wBAAgB,yBAAyB,CAAC,IAAI,EAAE;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,aAAa,CAAA;CAAE,GAAG,MAAM,CAMnG;AAKD,wBAAgB,sBAAsB,CAAC,IAAI,EAAE,MAAM,GAAG,iBAAiB,GAAG,SAAS,CAiClF;AAED,uFAAuF;AACvF,wBAAgB,sBAAsB,CAAC,QAAQ,EAAE,aAAa,EAAE,OAAO,EAAE,iBAAiB,GAAG,aAAa,CAczG;AAED,MAAM,WAAW,mBAAmB;IACnC,QAAQ,EAAE,aAAa,CAAC;IACxB,OAAO,CAAC,EAAE,iBAAiB,CAAC;IAC5B,qEAAqE;IACrE,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,OAAO,EAAE,MAAM,CAAC;CAChB;AAED;;;;GAIG;AACH,wBAAsB,aAAa,CAAC,IAAI,EAAE;IACzC,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,aAAa,CAAC;IACxB,QAAQ,EAAE,CAAC,KAAK,EAAE;QAAE,YAAY,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,CAAC;QAAC,MAAM,CAAC,EAAE,WAAW,CAAA;KAAE,KAAK,OAAO,CAAC;QAChG,IAAI,EAAE,MAAM,CAAC;QACb,OAAO,EAAE,MAAM,CAAC;QAChB,UAAU,EAAE,MAAM,CAAC;KACnB,CAAC,CAAC;IACH,MAAM,CAAC,EAAE,WAAW,CAAC;IACrB,cAAc,CAAC,EAAE,MAAM,CAAC;CACxB,GAAG,OAAO,CAAC,mBAAmB,CAAC,CAyC/B","sourcesContent":["import { runBoundedCompletion } from \"../autonomy/bounded-completion.ts\";\nimport type { RouteDecision } from \"../autonomy/contracts.ts\";\n\n/**\n * Routing-only judge: a bounded, tool-less completion (default: the medium model) that decides the\n * final cheap/medium/expensive tier for a user prompt, refining the regex classifier's baseline.\n * Core rule: planning is never cheap unless the judge explicitly deems the task trivial. The judge\n * proposes; the existing pipeline (model resolution, auth, escalation, gates) still decides.\n * Failure is honest: unparseable/timeout/unavailable falls back to the baseline with a visible\n * reasonCode, never silently.\n */\n\n/** Static across calls — callers pass cacheRetention \"short\" so only the variable tail is billed. */\nexport const ROUTE_JUDGE_SYSTEM_PROMPT = [\n\t\"You are a routing judge for a coding agent. You only route; you never answer the task.\",\n\t\"Pick which model tier should handle the user prompt:\",\n\t'- \"cheap\": trivial, mechanical, read-only lookups only.',\n\t'- \"medium\": normal implementation, scoped edits, tests, and NON-trivial planning/design.',\n\t'- \"expensive\": architecture, ambiguity, security/auth, destructive or release operations, high-impact changes.',\n\t\"Planning, design, and strategy prompts are NEVER cheap unless the task is genuinely trivial.\",\n\t\"Respond with STRICT JSON only - no prose:\",\n\t'{\"tier\":\"cheap\"|\"medium\"|\"expensive\",\"risk\":\"read-only\"|\"scoped-write\"|\"high-impact\"|\"approval-required\",\"trivial\":true|false,\"reason\":\"<short reason>\"}',\n].join(\"\\n\");\n\nexport const ROUTE_JUDGE_MAX_OUTPUT_TOKENS = 128;\nexport const ROUTE_JUDGE_MAX_WALL_CLOCK_MS = 10_000;\n\nexport interface RouteJudgeVerdict {\n\ttier: \"cheap\" | \"medium\" | \"expensive\";\n\trisk: RouteDecision[\"risk\"];\n\ttrivial: boolean;\n\treason: string;\n}\n\nexport function buildRouteJudgeUserPrompt(args: { prompt: string; baseline: RouteDecision }): string {\n\treturn [\n\t\t`Baseline (regex) verdict: tier=${args.baseline.tier}, risk=${args.baseline.risk}, reason=${args.baseline.reasonCode}.`,\n\t\t\"User prompt:\",\n\t\targs.prompt.slice(0, 4000),\n\t].join(\"\\n\");\n}\n\nconst JUDGE_TIERS: readonly string[] = [\"cheap\", \"medium\", \"expensive\"];\nconst JUDGE_RISKS: readonly string[] = [\"read-only\", \"scoped-write\", \"high-impact\", \"approval-required\"];\n\nexport function parseRouteJudgeVerdict(text: string): RouteJudgeVerdict | undefined {\n\tconst trimmed = text.trim();\n\tconst candidates: string[] = [trimmed];\n\tconst fenced = /```(?:json)?\\s*([\\s\\S]*?)```/.exec(trimmed);\n\tif (fenced?.[1]) candidates.push(fenced[1].trim());\n\tconst start = trimmed.indexOf(\"{\");\n\tconst end = trimmed.lastIndexOf(\"}\");\n\tif (start >= 0 && end > start) candidates.push(trimmed.slice(start, end + 1));\n\n\tfor (const candidate of candidates) {\n\t\tlet parsed: unknown;\n\t\ttry {\n\t\t\tparsed = JSON.parse(candidate);\n\t\t} catch {\n\t\t\tcontinue;\n\t\t}\n\t\tif (!parsed || typeof parsed !== \"object\" || Array.isArray(parsed)) continue;\n\t\tconst record = parsed as Record<string, unknown>;\n\t\t// The judge may never select the learning tier or anything outside the three foreground tiers.\n\t\tif (typeof record.tier !== \"string\" || !JUDGE_TIERS.includes(record.tier)) continue;\n\t\tconst risk =\n\t\t\ttypeof record.risk === \"string\" && JUDGE_RISKS.includes(record.risk)\n\t\t\t\t? (record.risk as RouteDecision[\"risk\"])\n\t\t\t\t: undefined;\n\t\tif (!risk) continue;\n\t\treturn {\n\t\t\ttier: record.tier as RouteJudgeVerdict[\"tier\"],\n\t\t\trisk,\n\t\t\ttrivial: record.trivial === true,\n\t\t\treason: typeof record.reason === \"string\" ? record.reason.slice(0, 200) : \"\",\n\t\t};\n\t}\n\treturn undefined;\n}\n\n/** Merge a judge verdict into the baseline decision (pure; never returns learning). */\nexport function applyRouteJudgeVerdict(baseline: RouteDecision, verdict: RouteJudgeVerdict): RouteDecision {\n\t// Enforce the core rule in code, not just in the judge prompt: downgrading a non-cheap\n\t// baseline to cheap requires an EXPLICIT trivial verdict. An untrusted judge saying\n\t// {tier:\"cheap\", trivial:false} for an elevated prompt keeps the baseline tier.\n\tconst tier =\n\t\tverdict.tier === \"cheap\" && baseline.tier !== \"cheap\" && !verdict.trivial ? baseline.tier : verdict.tier;\n\treturn {\n\t\t...baseline,\n\t\ttier,\n\t\trisk: verdict.risk,\n\t\tconfidence: Math.max(baseline.confidence, 0.75),\n\t\treasonCode: `judge_${tier}${verdict.trivial ? \"_trivial\" : \"\"}`,\n\t\treasons: [...baseline.reasons, `Route judge: ${verdict.reason || \"no reason given\"}`],\n\t};\n}\n\nexport interface RouteJudgeRunResult {\n\tdecision: RouteDecision;\n\tverdict?: RouteJudgeVerdict;\n\t/** Set when the judge could not decide and the baseline was kept. */\n\tfallbackReason?: string;\n\tcostUsd: number;\n}\n\n/**\n * Run the judge over a baseline decision. The completion executor is injected (production:\n * AgentSession.runIsolatedCompletion on the judge model). Never throws; every failure keeps the\n * baseline with a visible fallbackReason.\n */\nexport async function runRouteJudge(args: {\n\tprompt: string;\n\tbaseline: RouteDecision;\n\tcomplete: (input: { systemPrompt: string; userPrompt: string; signal?: AbortSignal }) => Promise<{\n\t\ttext: string;\n\t\tcostUsd: number;\n\t\tstopReason: string;\n\t}>;\n\tsignal?: AbortSignal;\n\tmaxWallClockMs?: number;\n}): Promise<RouteJudgeRunResult> {\n\tconst bounded = await runBoundedCompletion({\n\t\tmaxWallClockMs: args.maxWallClockMs ?? ROUTE_JUDGE_MAX_WALL_CLOCK_MS,\n\t\tsignal: args.signal,\n\t\texecute: (signal) =>\n\t\t\targs.complete({\n\t\t\t\tsystemPrompt: ROUTE_JUDGE_SYSTEM_PROMPT,\n\t\t\t\tuserPrompt: buildRouteJudgeUserPrompt({ prompt: args.prompt, baseline: args.baseline }),\n\t\t\t\tsignal,\n\t\t\t}),\n\t});\n\tconst costUsd = bounded.completion?.costUsd ?? 0;\n\n\tif (bounded.failure || !bounded.completion) {\n\t\treturn {\n\t\t\tdecision: {\n\t\t\t\t...args.baseline,\n\t\t\t\treasons: [...args.baseline.reasons, \"Route judge unavailable; baseline kept\"],\n\t\t\t},\n\t\t\tfallbackReason: bounded.failure ? `judge_${bounded.failure.reasonCode}` : \"judge_unavailable_fallback\",\n\t\t\tcostUsd,\n\t\t};\n\t}\n\tif (bounded.completion.stopReason === \"error\" || bounded.completion.stopReason === \"aborted\") {\n\t\treturn {\n\t\t\tdecision: { ...args.baseline, reasons: [...args.baseline.reasons, \"Route judge errored; baseline kept\"] },\n\t\t\tfallbackReason: \"judge_model_error\",\n\t\t\tcostUsd,\n\t\t};\n\t}\n\n\tconst verdict = parseRouteJudgeVerdict(bounded.completion.text);\n\tif (!verdict) {\n\t\treturn {\n\t\t\tdecision: { ...args.baseline, reasons: [...args.baseline.reasons, \"Route judge unparseable; baseline kept\"] },\n\t\t\tfallbackReason: \"judge_unparseable_fallback\",\n\t\t\tcostUsd,\n\t\t};\n\t}\n\n\treturn { decision: applyRouteJudgeVerdict(args.baseline, verdict), verdict, costUsd };\n}\n"]}
+{"version":3,"file":"route-judge.d.ts","sourceRoot":"","sources":["../../../src/core/model-router/route-judge.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,0BAA0B,CAAC;AAE9D;;;;;;;GAOG;AAEH,uGAAqG;AACrG,eAAO,MAAM,yBAAyB,QAS1B,CAAC;AAEb,eAAO,MAAM,6BAA6B,MAAM,CAAC;AACjD,eAAO,MAAM,6BAA6B,QAAS,CAAC;AAEpD,MAAM,WAAW,iBAAiB;IACjC,IAAI,EAAE,OAAO,GAAG,QAAQ,GAAG,WAAW,CAAC;IACvC,IAAI,EAAE,aAAa,CAAC,MAAM,CAAC,CAAC;IAC5B,OAAO,EAAE,OAAO,CAAC;IACjB,MAAM,EAAE,MAAM,CAAC;CACf;AAED,wBAAgB,yBAAyB,CAAC,IAAI,EAAE;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,aAAa,CAAA;CAAE,GAAG,MAAM,CAMnG;AAKD,wBAAgB,sBAAsB,CAAC,IAAI,EAAE,MAAM,GAAG,iBAAiB,GAAG,SAAS,CAiClF;AAED,uFAAuF;AACvF,wBAAgB,sBAAsB,CAAC,QAAQ,EAAE,aAAa,EAAE,OAAO,EAAE,iBAAiB,GAAG,aAAa,CAczG;AAED,MAAM,WAAW,mBAAmB;IACnC,QAAQ,EAAE,aAAa,CAAC;IACxB,OAAO,CAAC,EAAE,iBAAiB,CAAC;IAC5B,qEAAqE;IACrE,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,OAAO,EAAE,MAAM,CAAC;CAChB;AAUD;;;;GAIG;AACH,wBAAsB,aAAa,CAAC,IAAI,EAAE;IACzC,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,aAAa,CAAC;IACxB,QAAQ,EAAE,CAAC,KAAK,EAAE;QAAE,YAAY,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,CAAC;QAAC,MAAM,CAAC,EAAE,WAAW,CAAA;KAAE,KAAK,OAAO,CAAC;QAChG,IAAI,EAAE,MAAM,CAAC;QACb,OAAO,EAAE,MAAM,CAAC;QAChB,UAAU,EAAE,MAAM,CAAC;KACnB,CAAC,CAAC;IACH,MAAM,CAAC,EAAE,WAAW,CAAC;IACrB,cAAc,CAAC,EAAE,MAAM,CAAC;CACxB,GAAG,OAAO,CAAC,mBAAmB,CAAC,CAsC/B","sourcesContent":["import { runBoundedCompletion } from \"../autonomy/bounded-completion.ts\";\nimport type { RouteDecision } from \"../autonomy/contracts.ts\";\n\n/**\n * Routing-only judge: a bounded, tool-less completion (default: the medium model) that decides the\n * final cheap/medium/expensive tier for a user prompt, refining the regex classifier's baseline.\n * Core rule: planning is never cheap unless the judge explicitly deems the task trivial. The judge\n * proposes; the existing pipeline (model resolution, auth, escalation, gates) still decides.\n * Failure is honest: unparseable/timeout/unavailable falls back to the baseline with a visible\n * reasonCode, never silently.\n */\n\n/** Static across calls — callers pass cacheRetention \"short\" so only the variable tail is billed. */\nexport const ROUTE_JUDGE_SYSTEM_PROMPT = [\n\t\"You are a routing judge for a coding agent. You only route; you never answer the task.\",\n\t\"Pick which model tier should handle the user prompt:\",\n\t'- \"cheap\": trivial, mechanical, read-only lookups only.',\n\t'- \"medium\": normal implementation, scoped edits, tests, and NON-trivial planning/design.',\n\t'- \"expensive\": architecture, ambiguity, security/auth, destructive or release operations, high-impact changes.',\n\t\"Planning, design, and strategy prompts are NEVER cheap unless the task is genuinely trivial.\",\n\t\"Respond with STRICT JSON only - no prose:\",\n\t'{\"tier\":\"cheap\"|\"medium\"|\"expensive\",\"risk\":\"read-only\"|\"scoped-write\"|\"high-impact\"|\"approval-required\",\"trivial\":true|false,\"reason\":\"<short reason>\"}',\n].join(\"\\n\");\n\nexport const ROUTE_JUDGE_MAX_OUTPUT_TOKENS = 128;\nexport const ROUTE_JUDGE_MAX_WALL_CLOCK_MS = 10_000;\n\nexport interface RouteJudgeVerdict {\n\ttier: \"cheap\" | \"medium\" | \"expensive\";\n\trisk: RouteDecision[\"risk\"];\n\ttrivial: boolean;\n\treason: string;\n}\n\nexport function buildRouteJudgeUserPrompt(args: { prompt: string; baseline: RouteDecision }): string {\n\treturn [\n\t\t`Baseline (regex) verdict: tier=${args.baseline.tier}, risk=${args.baseline.risk}, reason=${args.baseline.reasonCode}.`,\n\t\t\"User prompt:\",\n\t\targs.prompt.slice(0, 4000),\n\t].join(\"\\n\");\n}\n\nconst JUDGE_TIERS: readonly string[] = [\"cheap\", \"medium\", \"expensive\"];\nconst JUDGE_RISKS: readonly string[] = [\"read-only\", \"scoped-write\", \"high-impact\", \"approval-required\"];\n\nexport function parseRouteJudgeVerdict(text: string): RouteJudgeVerdict | undefined {\n\tconst trimmed = text.trim();\n\tconst candidates: string[] = [trimmed];\n\tconst fenced = /```(?:json)?\\s*([\\s\\S]*?)```/.exec(trimmed);\n\tif (fenced?.[1]) candidates.push(fenced[1].trim());\n\tconst start = trimmed.indexOf(\"{\");\n\tconst end = trimmed.lastIndexOf(\"}\");\n\tif (start >= 0 && end > start) candidates.push(trimmed.slice(start, end + 1));\n\n\tfor (const candidate of candidates) {\n\t\tlet parsed: unknown;\n\t\ttry {\n\t\t\tparsed = JSON.parse(candidate);\n\t\t} catch {\n\t\t\tcontinue;\n\t\t}\n\t\tif (!parsed || typeof parsed !== \"object\" || Array.isArray(parsed)) continue;\n\t\tconst record = parsed as Record<string, unknown>;\n\t\t// The judge may never select the learning tier or anything outside the three foreground tiers.\n\t\tif (typeof record.tier !== \"string\" || !JUDGE_TIERS.includes(record.tier)) continue;\n\t\tconst risk =\n\t\t\ttypeof record.risk === \"string\" && JUDGE_RISKS.includes(record.risk)\n\t\t\t\t? (record.risk as RouteDecision[\"risk\"])\n\t\t\t\t: undefined;\n\t\tif (!risk) continue;\n\t\treturn {\n\t\t\ttier: record.tier as RouteJudgeVerdict[\"tier\"],\n\t\t\trisk,\n\t\t\ttrivial: record.trivial === true,\n\t\t\treason: typeof record.reason === \"string\" ? record.reason.slice(0, 200) : \"\",\n\t\t};\n\t}\n\treturn undefined;\n}\n\n/** Merge a judge verdict into the baseline decision (pure; never returns learning). */\nexport function applyRouteJudgeVerdict(baseline: RouteDecision, verdict: RouteJudgeVerdict): RouteDecision {\n\t// Enforce the core rule in code, not just in the judge prompt: downgrading a non-cheap\n\t// baseline to cheap requires an EXPLICIT trivial verdict. An untrusted judge saying\n\t// {tier:\"cheap\", trivial:false} for an elevated prompt keeps the baseline tier.\n\tconst tier =\n\t\tverdict.tier === \"cheap\" && baseline.tier !== \"cheap\" && !verdict.trivial ? baseline.tier : verdict.tier;\n\treturn {\n\t\t...baseline,\n\t\ttier,\n\t\trisk: verdict.risk,\n\t\tconfidence: Math.max(baseline.confidence, 0.75),\n\t\treasonCode: `judge_${tier}${verdict.trivial ? \"_trivial\" : \"\"}`,\n\t\treasons: [...baseline.reasons, `Route judge: ${verdict.reason || \"no reason given\"}`],\n\t};\n}\n\nexport interface RouteJudgeRunResult {\n\tdecision: RouteDecision;\n\tverdict?: RouteJudgeVerdict;\n\t/** Set when the judge could not decide and the baseline was kept. */\n\tfallbackReason?: string;\n\tcostUsd: number;\n}\n\nfunction judgeFallbackDecision(baseline: RouteDecision, reason: string): RouteDecision {\n\treturn {\n\t\t...baseline,\n\t\treasonCode: \"judge_unavailable_fallback\",\n\t\treasons: [...baseline.reasons, reason],\n\t};\n}\n\n/**\n * Run the judge over a baseline decision. The completion executor is injected (production:\n * AgentSession.runIsolatedCompletion on the judge model). Never throws; every failure keeps the\n * baseline with a visible fallbackReason.\n */\nexport async function runRouteJudge(args: {\n\tprompt: string;\n\tbaseline: RouteDecision;\n\tcomplete: (input: { systemPrompt: string; userPrompt: string; signal?: AbortSignal }) => Promise<{\n\t\ttext: string;\n\t\tcostUsd: number;\n\t\tstopReason: string;\n\t}>;\n\tsignal?: AbortSignal;\n\tmaxWallClockMs?: number;\n}): Promise<RouteJudgeRunResult> {\n\tconst bounded = await runBoundedCompletion({\n\t\tmaxWallClockMs: args.maxWallClockMs ?? ROUTE_JUDGE_MAX_WALL_CLOCK_MS,\n\t\tsignal: args.signal,\n\t\texecute: (signal) =>\n\t\t\targs.complete({\n\t\t\t\tsystemPrompt: ROUTE_JUDGE_SYSTEM_PROMPT,\n\t\t\t\tuserPrompt: buildRouteJudgeUserPrompt({ prompt: args.prompt, baseline: args.baseline }),\n\t\t\t\tsignal,\n\t\t\t}),\n\t});\n\tconst costUsd = bounded.completion?.costUsd ?? 0;\n\n\tif (bounded.failure || !bounded.completion) {\n\t\treturn {\n\t\t\tdecision: judgeFallbackDecision(args.baseline, \"Route judge unavailable; baseline kept\"),\n\t\t\tfallbackReason: bounded.failure ? `judge_${bounded.failure.reasonCode}` : \"judge_unavailable_fallback\",\n\t\t\tcostUsd,\n\t\t};\n\t}\n\tif (bounded.completion.stopReason === \"error\" || bounded.completion.stopReason === \"aborted\") {\n\t\treturn {\n\t\t\tdecision: judgeFallbackDecision(args.baseline, \"Route judge errored; baseline kept\"),\n\t\t\tfallbackReason: \"judge_model_error\",\n\t\t\tcostUsd,\n\t\t};\n\t}\n\n\tconst verdict = parseRouteJudgeVerdict(bounded.completion.text);\n\tif (!verdict) {\n\t\treturn {\n\t\t\tdecision: judgeFallbackDecision(args.baseline, \"Route judge unparseable; baseline kept\"),\n\t\t\tfallbackReason: \"judge_unparseable_fallback\",\n\t\t\tcostUsd,\n\t\t};\n\t}\n\n\treturn { decision: applyRouteJudgeVerdict(args.baseline, verdict), verdict, costUsd };\n}\n"]}

package/dist/core/model-router/route-judge.js

@@ -82,6 +82,13 @@ export function applyRouteJudgeVerdict(baseline, verdict) {
         reasons: [...baseline.reasons, `Route judge: ${verdict.reason || "no reason given"}`],
     };
 }
+function judgeFallbackDecision(baseline, reason) {
+    return {
+        ...baseline,
+        reasonCode: "judge_unavailable_fallback",
+        reasons: [...baseline.reasons, reason],
+    };
+}
 /**
  * Run the judge over a baseline decision. The completion executor is injected (production:
  * AgentSession.runIsolatedCompletion on the judge model). Never throws; every failure keeps the
@@ -100,17 +107,14 @@ export async function runRouteJudge(args) {
     const costUsd = bounded.completion?.costUsd ?? 0;
     if (bounded.failure || !bounded.completion) {
         return {
-            decision: {
-                ...args.baseline,
-                reasons: [...args.baseline.reasons, "Route judge unavailable; baseline kept"],
-            },
+            decision: judgeFallbackDecision(args.baseline, "Route judge unavailable; baseline kept"),
             fallbackReason: bounded.failure ? `judge_${bounded.failure.reasonCode}` : "judge_unavailable_fallback",
             costUsd,
         };
     }
     if (bounded.completion.stopReason === "error" || bounded.completion.stopReason === "aborted") {
         return {
-            decision: { ...args.baseline, reasons: [...args.baseline.reasons, "Route judge errored; baseline kept"] },
+            decision: judgeFallbackDecision(args.baseline, "Route judge errored; baseline kept"),
             fallbackReason: "judge_model_error",
             costUsd,
         };
@@ -118,7 +122,7 @@ export async function runRouteJudge(args) {
     const verdict = parseRouteJudgeVerdict(bounded.completion.text);
     if (!verdict) {
         return {
-            decision: { ...args.baseline, reasons: [...args.baseline.reasons, "Route judge unparseable; baseline kept"] },
+            decision: judgeFallbackDecision(args.baseline, "Route judge unparseable; baseline kept"),
             fallbackReason: "judge_unparseable_fallback",
             costUsd,
         };

package/dist/core/model-router/route-judge.js.map

@@ -1 +1 @@
-{"version":3,"file":"route-judge.js","sourceRoot":"","sources":["../../../src/core/model-router/route-judge.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,oBAAoB,EAAE,MAAM,mCAAmC,CAAC;AAGzE;;;;;;;GAOG;AAEH,uGAAqG;AACrG,MAAM,CAAC,MAAM,yBAAyB,GAAG;IACxC,wFAAwF;IACxF,sDAAsD;IACtD,yDAAyD;IACzD,0FAA0F;IAC1F,gHAAgH;IAChH,8FAA8F;IAC9F,2CAA2C;IAC3C,0JAA0J;CAC1J,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAEb,MAAM,CAAC,MAAM,6BAA6B,GAAG,GAAG,CAAC;AACjD,MAAM,CAAC,MAAM,6BAA6B,GAAG,MAAM,CAAC;AASpD,MAAM,UAAU,yBAAyB,CAAC,IAAiD,EAAU;IACpG,OAAO;QACN,kCAAkC,IAAI,CAAC,QAAQ,CAAC,IAAI,UAAU,IAAI,CAAC,QAAQ,CAAC,IAAI,YAAY,IAAI,CAAC,QAAQ,CAAC,UAAU,GAAG;QACvH,cAAc;QACd,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC;KAC1B,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAAA,CACb;AAED,MAAM,WAAW,GAAsB,CAAC,OAAO,EAAE,QAAQ,EAAE,WAAW,CAAC,CAAC;AACxE,MAAM,WAAW,GAAsB,CAAC,WAAW,EAAE,cAAc,EAAE,aAAa,EAAE,mBAAmB,CAAC,CAAC;AAEzG,MAAM,UAAU,sBAAsB,CAAC,IAAY,EAAiC;IACnF,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;IAC5B,MAAM,UAAU,GAAa,CAAC,OAAO,CAAC,CAAC;IACvC,MAAM,MAAM,GAAG,8BAA8B,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;IAC5D,IAAI,MAAM,EAAE,CAAC,CAAC,CAAC;QAAE,UAAU,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;IACnD,MAAM,KAAK,GAAG,OAAO,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;IACnC,MAAM,GAAG,GAAG,OAAO,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC;IACrC,IAAI,KAAK,IAAI,CAAC,IAAI,GAAG,GAAG,KAAK;QAAE,UAAU,CAAC,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,KAAK,EAAE,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC;IAE9E,KAAK,MAAM,SAAS,IAAI,UAAU,EAAE,CAAC;QACpC,IAAI,MAAe,CAAC;QACpB,IAAI,CAAC;YACJ,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;QAChC,CAAC;QAAC,MAAM,CAAC;YACR,SAAS;QACV,CAAC;QACD,IAAI,CAAC,MAAM,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC;YAAE,SAAS;QAC7E,MAAM,MAAM,GAAG,MAAiC,CAAC;QACjD,+FAA+F;QAC/F,IAAI,OAAO,MAAM,CAAC,IAAI,KAAK,QAAQ,IAAI,CAAC,WAAW,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC;YAAE,SAAS;QACpF,MAAM,IAAI,GACT,OAAO,MAAM,CAAC,IAAI,KAAK,QAAQ,IAAI,WAAW,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC;YACnE,CAAC,CAAE,MAAM,CAAC,IAA8B;YACxC,CAAC,CAAC,SAAS,CAAC;QACd,IAAI,CAAC,IAAI;YAAE,SAAS;QACpB,OAAO;YACN,IAAI,EAAE,MAAM,CAAC,IAAiC;YAC9C,IAAI;YACJ,OAAO,EAAE,MAAM,CAAC,OAAO,KAAK,IAAI;YAChC,MAAM,EAAE,OAAO,MAAM,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE;SAC5E,CAAC;IACH,CAAC;IACD,OAAO,SAAS,CAAC;AAAA,CACjB;AAED,uFAAuF;AACvF,MAAM,UAAU,sBAAsB,CAAC,QAAuB,EAAE,OAA0B,EAAiB;IAC1G,uFAAuF;IACvF,oFAAoF;IACpF,gFAAgF;IAChF,MAAM,IAAI,GACT,OAAO,CAAC,IAAI,KAAK,OAAO,IAAI,QAAQ,CAAC,IAAI,KAAK,OAAO,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC;IAC1G,OAAO;QACN,GAAG,QAAQ;QACX,IAAI;QACJ,IAAI,EAAE,OAAO,CAAC,IAAI;QAClB,UAAU,EAAE,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,UAAU,EAAE,IAAI,CAAC;QAC/C,UAAU,EAAE,SAAS,IAAI,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE,EAAE;QAC/D,OAAO,EAAE,CAAC,GAAG,QAAQ,CAAC,OAAO,EAAE,gBAAgB,OAAO,CAAC,MAAM,IAAI,iBAAiB,EAAE,CAAC;KACrF,CAAC;AAAA,CACF;AAUD;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,aAAa,CAAC,IAUnC,EAAgC;IAChC,MAAM,OAAO,GAAG,MAAM,oBAAoB,CAAC;QAC1C,cAAc,EAAE,IAAI,CAAC,cAAc,IAAI,6BAA6B;QACpE,MAAM,EAAE,IAAI,CAAC,MAAM;QACnB,OAAO,EAAE,CAAC,MAAM,EAAE,EAAE,CACnB,IAAI,CAAC,QAAQ,CAAC;YACb,YAAY,EAAE,yBAAyB;YACvC,UAAU,EAAE,yBAAyB,CAAC,EAAE,MAAM,EAAE,IAAI,CAAC,MAAM,EAAE,QAAQ,EAAE,IAAI,CAAC,QAAQ,EAAE,CAAC;YACvF,MAAM;SACN,CAAC;KACH,CAAC,CAAC;IACH,MAAM,OAAO,GAAG,OAAO,CAAC,UAAU,EAAE,OAAO,IAAI,CAAC,CAAC;IAEjD,IAAI,OAAO,CAAC,OAAO,IAAI,CAAC,OAAO,CAAC,UAAU,EAAE,CAAC;QAC5C,OAAO;YACN,QAAQ,EAAE;gBACT,GAAG,IAAI,CAAC,QAAQ;gBAChB,OAAO,EAAE,CAAC,GAAG,IAAI,CAAC,QAAQ,CAAC,OAAO,EAAE,wCAAwC,CAAC;aAC7E;YACD,cAAc,EAAE,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,OAAO,CAAC,OAAO,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,4BAA4B;YACtG,OAAO;SACP,CAAC;IACH,CAAC;IACD,IAAI,OAAO,CAAC,UAAU,CAAC,UAAU,KAAK,OAAO,IAAI,OAAO,CAAC,UAAU,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;QAC9F,OAAO;YACN,QAAQ,EAAE,EAAE,GAAG,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,CAAC,GAAG,IAAI,CAAC,QAAQ,CAAC,OAAO,EAAE,oCAAoC,CAAC,EAAE;YACzG,cAAc,EAAE,mBAAmB;YACnC,OAAO;SACP,CAAC;IACH,CAAC;IAED,MAAM,OAAO,GAAG,sBAAsB,CAAC,OAAO,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC;IAChE,IAAI,CAAC,OAAO,EAAE,CAAC;QACd,OAAO;YACN,QAAQ,EAAE,EAAE,GAAG,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,CAAC,GAAG,IAAI,CAAC,QAAQ,CAAC,OAAO,EAAE,wCAAwC,CAAC,EAAE;YAC7G,cAAc,EAAE,4BAA4B;YAC5C,OAAO;SACP,CAAC;IACH,CAAC;IAED,OAAO,EAAE,QAAQ,EAAE,sBAAsB,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,CAAC,EAAE,OAAO,EAAE,OAAO,EAAE,CAAC;AAAA,CACtF","sourcesContent":["import { runBoundedCompletion } from \"../autonomy/bounded-completion.ts\";\nimport type { RouteDecision } from \"../autonomy/contracts.ts\";\n\n/**\n * Routing-only judge: a bounded, tool-less completion (default: the medium model) that decides the\n * final cheap/medium/expensive tier for a user prompt, refining the regex classifier's baseline.\n * Core rule: planning is never cheap unless the judge explicitly deems the task trivial. The judge\n * proposes; the existing pipeline (model resolution, auth, escalation, gates) still decides.\n * Failure is honest: unparseable/timeout/unavailable falls back to the baseline with a visible\n * reasonCode, never silently.\n */\n\n/** Static across calls — callers pass cacheRetention \"short\" so only the variable tail is billed. */\nexport const ROUTE_JUDGE_SYSTEM_PROMPT = [\n\t\"You are a routing judge for a coding agent. You only route; you never answer the task.\",\n\t\"Pick which model tier should handle the user prompt:\",\n\t'- \"cheap\": trivial, mechanical, read-only lookups only.',\n\t'- \"medium\": normal implementation, scoped edits, tests, and NON-trivial planning/design.',\n\t'- \"expensive\": architecture, ambiguity, security/auth, destructive or release operations, high-impact changes.',\n\t\"Planning, design, and strategy prompts are NEVER cheap unless the task is genuinely trivial.\",\n\t\"Respond with STRICT JSON only - no prose:\",\n\t'{\"tier\":\"cheap\"|\"medium\"|\"expensive\",\"risk\":\"read-only\"|\"scoped-write\"|\"high-impact\"|\"approval-required\",\"trivial\":true|false,\"reason\":\"<short reason>\"}',\n].join(\"\\n\");\n\nexport const ROUTE_JUDGE_MAX_OUTPUT_TOKENS = 128;\nexport const ROUTE_JUDGE_MAX_WALL_CLOCK_MS = 10_000;\n\nexport interface RouteJudgeVerdict {\n\ttier: \"cheap\" | \"medium\" | \"expensive\";\n\trisk: RouteDecision[\"risk\"];\n\ttrivial: boolean;\n\treason: string;\n}\n\nexport function buildRouteJudgeUserPrompt(args: { prompt: string; baseline: RouteDecision }): string {\n\treturn [\n\t\t`Baseline (regex) verdict: tier=${args.baseline.tier}, risk=${args.baseline.risk}, reason=${args.baseline.reasonCode}.`,\n\t\t\"User prompt:\",\n\t\targs.prompt.slice(0, 4000),\n\t].join(\"\\n\");\n}\n\nconst JUDGE_TIERS: readonly string[] = [\"cheap\", \"medium\", \"expensive\"];\nconst JUDGE_RISKS: readonly string[] = [\"read-only\", \"scoped-write\", \"high-impact\", \"approval-required\"];\n\nexport function parseRouteJudgeVerdict(text: string): RouteJudgeVerdict | undefined {\n\tconst trimmed = text.trim();\n\tconst candidates: string[] = [trimmed];\n\tconst fenced = /```(?:json)?\\s*([\\s\\S]*?)```/.exec(trimmed);\n\tif (fenced?.[1]) candidates.push(fenced[1].trim());\n\tconst start = trimmed.indexOf(\"{\");\n\tconst end = trimmed.lastIndexOf(\"}\");\n\tif (start >= 0 && end > start) candidates.push(trimmed.slice(start, end + 1));\n\n\tfor (const candidate of candidates) {\n\t\tlet parsed: unknown;\n\t\ttry {\n\t\t\tparsed = JSON.parse(candidate);\n\t\t} catch {\n\t\t\tcontinue;\n\t\t}\n\t\tif (!parsed || typeof parsed !== \"object\" || Array.isArray(parsed)) continue;\n\t\tconst record = parsed as Record<string, unknown>;\n\t\t// The judge may never select the learning tier or anything outside the three foreground tiers.\n\t\tif (typeof record.tier !== \"string\" || !JUDGE_TIERS.includes(record.tier)) continue;\n\t\tconst risk =\n\t\t\ttypeof record.risk === \"string\" && JUDGE_RISKS.includes(record.risk)\n\t\t\t\t? (record.risk as RouteDecision[\"risk\"])\n\t\t\t\t: undefined;\n\t\tif (!risk) continue;\n\t\treturn {\n\t\t\ttier: record.tier as RouteJudgeVerdict[\"tier\"],\n\t\t\trisk,\n\t\t\ttrivial: record.trivial === true,\n\t\t\treason: typeof record.reason === \"string\" ? record.reason.slice(0, 200) : \"\",\n\t\t};\n\t}\n\treturn undefined;\n}\n\n/** Merge a judge verdict into the baseline decision (pure; never returns learning). */\nexport function applyRouteJudgeVerdict(baseline: RouteDecision, verdict: RouteJudgeVerdict): RouteDecision {\n\t// Enforce the core rule in code, not just in the judge prompt: downgrading a non-cheap\n\t// baseline to cheap requires an EXPLICIT trivial verdict. An untrusted judge saying\n\t// {tier:\"cheap\", trivial:false} for an elevated prompt keeps the baseline tier.\n\tconst tier =\n\t\tverdict.tier === \"cheap\" && baseline.tier !== \"cheap\" && !verdict.trivial ? baseline.tier : verdict.tier;\n\treturn {\n\t\t...baseline,\n\t\ttier,\n\t\trisk: verdict.risk,\n\t\tconfidence: Math.max(baseline.confidence, 0.75),\n\t\treasonCode: `judge_${tier}${verdict.trivial ? \"_trivial\" : \"\"}`,\n\t\treasons: [...baseline.reasons, `Route judge: ${verdict.reason || \"no reason given\"}`],\n\t};\n}\n\nexport interface RouteJudgeRunResult {\n\tdecision: RouteDecision;\n\tverdict?: RouteJudgeVerdict;\n\t/** Set when the judge could not decide and the baseline was kept. */\n\tfallbackReason?: string;\n\tcostUsd: number;\n}\n\n/**\n * Run the judge over a baseline decision. The completion executor is injected (production:\n * AgentSession.runIsolatedCompletion on the judge model). Never throws; every failure keeps the\n * baseline with a visible fallbackReason.\n */\nexport async function runRouteJudge(args: {\n\tprompt: string;\n\tbaseline: RouteDecision;\n\tcomplete: (input: { systemPrompt: string; userPrompt: string; signal?: AbortSignal }) => Promise<{\n\t\ttext: string;\n\t\tcostUsd: number;\n\t\tstopReason: string;\n\t}>;\n\tsignal?: AbortSignal;\n\tmaxWallClockMs?: number;\n}): Promise<RouteJudgeRunResult> {\n\tconst bounded = await runBoundedCompletion({\n\t\tmaxWallClockMs: args.maxWallClockMs ?? ROUTE_JUDGE_MAX_WALL_CLOCK_MS,\n\t\tsignal: args.signal,\n\t\texecute: (signal) =>\n\t\t\targs.complete({\n\t\t\t\tsystemPrompt: ROUTE_JUDGE_SYSTEM_PROMPT,\n\t\t\t\tuserPrompt: buildRouteJudgeUserPrompt({ prompt: args.prompt, baseline: args.baseline }),\n\t\t\t\tsignal,\n\t\t\t}),\n\t});\n\tconst costUsd = bounded.completion?.costUsd ?? 0;\n\n\tif (bounded.failure || !bounded.completion) {\n\t\treturn {\n\t\t\tdecision: {\n\t\t\t\t...args.baseline,\n\t\t\t\treasons: [...args.baseline.reasons, \"Route judge unavailable; baseline kept\"],\n\t\t\t},\n\t\t\tfallbackReason: bounded.failure ? `judge_${bounded.failure.reasonCode}` : \"judge_unavailable_fallback\",\n\t\t\tcostUsd,\n\t\t};\n\t}\n\tif (bounded.completion.stopReason === \"error\" || bounded.completion.stopReason === \"aborted\") {\n\t\treturn {\n\t\t\tdecision: { ...args.baseline, reasons: [...args.baseline.reasons, \"Route judge errored; baseline kept\"] },\n\t\t\tfallbackReason: \"judge_model_error\",\n\t\t\tcostUsd,\n\t\t};\n\t}\n\n\tconst verdict = parseRouteJudgeVerdict(bounded.completion.text);\n\tif (!verdict) {\n\t\treturn {\n\t\t\tdecision: { ...args.baseline, reasons: [...args.baseline.reasons, \"Route judge unparseable; baseline kept\"] },\n\t\t\tfallbackReason: \"judge_unparseable_fallback\",\n\t\t\tcostUsd,\n\t\t};\n\t}\n\n\treturn { decision: applyRouteJudgeVerdict(args.baseline, verdict), verdict, costUsd };\n}\n"]}
+{"version":3,"file":"route-judge.js","sourceRoot":"","sources":["../../../src/core/model-router/route-judge.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,oBAAoB,EAAE,MAAM,mCAAmC,CAAC;AAGzE;;;;;;;GAOG;AAEH,uGAAqG;AACrG,MAAM,CAAC,MAAM,yBAAyB,GAAG;IACxC,wFAAwF;IACxF,sDAAsD;IACtD,yDAAyD;IACzD,0FAA0F;IAC1F,gHAAgH;IAChH,8FAA8F;IAC9F,2CAA2C;IAC3C,0JAA0J;CAC1J,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAEb,MAAM,CAAC,MAAM,6BAA6B,GAAG,GAAG,CAAC;AACjD,MAAM,CAAC,MAAM,6BAA6B,GAAG,MAAM,CAAC;AASpD,MAAM,UAAU,yBAAyB,CAAC,IAAiD,EAAU;IACpG,OAAO;QACN,kCAAkC,IAAI,CAAC,QAAQ,CAAC,IAAI,UAAU,IAAI,CAAC,QAAQ,CAAC,IAAI,YAAY,IAAI,CAAC,QAAQ,CAAC,UAAU,GAAG;QACvH,cAAc;QACd,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC;KAC1B,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAAA,CACb;AAED,MAAM,WAAW,GAAsB,CAAC,OAAO,EAAE,QAAQ,EAAE,WAAW,CAAC,CAAC;AACxE,MAAM,WAAW,GAAsB,CAAC,WAAW,EAAE,cAAc,EAAE,aAAa,EAAE,mBAAmB,CAAC,CAAC;AAEzG,MAAM,UAAU,sBAAsB,CAAC,IAAY,EAAiC;IACnF,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;IAC5B,MAAM,UAAU,GAAa,CAAC,OAAO,CAAC,CAAC;IACvC,MAAM,MAAM,GAAG,8BAA8B,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;IAC5D,IAAI,MAAM,EAAE,CAAC,CAAC,CAAC;QAAE,UAAU,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;IACnD,MAAM,KAAK,GAAG,OAAO,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;IACnC,MAAM,GAAG,GAAG,OAAO,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC;IACrC,IAAI,KAAK,IAAI,CAAC,IAAI,GAAG,GAAG,KAAK;QAAE,UAAU,CAAC,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,KAAK,EAAE,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC;IAE9E,KAAK,MAAM,SAAS,IAAI,UAAU,EAAE,CAAC;QACpC,IAAI,MAAe,CAAC;QACpB,IAAI,CAAC;YACJ,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;QAChC,CAAC;QAAC,MAAM,CAAC;YACR,SAAS;QACV,CAAC;QACD,IAAI,CAAC,MAAM,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC;YAAE,SAAS;QAC7E,MAAM,MAAM,GAAG,MAAiC,CAAC;QACjD,+FAA+F;QAC/F,IAAI,OAAO,MAAM,CAAC,IAAI,KAAK,QAAQ,IAAI,CAAC,WAAW,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC;YAAE,SAAS;QACpF,MAAM,IAAI,GACT,OAAO,MAAM,CAAC,IAAI,KAAK,QAAQ,IAAI,WAAW,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC;YACnE,CAAC,CAAE,MAAM,CAAC,IAA8B;YACxC,CAAC,CAAC,SAAS,CAAC;QACd,IAAI,CAAC,IAAI;YAAE,SAAS;QACpB,OAAO;YACN,IAAI,EAAE,MAAM,CAAC,IAAiC;YAC9C,IAAI;YACJ,OAAO,EAAE,MAAM,CAAC,OAAO,KAAK,IAAI;YAChC,MAAM,EAAE,OAAO,MAAM,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE;SAC5E,CAAC;IACH,CAAC;IACD,OAAO,SAAS,CAAC;AAAA,CACjB;AAED,uFAAuF;AACvF,MAAM,UAAU,sBAAsB,CAAC,QAAuB,EAAE,OAA0B,EAAiB;IAC1G,uFAAuF;IACvF,oFAAoF;IACpF,gFAAgF;IAChF,MAAM,IAAI,GACT,OAAO,CAAC,IAAI,KAAK,OAAO,IAAI,QAAQ,CAAC,IAAI,KAAK,OAAO,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC;IAC1G,OAAO;QACN,GAAG,QAAQ;QACX,IAAI;QACJ,IAAI,EAAE,OAAO,CAAC,IAAI;QAClB,UAAU,EAAE,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,UAAU,EAAE,IAAI,CAAC;QAC/C,UAAU,EAAE,SAAS,IAAI,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE,EAAE;QAC/D,OAAO,EAAE,CAAC,GAAG,QAAQ,CAAC,OAAO,EAAE,gBAAgB,OAAO,CAAC,MAAM,IAAI,iBAAiB,EAAE,CAAC;KACrF,CAAC;AAAA,CACF;AAUD,SAAS,qBAAqB,CAAC,QAAuB,EAAE,MAAc,EAAiB;IACtF,OAAO;QACN,GAAG,QAAQ;QACX,UAAU,EAAE,4BAA4B;QACxC,OAAO,EAAE,CAAC,GAAG,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;KACtC,CAAC;AAAA,CACF;AAED;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,aAAa,CAAC,IAUnC,EAAgC;IAChC,MAAM,OAAO,GAAG,MAAM,oBAAoB,CAAC;QAC1C,cAAc,EAAE,IAAI,CAAC,cAAc,IAAI,6BAA6B;QACpE,MAAM,EAAE,IAAI,CAAC,MAAM;QACnB,OAAO,EAAE,CAAC,MAAM,EAAE,EAAE,CACnB,IAAI,CAAC,QAAQ,CAAC;YACb,YAAY,EAAE,yBAAyB;YACvC,UAAU,EAAE,yBAAyB,CAAC,EAAE,MAAM,EAAE,IAAI,CAAC,MAAM,EAAE,QAAQ,EAAE,IAAI,CAAC,QAAQ,EAAE,CAAC;YACvF,MAAM;SACN,CAAC;KACH,CAAC,CAAC;IACH,MAAM,OAAO,GAAG,OAAO,CAAC,UAAU,EAAE,OAAO,IAAI,CAAC,CAAC;IAEjD,IAAI,OAAO,CAAC,OAAO,IAAI,CAAC,OAAO,CAAC,UAAU,EAAE,CAAC;QAC5C,OAAO;YACN,QAAQ,EAAE,qBAAqB,CAAC,IAAI,CAAC,QAAQ,EAAE,wCAAwC,CAAC;YACxF,cAAc,EAAE,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,OAAO,CAAC,OAAO,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,4BAA4B;YACtG,OAAO;SACP,CAAC;IACH,CAAC;IACD,IAAI,OAAO,CAAC,UAAU,CAAC,UAAU,KAAK,OAAO,IAAI,OAAO,CAAC,UAAU,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;QAC9F,OAAO;YACN,QAAQ,EAAE,qBAAqB,CAAC,IAAI,CAAC,QAAQ,EAAE,oCAAoC,CAAC;YACpF,cAAc,EAAE,mBAAmB;YACnC,OAAO;SACP,CAAC;IACH,CAAC;IAED,MAAM,OAAO,GAAG,sBAAsB,CAAC,OAAO,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC;IAChE,IAAI,CAAC,OAAO,EAAE,CAAC;QACd,OAAO;YACN,QAAQ,EAAE,qBAAqB,CAAC,IAAI,CAAC,QAAQ,EAAE,wCAAwC,CAAC;YACxF,cAAc,EAAE,4BAA4B;YAC5C,OAAO;SACP,CAAC;IACH,CAAC;IAED,OAAO,EAAE,QAAQ,EAAE,sBAAsB,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,CAAC,EAAE,OAAO,EAAE,OAAO,EAAE,CAAC;AAAA,CACtF","sourcesContent":["import { runBoundedCompletion } from \"../autonomy/bounded-completion.ts\";\nimport type { RouteDecision } from \"../autonomy/contracts.ts\";\n\n/**\n * Routing-only judge: a bounded, tool-less completion (default: the medium model) that decides the\n * final cheap/medium/expensive tier for a user prompt, refining the regex classifier's baseline.\n * Core rule: planning is never cheap unless the judge explicitly deems the task trivial. The judge\n * proposes; the existing pipeline (model resolution, auth, escalation, gates) still decides.\n * Failure is honest: unparseable/timeout/unavailable falls back to the baseline with a visible\n * reasonCode, never silently.\n */\n\n/** Static across calls — callers pass cacheRetention \"short\" so only the variable tail is billed. */\nexport const ROUTE_JUDGE_SYSTEM_PROMPT = [\n\t\"You are a routing judge for a coding agent. You only route; you never answer the task.\",\n\t\"Pick which model tier should handle the user prompt:\",\n\t'- \"cheap\": trivial, mechanical, read-only lookups only.',\n\t'- \"medium\": normal implementation, scoped edits, tests, and NON-trivial planning/design.',\n\t'- \"expensive\": architecture, ambiguity, security/auth, destructive or release operations, high-impact changes.',\n\t\"Planning, design, and strategy prompts are NEVER cheap unless the task is genuinely trivial.\",\n\t\"Respond with STRICT JSON only - no prose:\",\n\t'{\"tier\":\"cheap\"|\"medium\"|\"expensive\",\"risk\":\"read-only\"|\"scoped-write\"|\"high-impact\"|\"approval-required\",\"trivial\":true|false,\"reason\":\"<short reason>\"}',\n].join(\"\\n\");\n\nexport const ROUTE_JUDGE_MAX_OUTPUT_TOKENS = 128;\nexport const ROUTE_JUDGE_MAX_WALL_CLOCK_MS = 10_000;\n\nexport interface RouteJudgeVerdict {\n\ttier: \"cheap\" | \"medium\" | \"expensive\";\n\trisk: RouteDecision[\"risk\"];\n\ttrivial: boolean;\n\treason: string;\n}\n\nexport function buildRouteJudgeUserPrompt(args: { prompt: string; baseline: RouteDecision }): string {\n\treturn [\n\t\t`Baseline (regex) verdict: tier=${args.baseline.tier}, risk=${args.baseline.risk}, reason=${args.baseline.reasonCode}.`,\n\t\t\"User prompt:\",\n\t\targs.prompt.slice(0, 4000),\n\t].join(\"\\n\");\n}\n\nconst JUDGE_TIERS: readonly string[] = [\"cheap\", \"medium\", \"expensive\"];\nconst JUDGE_RISKS: readonly string[] = [\"read-only\", \"scoped-write\", \"high-impact\", \"approval-required\"];\n\nexport function parseRouteJudgeVerdict(text: string): RouteJudgeVerdict | undefined {\n\tconst trimmed = text.trim();\n\tconst candidates: string[] = [trimmed];\n\tconst fenced = /```(?:json)?\\s*([\\s\\S]*?)```/.exec(trimmed);\n\tif (fenced?.[1]) candidates.push(fenced[1].trim());\n\tconst start = trimmed.indexOf(\"{\");\n\tconst end = trimmed.lastIndexOf(\"}\");\n\tif (start >= 0 && end > start) candidates.push(trimmed.slice(start, end + 1));\n\n\tfor (const candidate of candidates) {\n\t\tlet parsed: unknown;\n\t\ttry {\n\t\t\tparsed = JSON.parse(candidate);\n\t\t} catch {\n\t\t\tcontinue;\n\t\t}\n\t\tif (!parsed || typeof parsed !== \"object\" || Array.isArray(parsed)) continue;\n\t\tconst record = parsed as Record<string, unknown>;\n\t\t// The judge may never select the learning tier or anything outside the three foreground tiers.\n\t\tif (typeof record.tier !== \"string\" || !JUDGE_TIERS.includes(record.tier)) continue;\n\t\tconst risk =\n\t\t\ttypeof record.risk === \"string\" && JUDGE_RISKS.includes(record.risk)\n\t\t\t\t? (record.risk as RouteDecision[\"risk\"])\n\t\t\t\t: undefined;\n\t\tif (!risk) continue;\n\t\treturn {\n\t\t\ttier: record.tier as RouteJudgeVerdict[\"tier\"],\n\t\t\trisk,\n\t\t\ttrivial: record.trivial === true,\n\t\t\treason: typeof record.reason === \"string\" ? record.reason.slice(0, 200) : \"\",\n\t\t};\n\t}\n\treturn undefined;\n}\n\n/** Merge a judge verdict into the baseline decision (pure; never returns learning). */\nexport function applyRouteJudgeVerdict(baseline: RouteDecision, verdict: RouteJudgeVerdict): RouteDecision {\n\t// Enforce the core rule in code, not just in the judge prompt: downgrading a non-cheap\n\t// baseline to cheap requires an EXPLICIT trivial verdict. An untrusted judge saying\n\t// {tier:\"cheap\", trivial:false} for an elevated prompt keeps the baseline tier.\n\tconst tier =\n\t\tverdict.tier === \"cheap\" && baseline.tier !== \"cheap\" && !verdict.trivial ? baseline.tier : verdict.tier;\n\treturn {\n\t\t...baseline,\n\t\ttier,\n\t\trisk: verdict.risk,\n\t\tconfidence: Math.max(baseline.confidence, 0.75),\n\t\treasonCode: `judge_${tier}${verdict.trivial ? \"_trivial\" : \"\"}`,\n\t\treasons: [...baseline.reasons, `Route judge: ${verdict.reason || \"no reason given\"}`],\n\t};\n}\n\nexport interface RouteJudgeRunResult {\n\tdecision: RouteDecision;\n\tverdict?: RouteJudgeVerdict;\n\t/** Set when the judge could not decide and the baseline was kept. */\n\tfallbackReason?: string;\n\tcostUsd: number;\n}\n\nfunction judgeFallbackDecision(baseline: RouteDecision, reason: string): RouteDecision {\n\treturn {\n\t\t...baseline,\n\t\treasonCode: \"judge_unavailable_fallback\",\n\t\treasons: [...baseline.reasons, reason],\n\t};\n}\n\n/**\n * Run the judge over a baseline decision. The completion executor is injected (production:\n * AgentSession.runIsolatedCompletion on the judge model). Never throws; every failure keeps the\n * baseline with a visible fallbackReason.\n */\nexport async function runRouteJudge(args: {\n\tprompt: string;\n\tbaseline: RouteDecision;\n\tcomplete: (input: { systemPrompt: string; userPrompt: string; signal?: AbortSignal }) => Promise<{\n\t\ttext: string;\n\t\tcostUsd: number;\n\t\tstopReason: string;\n\t}>;\n\tsignal?: AbortSignal;\n\tmaxWallClockMs?: number;\n}): Promise<RouteJudgeRunResult> {\n\tconst bounded = await runBoundedCompletion({\n\t\tmaxWallClockMs: args.maxWallClockMs ?? ROUTE_JUDGE_MAX_WALL_CLOCK_MS,\n\t\tsignal: args.signal,\n\t\texecute: (signal) =>\n\t\t\targs.complete({\n\t\t\t\tsystemPrompt: ROUTE_JUDGE_SYSTEM_PROMPT,\n\t\t\t\tuserPrompt: buildRouteJudgeUserPrompt({ prompt: args.prompt, baseline: args.baseline }),\n\t\t\t\tsignal,\n\t\t\t}),\n\t});\n\tconst costUsd = bounded.completion?.costUsd ?? 0;\n\n\tif (bounded.failure || !bounded.completion) {\n\t\treturn {\n\t\t\tdecision: judgeFallbackDecision(args.baseline, \"Route judge unavailable; baseline kept\"),\n\t\t\tfallbackReason: bounded.failure ? `judge_${bounded.failure.reasonCode}` : \"judge_unavailable_fallback\",\n\t\t\tcostUsd,\n\t\t};\n\t}\n\tif (bounded.completion.stopReason === \"error\" || bounded.completion.stopReason === \"aborted\") {\n\t\treturn {\n\t\t\tdecision: judgeFallbackDecision(args.baseline, \"Route judge errored; baseline kept\"),\n\t\t\tfallbackReason: \"judge_model_error\",\n\t\t\tcostUsd,\n\t\t};\n\t}\n\n\tconst verdict = parseRouteJudgeVerdict(bounded.completion.text);\n\tif (!verdict) {\n\t\treturn {\n\t\t\tdecision: judgeFallbackDecision(args.baseline, \"Route judge unparseable; baseline kept\"),\n\t\t\tfallbackReason: \"judge_unparseable_fallback\",\n\t\t\tcostUsd,\n\t\t};\n\t}\n\n\treturn { decision: applyRouteJudgeVerdict(args.baseline, verdict), verdict, costUsd };\n}\n"]}

package/dist/core/model-router/tool-escalation.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"tool-escalation.d.ts","sourceRoot":"","sources":["../../../src/core/model-router/tool-escalation.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,0BAA0B,CAAC;AA2F1D,wBAAgB,6BAA6B,CAAC,OAAO,EAAE;IACtD,IAAI,EAAE,SAAS,CAAC;IAChB,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,2EAA2E;IAC3E,UAAU,CAAC,EAAE,MAAM,CAAC;CACpB,GAAG,OAAO,CAeV","sourcesContent":["import type { ModelTier } from \"../autonomy/contracts.ts\";\n\nconst READ_ONLY_TOOL_NAMES = new Set([\n\t\"read\",\n\t\"grep\",\n\t\"find\",\n\t\"ls\",\n\t\"list\",\n\t\"search\",\n\t\"glob\",\n\t\"view_file\",\n\t\"list_dir\",\n\t\"grep_search\",\n\t\"search_web\",\n\t\"read_url_content\",\n\t\"read_browser_page\",\n]);\n\nconst SHELL_TOOL_NAMES = new Set([\"bash\", \"exec\", \"execute\", \"run\", \"run_command\", \"shell\"]);\n\nconst READ_ONLY_COMMANDS = new Set([\n\t\"awk\",\n\t\"cat\",\n\t\"date\",\n\t\"df\",\n\t\"du\",\n\t\"env\",\n\t\"git\",\n\t\"grep\",\n\t\"head\",\n\t\"jq\",\n\t\"ls\",\n\t\"node\",\n\t\"npm\",\n\t\"pnpm\",\n\t\"pwd\",\n\t\"rg\",\n\t\"sed\",\n\t\"tail\",\n\t\"test\",\n\t\"tsc\",\n\t\"wc\",\n\t\"which\",\n\t\"yarn\",\n]);\n\nconst READ_ONLY_GIT_SUBCOMMANDS = new Set([\"branch\", \"diff\", \"log\", \"rev-parse\", \"show\", \"status\", \"tag\"]);\nconst READ_ONLY_NPM_SUBCOMMANDS = new Set([\"info\", \"list\", \"ls\", \"outdated\", \"view\", \"whoami\"]);\nconst MUTATING_SHELL_TOKEN_RE =\n\t/(^|\\s)(>|>>|2>|&>|tee\\b|rm\\b|mv\\b|cp\\b|mkdir\\b|touch\\b|chmod\\b|chown\\b|install\\b|commit\\b|push\\b|publish\\b|deploy\\b|apply\\b|add\\b|checkout\\b|switch\\b|reset\\b|clean\\b|stash\\b|merge\\b|rebase\\b|npm\\s+(?:i|install|ci|update|publish|run)\\b|pnpm\\s+(?:i|install|update|publish|run)\\b|yarn\\s+(?:add|install|upgrade|publish|run)\\b)/i;\nconst MUTATING_TOOL_NAME_RE =\n\t/(bash|exec|execute|run|shell|write|edit|patch|replace|delete|remove|move|rename|create|mkdir|touch|install|commit|push|publish|deploy|apply)/i;\n\nfunction getShellCommand(args: unknown): string | undefined {\n\tif (!args || typeof args !== \"object\") return undefined;\n\tconst record = args as Record<string, unknown>;\n\tconst command = record.command ?? record.cmd ?? record.shellCommand;\n\treturn typeof command === \"string\" ? command.trim() : undefined;\n}\n\nfunction commandName(segment: string): string | undefined {\n\tconst first = segment.trim().match(/^[A-Za-z0-9_./-]+/)?.[0];\n\tif (!first) return undefined;\n\tconst parts = first.split(\"/\");\n\treturn parts[parts.length - 1]?.toLowerCase();\n}\n\nfunction commandArg(segment: string, index: number): string | undefined {\n\treturn segment.trim().split(/\\s+/)[index]?.toLowerCase();\n}\n\nfunction isReadOnlyShellSegment(segment: string): boolean {\n\tconst name = commandName(segment);\n\tif (!name || !READ_ONLY_COMMANDS.has(name)) return false;\n\tif (name === \"git\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_GIT_SUBCOMMANDS.has(subcommand));\n\t}\n\tif (name === \"npm\" || name === \"pnpm\" || name === \"yarn\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_NPM_SUBCOMMANDS.has(subcommand));\n\t}\n\treturn true;\n}\n\nfunction isReadOnlyShellCommand(command: string): boolean {\n\tif (!command || MUTATING_SHELL_TOKEN_RE.test(command)) return false;\n\tconst segments = command.split(/\\s*(?:&&|\\|\\||[;|\\r\\n])\\s*/).map((segment) => segment.trim());\n\treturn segments.length > 0 && segments.every((segment) => segment.length > 0 && isReadOnlyShellSegment(segment));\n}\n\nexport function shouldEscalateModelRouterTool(options: {\n\ttier: ModelTier;\n\ttoolName: string;\n\targs?: unknown;\n\t/** The route's reasonCode; executor-lane turns carry \"executor_direct\". */\n\treasonCode?: string;\n}): boolean {\n\tif (options.tier !== \"cheap\") return false;\n\tconst toolName = options.toolName.trim().toLowerCase();\n\t// Executor-lane turns (G16) exist to run exactly one tool: run_toolkit_script, which enforces\n\t// its own safety (danger confirmation, structural exit-code contract). Escalating on it would\n\t// abort every executor turn at the moment it does its job. Any OTHER mutating tool still\n\t// escalates to the expensive model as usual.\n\tif (options.reasonCode === \"executor_direct\" && toolName === \"run_toolkit_script\") return false;\n\tif (!toolName) return true;\n\tif (READ_ONLY_TOOL_NAMES.has(toolName)) return false;\n\tif (SHELL_TOOL_NAMES.has(toolName)) {\n\t\tconst command = getShellCommand(options.args);\n\t\treturn command ? !isReadOnlyShellCommand(command) : true;\n\t}\n\treturn MUTATING_TOOL_NAME_RE.test(toolName) || !toolName.startsWith(\"read_\");\n}\n"]}
+{"version":3,"file":"tool-escalation.d.ts","sourceRoot":"","sources":["../../../src/core/model-router/tool-escalation.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,0BAA0B,CAAC;AA6F1D,wBAAgB,6BAA6B,CAAC,OAAO,EAAE;IACtD,IAAI,EAAE,SAAS,CAAC;IAChB,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,2EAA2E;IAC3E,UAAU,CAAC,EAAE,MAAM,CAAC;CACpB,GAAG,OAAO,CAeV","sourcesContent":["import type { ModelTier } from \"../autonomy/contracts.ts\";\n\nconst READ_ONLY_TOOL_NAMES = new Set([\n\t\"read\",\n\t\"grep\",\n\t\"find\",\n\t\"ls\",\n\t\"list\",\n\t\"search\",\n\t\"glob\",\n\t\"view_file\",\n\t\"list_dir\",\n\t\"grep_search\",\n\t\"search_web\",\n\t\"read_url_content\",\n\t\"read_browser_page\",\n]);\n\nconst SHELL_TOOL_NAMES = new Set([\"bash\", \"exec\", \"execute\", \"run\", \"run_command\", \"shell\"]);\n\nconst READ_ONLY_COMMANDS = new Set([\n\t\"awk\",\n\t\"cat\",\n\t\"date\",\n\t\"df\",\n\t\"du\",\n\t\"env\",\n\t\"git\",\n\t\"grep\",\n\t\"head\",\n\t\"jq\",\n\t\"ls\",\n\t\"node\",\n\t\"npm\",\n\t\"pnpm\",\n\t\"pwd\",\n\t\"rg\",\n\t\"sed\",\n\t\"tail\",\n\t\"test\",\n\t\"tsc\",\n\t\"wc\",\n\t\"which\",\n\t\"yarn\",\n]);\n\nconst READ_ONLY_GIT_SUBCOMMANDS = new Set([\"branch\", \"diff\", \"log\", \"rev-parse\", \"show\", \"status\", \"tag\"]);\nconst READ_ONLY_NPM_SUBCOMMANDS = new Set([\"info\", \"list\", \"ls\", \"outdated\", \"view\", \"whoami\"]);\nconst MUTATING_SHELL_TOKEN_RE =\n\t/(^|\\s)(>|>>|2>|&>|tee\\b|rm\\b|mv\\b|cp\\b|mkdir\\b|touch\\b|chmod\\b|chown\\b|install\\b|commit\\b|push\\b|publish\\b|deploy\\b|apply\\b|add\\b|checkout\\b|switch\\b|reset\\b|clean\\b|stash\\b|merge\\b|rebase\\b|npm\\s+(?:i|install|ci|update|publish|run)\\b|pnpm\\s+(?:i|install|update|publish|run)\\b|yarn\\s+(?:add|install|upgrade|publish|run)\\b)/i;\nconst UNSAFE_NESTED_SHELL_EXECUTION_RE = /(`|\\$\\(|\\bfind\\b[\\s\\S]*\\s-exec(?:dir)?\\b|\\bxargs\\b)/i;\nconst MUTATING_TOOL_NAME_RE =\n\t/(bash|exec|execute|run|shell|write|edit|patch|replace|delete|remove|move|rename|create|mkdir|touch|install|commit|push|publish|deploy|apply)/i;\n\nfunction getShellCommand(args: unknown): string | undefined {\n\tif (!args || typeof args !== \"object\") return undefined;\n\tconst record = args as Record<string, unknown>;\n\tconst command = record.command ?? record.cmd ?? record.shellCommand;\n\treturn typeof command === \"string\" ? command.trim() : undefined;\n}\n\nfunction commandName(segment: string): string | undefined {\n\tconst first = segment.trim().match(/^[A-Za-z0-9_./-]+/)?.[0];\n\tif (!first) return undefined;\n\tconst parts = first.split(\"/\");\n\treturn parts[parts.length - 1]?.toLowerCase();\n}\n\nfunction commandArg(segment: string, index: number): string | undefined {\n\treturn segment.trim().split(/\\s+/)[index]?.toLowerCase();\n}\n\nfunction isReadOnlyShellSegment(segment: string): boolean {\n\tconst name = commandName(segment);\n\tif (!name || !READ_ONLY_COMMANDS.has(name)) return false;\n\tif (name === \"git\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_GIT_SUBCOMMANDS.has(subcommand));\n\t}\n\tif (name === \"npm\" || name === \"pnpm\" || name === \"yarn\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_NPM_SUBCOMMANDS.has(subcommand));\n\t}\n\treturn true;\n}\n\nfunction isReadOnlyShellCommand(command: string): boolean {\n\tif (!command || MUTATING_SHELL_TOKEN_RE.test(command) || UNSAFE_NESTED_SHELL_EXECUTION_RE.test(command))\n\t\treturn false;\n\tconst segments = command.split(/\\s*(?:&&|\\|\\||[;|\\r\\n])\\s*/).map((segment) => segment.trim());\n\treturn segments.length > 0 && segments.every((segment) => segment.length > 0 && isReadOnlyShellSegment(segment));\n}\n\nexport function shouldEscalateModelRouterTool(options: {\n\ttier: ModelTier;\n\ttoolName: string;\n\targs?: unknown;\n\t/** The route's reasonCode; executor-lane turns carry \"executor_direct\". */\n\treasonCode?: string;\n}): boolean {\n\tif (options.tier !== \"cheap\") return false;\n\tconst toolName = options.toolName.trim().toLowerCase();\n\t// Executor-lane turns (G16) exist to run exactly one tool: run_toolkit_script, which enforces\n\t// its own safety (danger confirmation, structural exit-code contract). Escalating on it would\n\t// abort every executor turn at the moment it does its job. Any OTHER mutating tool still\n\t// escalates to the expensive model as usual.\n\tif (options.reasonCode === \"executor_direct\" && toolName === \"run_toolkit_script\") return false;\n\tif (!toolName) return true;\n\tif (READ_ONLY_TOOL_NAMES.has(toolName)) return false;\n\tif (SHELL_TOOL_NAMES.has(toolName)) {\n\t\tconst command = getShellCommand(options.args);\n\t\treturn command ? !isReadOnlyShellCommand(command) : true;\n\t}\n\treturn MUTATING_TOOL_NAME_RE.test(toolName) || !toolName.startsWith(\"read_\");\n}\n"]}

package/dist/core/model-router/tool-escalation.js

@@ -42,6 +42,7 @@ const READ_ONLY_COMMANDS = new Set([
 const READ_ONLY_GIT_SUBCOMMANDS = new Set(["branch", "diff", "log", "rev-parse", "show", "status", "tag"]);
 const READ_ONLY_NPM_SUBCOMMANDS = new Set(["info", "list", "ls", "outdated", "view", "whoami"]);
 const MUTATING_SHELL_TOKEN_RE = /(^|\s)(>|>>|2>|&>|tee\b|rm\b|mv\b|cp\b|mkdir\b|touch\b|chmod\b|chown\b|install\b|commit\b|push\b|publish\b|deploy\b|apply\b|add\b|checkout\b|switch\b|reset\b|clean\b|stash\b|merge\b|rebase\b|npm\s+(?:i|install|ci|update|publish|run)\b|pnpm\s+(?:i|install|update|publish|run)\b|yarn\s+(?:add|install|upgrade|publish|run)\b)/i;
+const UNSAFE_NESTED_SHELL_EXECUTION_RE = /(`|\$\(|\bfind\b[\s\S]*\s-exec(?:dir)?\b|\bxargs\b)/i;
 const MUTATING_TOOL_NAME_RE = /(bash|exec|execute|run|shell|write|edit|patch|replace|delete|remove|move|rename|create|mkdir|touch|install|commit|push|publish|deploy|apply)/i;
 function getShellCommand(args) {
     if (!args || typeof args !== "object")
@@ -75,7 +76,7 @@ function isReadOnlyShellSegment(segment) {
     return true;
 }
 function isReadOnlyShellCommand(command) {
-    if (!command || MUTATING_SHELL_TOKEN_RE.test(command))
+    if (!command || MUTATING_SHELL_TOKEN_RE.test(command) || UNSAFE_NESTED_SHELL_EXECUTION_RE.test(command))
         return false;
     const segments = command.split(/\s*(?:&&|\|\||[;|\r\n])\s*/).map((segment) => segment.trim());
     return segments.length > 0 && segments.every((segment) => segment.length > 0 && isReadOnlyShellSegment(segment));

package/dist/core/model-router/tool-escalation.js.map

@@ -1 +1 @@
-{"version":3,"file":"tool-escalation.js","sourceRoot":"","sources":["../../../src/core/model-router/tool-escalation.ts"],"names":[],"mappings":"AAEA,MAAM,oBAAoB,GAAG,IAAI,GAAG,CAAC;IACpC,MAAM;IACN,MAAM;IACN,MAAM;IACN,IAAI;IACJ,MAAM;IACN,QAAQ;IACR,MAAM;IACN,WAAW;IACX,UAAU;IACV,aAAa;IACb,YAAY;IACZ,kBAAkB;IAClB,mBAAmB;CACnB,CAAC,CAAC;AAEH,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAAC,CAAC,MAAM,EAAE,MAAM,EAAE,SAAS,EAAE,KAAK,EAAE,aAAa,EAAE,OAAO,CAAC,CAAC,CAAC;AAE7F,MAAM,kBAAkB,GAAG,IAAI,GAAG,CAAC;IAClC,KAAK;IACL,KAAK;IACL,MAAM;IACN,IAAI;IACJ,IAAI;IACJ,KAAK;IACL,KAAK;IACL,MAAM;IACN,MAAM;IACN,IAAI;IACJ,IAAI;IACJ,MAAM;IACN,KAAK;IACL,MAAM;IACN,KAAK;IACL,IAAI;IACJ,KAAK;IACL,MAAM;IACN,MAAM;IACN,KAAK;IACL,IAAI;IACJ,OAAO;IACP,MAAM;CACN,CAAC,CAAC;AAEH,MAAM,yBAAyB,GAAG,IAAI,GAAG,CAAC,CAAC,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,WAAW,EAAE,MAAM,EAAE,QAAQ,EAAE,KAAK,CAAC,CAAC,CAAC;AAC3G,MAAM,yBAAyB,GAAG,IAAI,GAAG,CAAC,CAAC,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,QAAQ,CAAC,CAAC,CAAC;AAChG,MAAM,uBAAuB,GAC5B,qUAAqU,CAAC;AACvU,MAAM,qBAAqB,GAC1B,+IAA+I,CAAC;AAEjJ,SAAS,eAAe,CAAC,IAAa,EAAsB;IAC3D,IAAI,CAAC,IAAI,IAAI,OAAO,IAAI,KAAK,QAAQ;QAAE,OAAO,SAAS,CAAC;IACxD,MAAM,MAAM,GAAG,IAA+B,CAAC;IAC/C,MAAM,OAAO,GAAG,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC,GAAG,IAAI,MAAM,CAAC,YAAY,CAAC;IACpE,OAAO,OAAO,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;AAAA,CAChE;AAED,SAAS,WAAW,CAAC,OAAe,EAAsB;IACzD,MAAM,KAAK,GAAG,OAAO,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,mBAAmB,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;IAC7D,IAAI,CAAC,KAAK;QAAE,OAAO,SAAS,CAAC;IAC7B,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IAC/B,OAAO,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC;AAAA,CAC9C;AAED,SAAS,UAAU,CAAC,OAAe,EAAE,KAAa,EAAsB;IACvE,OAAO,OAAO,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,EAAE,WAAW,EAAE,CAAC;AAAA,CACzD;AAED,SAAS,sBAAsB,CAAC,OAAe,EAAW;IACzD,MAAM,IAAI,GAAG,WAAW,CAAC,OAAO,CAAC,CAAC;IAClC,IAAI,CAAC,IAAI,IAAI,CAAC,kBAAkB,CAAC,GAAG,CAAC,IAAI,CAAC;QAAE,OAAO,KAAK,CAAC;IACzD,IAAI,IAAI,KAAK,KAAK,EAAE,CAAC;QACpB,MAAM,UAAU,GAAG,UAAU,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QAC1C,OAAO,OAAO,CAAC,UAAU,IAAI,yBAAyB,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC;IACzE,CAAC;IACD,IAAI,IAAI,KAAK,KAAK,IAAI,IAAI,KAAK,MAAM,IAAI,IAAI,KAAK,MAAM,EAAE,CAAC;QAC1D,MAAM,UAAU,GAAG,UAAU,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QAC1C,OAAO,OAAO,CAAC,UAAU,IAAI,yBAAyB,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC;IACzE,CAAC;IACD,OAAO,IAAI,CAAC;AAAA,CACZ;AAED,SAAS,sBAAsB,CAAC,OAAe,EAAW;IACzD,IAAI,CAAC,OAAO,IAAI,uBAAuB,CAAC,IAAI,CAAC,OAAO,CAAC;QAAE,OAAO,KAAK,CAAC;IACpE,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,CAAC,4BAA4B,CAAC,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;IAC9F,OAAO,QAAQ,CAAC,MAAM,GAAG,CAAC,IAAI,QAAQ,CAAC,KAAK,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,IAAI,sBAAsB,CAAC,OAAO,CAAC,CAAC,CAAC;AAAA,CACjH;AAED,MAAM,UAAU,6BAA6B,CAAC,OAM7C,EAAW;IACX,IAAI,OAAO,CAAC,IAAI,KAAK,OAAO;QAAE,OAAO,KAAK,CAAC;IAC3C,MAAM,QAAQ,GAAG,OAAO,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IACvD,8FAA8F;IAC9F,8FAA8F;IAC9F,yFAAyF;IACzF,6CAA6C;IAC7C,IAAI,OAAO,CAAC,UAAU,KAAK,iBAAiB,IAAI,QAAQ,KAAK,oBAAoB;QAAE,OAAO,KAAK,CAAC;IAChG,IAAI,CAAC,QAAQ;QAAE,OAAO,IAAI,CAAC;IAC3B,IAAI,oBAAoB,CAAC,GAAG,CAAC,QAAQ,CAAC;QAAE,OAAO,KAAK,CAAC;IACrD,IAAI,gBAAgB,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;QACpC,MAAM,OAAO,GAAG,eAAe,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;QAC9C,OAAO,OAAO,CAAC,CAAC,CAAC,CAAC,sBAAsB,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAC1D,CAAC;IACD,OAAO,qBAAqB,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,QAAQ,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC;AAAA,CAC7E","sourcesContent":["import type { ModelTier } from \"../autonomy/contracts.ts\";\n\nconst READ_ONLY_TOOL_NAMES = new Set([\n\t\"read\",\n\t\"grep\",\n\t\"find\",\n\t\"ls\",\n\t\"list\",\n\t\"search\",\n\t\"glob\",\n\t\"view_file\",\n\t\"list_dir\",\n\t\"grep_search\",\n\t\"search_web\",\n\t\"read_url_content\",\n\t\"read_browser_page\",\n]);\n\nconst SHELL_TOOL_NAMES = new Set([\"bash\", \"exec\", \"execute\", \"run\", \"run_command\", \"shell\"]);\n\nconst READ_ONLY_COMMANDS = new Set([\n\t\"awk\",\n\t\"cat\",\n\t\"date\",\n\t\"df\",\n\t\"du\",\n\t\"env\",\n\t\"git\",\n\t\"grep\",\n\t\"head\",\n\t\"jq\",\n\t\"ls\",\n\t\"node\",\n\t\"npm\",\n\t\"pnpm\",\n\t\"pwd\",\n\t\"rg\",\n\t\"sed\",\n\t\"tail\",\n\t\"test\",\n\t\"tsc\",\n\t\"wc\",\n\t\"which\",\n\t\"yarn\",\n]);\n\nconst READ_ONLY_GIT_SUBCOMMANDS = new Set([\"branch\", \"diff\", \"log\", \"rev-parse\", \"show\", \"status\", \"tag\"]);\nconst READ_ONLY_NPM_SUBCOMMANDS = new Set([\"info\", \"list\", \"ls\", \"outdated\", \"view\", \"whoami\"]);\nconst MUTATING_SHELL_TOKEN_RE =\n\t/(^|\\s)(>|>>|2>|&>|tee\\b|rm\\b|mv\\b|cp\\b|mkdir\\b|touch\\b|chmod\\b|chown\\b|install\\b|commit\\b|push\\b|publish\\b|deploy\\b|apply\\b|add\\b|checkout\\b|switch\\b|reset\\b|clean\\b|stash\\b|merge\\b|rebase\\b|npm\\s+(?:i|install|ci|update|publish|run)\\b|pnpm\\s+(?:i|install|update|publish|run)\\b|yarn\\s+(?:add|install|upgrade|publish|run)\\b)/i;\nconst MUTATING_TOOL_NAME_RE =\n\t/(bash|exec|execute|run|shell|write|edit|patch|replace|delete|remove|move|rename|create|mkdir|touch|install|commit|push|publish|deploy|apply)/i;\n\nfunction getShellCommand(args: unknown): string | undefined {\n\tif (!args || typeof args !== \"object\") return undefined;\n\tconst record = args as Record<string, unknown>;\n\tconst command = record.command ?? record.cmd ?? record.shellCommand;\n\treturn typeof command === \"string\" ? command.trim() : undefined;\n}\n\nfunction commandName(segment: string): string | undefined {\n\tconst first = segment.trim().match(/^[A-Za-z0-9_./-]+/)?.[0];\n\tif (!first) return undefined;\n\tconst parts = first.split(\"/\");\n\treturn parts[parts.length - 1]?.toLowerCase();\n}\n\nfunction commandArg(segment: string, index: number): string | undefined {\n\treturn segment.trim().split(/\\s+/)[index]?.toLowerCase();\n}\n\nfunction isReadOnlyShellSegment(segment: string): boolean {\n\tconst name = commandName(segment);\n\tif (!name || !READ_ONLY_COMMANDS.has(name)) return false;\n\tif (name === \"git\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_GIT_SUBCOMMANDS.has(subcommand));\n\t}\n\tif (name === \"npm\" || name === \"pnpm\" || name === \"yarn\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_NPM_SUBCOMMANDS.has(subcommand));\n\t}\n\treturn true;\n}\n\nfunction isReadOnlyShellCommand(command: string): boolean {\n\tif (!command || MUTATING_SHELL_TOKEN_RE.test(command)) return false;\n\tconst segments = command.split(/\\s*(?:&&|\\|\\||[;|\\r\\n])\\s*/).map((segment) => segment.trim());\n\treturn segments.length > 0 && segments.every((segment) => segment.length > 0 && isReadOnlyShellSegment(segment));\n}\n\nexport function shouldEscalateModelRouterTool(options: {\n\ttier: ModelTier;\n\ttoolName: string;\n\targs?: unknown;\n\t/** The route's reasonCode; executor-lane turns carry \"executor_direct\". */\n\treasonCode?: string;\n}): boolean {\n\tif (options.tier !== \"cheap\") return false;\n\tconst toolName = options.toolName.trim().toLowerCase();\n\t// Executor-lane turns (G16) exist to run exactly one tool: run_toolkit_script, which enforces\n\t// its own safety (danger confirmation, structural exit-code contract). Escalating on it would\n\t// abort every executor turn at the moment it does its job. Any OTHER mutating tool still\n\t// escalates to the expensive model as usual.\n\tif (options.reasonCode === \"executor_direct\" && toolName === \"run_toolkit_script\") return false;\n\tif (!toolName) return true;\n\tif (READ_ONLY_TOOL_NAMES.has(toolName)) return false;\n\tif (SHELL_TOOL_NAMES.has(toolName)) {\n\t\tconst command = getShellCommand(options.args);\n\t\treturn command ? !isReadOnlyShellCommand(command) : true;\n\t}\n\treturn MUTATING_TOOL_NAME_RE.test(toolName) || !toolName.startsWith(\"read_\");\n}\n"]}
+{"version":3,"file":"tool-escalation.js","sourceRoot":"","sources":["../../../src/core/model-router/tool-escalation.ts"],"names":[],"mappings":"AAEA,MAAM,oBAAoB,GAAG,IAAI,GAAG,CAAC;IACpC,MAAM;IACN,MAAM;IACN,MAAM;IACN,IAAI;IACJ,MAAM;IACN,QAAQ;IACR,MAAM;IACN,WAAW;IACX,UAAU;IACV,aAAa;IACb,YAAY;IACZ,kBAAkB;IAClB,mBAAmB;CACnB,CAAC,CAAC;AAEH,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAAC,CAAC,MAAM,EAAE,MAAM,EAAE,SAAS,EAAE,KAAK,EAAE,aAAa,EAAE,OAAO,CAAC,CAAC,CAAC;AAE7F,MAAM,kBAAkB,GAAG,IAAI,GAAG,CAAC;IAClC,KAAK;IACL,KAAK;IACL,MAAM;IACN,IAAI;IACJ,IAAI;IACJ,KAAK;IACL,KAAK;IACL,MAAM;IACN,MAAM;IACN,IAAI;IACJ,IAAI;IACJ,MAAM;IACN,KAAK;IACL,MAAM;IACN,KAAK;IACL,IAAI;IACJ,KAAK;IACL,MAAM;IACN,MAAM;IACN,KAAK;IACL,IAAI;IACJ,OAAO;IACP,MAAM;CACN,CAAC,CAAC;AAEH,MAAM,yBAAyB,GAAG,IAAI,GAAG,CAAC,CAAC,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,WAAW,EAAE,MAAM,EAAE,QAAQ,EAAE,KAAK,CAAC,CAAC,CAAC;AAC3G,MAAM,yBAAyB,GAAG,IAAI,GAAG,CAAC,CAAC,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,QAAQ,CAAC,CAAC,CAAC;AAChG,MAAM,uBAAuB,GAC5B,qUAAqU,CAAC;AACvU,MAAM,gCAAgC,GAAG,sDAAsD,CAAC;AAChG,MAAM,qBAAqB,GAC1B,+IAA+I,CAAC;AAEjJ,SAAS,eAAe,CAAC,IAAa,EAAsB;IAC3D,IAAI,CAAC,IAAI,IAAI,OAAO,IAAI,KAAK,QAAQ;QAAE,OAAO,SAAS,CAAC;IACxD,MAAM,MAAM,GAAG,IAA+B,CAAC;IAC/C,MAAM,OAAO,GAAG,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC,GAAG,IAAI,MAAM,CAAC,YAAY,CAAC;IACpE,OAAO,OAAO,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;AAAA,CAChE;AAED,SAAS,WAAW,CAAC,OAAe,EAAsB;IACzD,MAAM,KAAK,GAAG,OAAO,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,mBAAmB,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;IAC7D,IAAI,CAAC,KAAK;QAAE,OAAO,SAAS,CAAC;IAC7B,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IAC/B,OAAO,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC;AAAA,CAC9C;AAED,SAAS,UAAU,CAAC,OAAe,EAAE,KAAa,EAAsB;IACvE,OAAO,OAAO,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,EAAE,WAAW,EAAE,CAAC;AAAA,CACzD;AAED,SAAS,sBAAsB,CAAC,OAAe,EAAW;IACzD,MAAM,IAAI,GAAG,WAAW,CAAC,OAAO,CAAC,CAAC;IAClC,IAAI,CAAC,IAAI,IAAI,CAAC,kBAAkB,CAAC,GAAG,CAAC,IAAI,CAAC;QAAE,OAAO,KAAK,CAAC;IACzD,IAAI,IAAI,KAAK,KAAK,EAAE,CAAC;QACpB,MAAM,UAAU,GAAG,UAAU,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QAC1C,OAAO,OAAO,CAAC,UAAU,IAAI,yBAAyB,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC;IACzE,CAAC;IACD,IAAI,IAAI,KAAK,KAAK,IAAI,IAAI,KAAK,MAAM,IAAI,IAAI,KAAK,MAAM,EAAE,CAAC;QAC1D,MAAM,UAAU,GAAG,UAAU,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QAC1C,OAAO,OAAO,CAAC,UAAU,IAAI,yBAAyB,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC;IACzE,CAAC;IACD,OAAO,IAAI,CAAC;AAAA,CACZ;AAED,SAAS,sBAAsB,CAAC,OAAe,EAAW;IACzD,IAAI,CAAC,OAAO,IAAI,uBAAuB,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,gCAAgC,CAAC,IAAI,CAAC,OAAO,CAAC;QACtG,OAAO,KAAK,CAAC;IACd,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,CAAC,4BAA4B,CAAC,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;IAC9F,OAAO,QAAQ,CAAC,MAAM,GAAG,CAAC,IAAI,QAAQ,CAAC,KAAK,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,IAAI,sBAAsB,CAAC,OAAO,CAAC,CAAC,CAAC;AAAA,CACjH;AAED,MAAM,UAAU,6BAA6B,CAAC,OAM7C,EAAW;IACX,IAAI,OAAO,CAAC,IAAI,KAAK,OAAO;QAAE,OAAO,KAAK,CAAC;IAC3C,MAAM,QAAQ,GAAG,OAAO,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IACvD,8FAA8F;IAC9F,8FAA8F;IAC9F,yFAAyF;IACzF,6CAA6C;IAC7C,IAAI,OAAO,CAAC,UAAU,KAAK,iBAAiB,IAAI,QAAQ,KAAK,oBAAoB;QAAE,OAAO,KAAK,CAAC;IAChG,IAAI,CAAC,QAAQ;QAAE,OAAO,IAAI,CAAC;IAC3B,IAAI,oBAAoB,CAAC,GAAG,CAAC,QAAQ,CAAC;QAAE,OAAO,KAAK,CAAC;IACrD,IAAI,gBAAgB,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;QACpC,MAAM,OAAO,GAAG,eAAe,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;QAC9C,OAAO,OAAO,CAAC,CAAC,CAAC,CAAC,sBAAsB,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAC1D,CAAC;IACD,OAAO,qBAAqB,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,QAAQ,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC;AAAA,CAC7E","sourcesContent":["import type { ModelTier } from \"../autonomy/contracts.ts\";\n\nconst READ_ONLY_TOOL_NAMES = new Set([\n\t\"read\",\n\t\"grep\",\n\t\"find\",\n\t\"ls\",\n\t\"list\",\n\t\"search\",\n\t\"glob\",\n\t\"view_file\",\n\t\"list_dir\",\n\t\"grep_search\",\n\t\"search_web\",\n\t\"read_url_content\",\n\t\"read_browser_page\",\n]);\n\nconst SHELL_TOOL_NAMES = new Set([\"bash\", \"exec\", \"execute\", \"run\", \"run_command\", \"shell\"]);\n\nconst READ_ONLY_COMMANDS = new Set([\n\t\"awk\",\n\t\"cat\",\n\t\"date\",\n\t\"df\",\n\t\"du\",\n\t\"env\",\n\t\"git\",\n\t\"grep\",\n\t\"head\",\n\t\"jq\",\n\t\"ls\",\n\t\"node\",\n\t\"npm\",\n\t\"pnpm\",\n\t\"pwd\",\n\t\"rg\",\n\t\"sed\",\n\t\"tail\",\n\t\"test\",\n\t\"tsc\",\n\t\"wc\",\n\t\"which\",\n\t\"yarn\",\n]);\n\nconst READ_ONLY_GIT_SUBCOMMANDS = new Set([\"branch\", \"diff\", \"log\", \"rev-parse\", \"show\", \"status\", \"tag\"]);\nconst READ_ONLY_NPM_SUBCOMMANDS = new Set([\"info\", \"list\", \"ls\", \"outdated\", \"view\", \"whoami\"]);\nconst MUTATING_SHELL_TOKEN_RE =\n\t/(^|\\s)(>|>>|2>|&>|tee\\b|rm\\b|mv\\b|cp\\b|mkdir\\b|touch\\b|chmod\\b|chown\\b|install\\b|commit\\b|push\\b|publish\\b|deploy\\b|apply\\b|add\\b|checkout\\b|switch\\b|reset\\b|clean\\b|stash\\b|merge\\b|rebase\\b|npm\\s+(?:i|install|ci|update|publish|run)\\b|pnpm\\s+(?:i|install|update|publish|run)\\b|yarn\\s+(?:add|install|upgrade|publish|run)\\b)/i;\nconst UNSAFE_NESTED_SHELL_EXECUTION_RE = /(`|\\$\\(|\\bfind\\b[\\s\\S]*\\s-exec(?:dir)?\\b|\\bxargs\\b)/i;\nconst MUTATING_TOOL_NAME_RE =\n\t/(bash|exec|execute|run|shell|write|edit|patch|replace|delete|remove|move|rename|create|mkdir|touch|install|commit|push|publish|deploy|apply)/i;\n\nfunction getShellCommand(args: unknown): string | undefined {\n\tif (!args || typeof args !== \"object\") return undefined;\n\tconst record = args as Record<string, unknown>;\n\tconst command = record.command ?? record.cmd ?? record.shellCommand;\n\treturn typeof command === \"string\" ? command.trim() : undefined;\n}\n\nfunction commandName(segment: string): string | undefined {\n\tconst first = segment.trim().match(/^[A-Za-z0-9_./-]+/)?.[0];\n\tif (!first) return undefined;\n\tconst parts = first.split(\"/\");\n\treturn parts[parts.length - 1]?.toLowerCase();\n}\n\nfunction commandArg(segment: string, index: number): string | undefined {\n\treturn segment.trim().split(/\\s+/)[index]?.toLowerCase();\n}\n\nfunction isReadOnlyShellSegment(segment: string): boolean {\n\tconst name = commandName(segment);\n\tif (!name || !READ_ONLY_COMMANDS.has(name)) return false;\n\tif (name === \"git\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_GIT_SUBCOMMANDS.has(subcommand));\n\t}\n\tif (name === \"npm\" || name === \"pnpm\" || name === \"yarn\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_NPM_SUBCOMMANDS.has(subcommand));\n\t}\n\treturn true;\n}\n\nfunction isReadOnlyShellCommand(command: string): boolean {\n\tif (!command || MUTATING_SHELL_TOKEN_RE.test(command) || UNSAFE_NESTED_SHELL_EXECUTION_RE.test(command))\n\t\treturn false;\n\tconst segments = command.split(/\\s*(?:&&|\\|\\||[;|\\r\\n])\\s*/).map((segment) => segment.trim());\n\treturn segments.length > 0 && segments.every((segment) => segment.length > 0 && isReadOnlyShellSegment(segment));\n}\n\nexport function shouldEscalateModelRouterTool(options: {\n\ttier: ModelTier;\n\ttoolName: string;\n\targs?: unknown;\n\t/** The route's reasonCode; executor-lane turns carry \"executor_direct\". */\n\treasonCode?: string;\n}): boolean {\n\tif (options.tier !== \"cheap\") return false;\n\tconst toolName = options.toolName.trim().toLowerCase();\n\t// Executor-lane turns (G16) exist to run exactly one tool: run_toolkit_script, which enforces\n\t// its own safety (danger confirmation, structural exit-code contract). Escalating on it would\n\t// abort every executor turn at the moment it does its job. Any OTHER mutating tool still\n\t// escalates to the expensive model as usual.\n\tif (options.reasonCode === \"executor_direct\" && toolName === \"run_toolkit_script\") return false;\n\tif (!toolName) return true;\n\tif (READ_ONLY_TOOL_NAMES.has(toolName)) return false;\n\tif (SHELL_TOOL_NAMES.has(toolName)) {\n\t\tconst command = getShellCommand(options.args);\n\t\treturn command ? !isReadOnlyShellCommand(command) : true;\n\t}\n\treturn MUTATING_TOOL_NAME_RE.test(toolName) || !toolName.startsWith(\"read_\");\n}\n"]}