niceeval 0.12.1 → 0.12.2-canary.31

package/dist/o11y/record/validation.d.mts

@@ -0,0 +1,62 @@
+import { type AttemptObservabilityFamilySchemaIdV1, type AttemptReferenceTargetV1, type AttemptReferencesForFamilyV1, type ObservabilityEntityIdV1, type ObservabilityOwnerV1, type ObservabilityReferenceTargetV1, type RunObservabilityFamilySchemaIdV1, type RunReferenceTargetV1, type RunReferencesForFamilyV1 } from "./model.mjs";
+import { type ObservabilityRecordContractError } from "./errors.mjs";
+declare const observabilityFamilyValidationTypeId: unique symbol;
+/**
+ * A package-created proof that one family has completed its local validation.
+ * Payloads remain in their owning family; this proof carries only entity and
+ * direct-reference facts needed by the whole-Run contract.
+ */
+export interface ObservabilityFamilyValidationV1<Owner extends ObservabilityOwnerV1> {
+    readonly [observabilityFamilyValidationTypeId]: () => Owner;
+}
+export interface AttemptReferenceSourceV1<Family extends AttemptObservabilityFamilySchemaIdV1> {
+    readonly sourceId: ObservabilityEntityIdV1;
+    readonly refs: readonly AttemptReferencesForFamilyV1<Family>[];
+}
+export interface RunReferenceSourceV1<Family extends RunObservabilityFamilySchemaIdV1> {
+    readonly sourceId: ObservabilityEntityIdV1;
+    readonly refs: readonly RunReferencesForFamilyV1<Family>[];
+}
+export interface AttemptObservabilityFamilyValidationInputV1<Family extends AttemptObservabilityFamilySchemaIdV1> {
+    readonly schemaId: Family;
+    readonly entities: readonly AttemptReferenceTargetV1[];
+    readonly references: readonly AttemptReferenceSourceV1<Family>[];
+    /** Family-local timing/source-frame errors use the stable aggregate union. */
+    readonly localErrors?: readonly ObservabilityRecordContractError[];
+}
+export interface RunObservabilityFamilyValidationInputV1<Family extends RunObservabilityFamilySchemaIdV1> {
+    readonly schemaId: Family;
+    readonly entities: readonly RunReferenceTargetV1[];
+    readonly references: readonly RunReferenceSourceV1<Family>[];
+    /** Family-local timing/source-frame errors use the stable aggregate union. */
+    readonly localErrors?: readonly ObservabilityRecordContractError[];
+}
+export declare const REQUIRED_ATTEMPT_OBSERVABILITY_FAMILIES_V1: readonly ["niceeval.conversation/v1", "niceeval.commands/v1", "niceeval.usage/v1", "niceeval.timing/v1", "niceeval.diagnostics/v1"];
+export declare const REQUIRED_RUN_OBSERVABILITY_FAMILIES_V1: readonly ["niceeval.timing/v1", "niceeval.diagnostics/v1"];
+export declare function compareObservabilityFamilySchemaIdV1(left: string, right: string): number;
+export declare function compareObservabilityEntityTargetV1(left: ObservabilityReferenceTargetV1, right: ObservabilityReferenceTargetV1): number;
+/** A pure convenience for family encoders that need a canonical entity order. */
+export declare function sortObservabilityEntityTargetsV1(targets: readonly ObservabilityReferenceTargetV1[]): readonly ObservabilityReferenceTargetV1[];
+export declare function compareObservabilityRecordContractErrorV1(left: ObservabilityRecordContractError, right: ObservabilityRecordContractError): number;
+export declare function sortObservabilityRecordContractErrorsV1(errors: readonly ObservabilityRecordContractError[]): readonly ObservabilityRecordContractError[];
+export declare function validateAttemptObservabilityFamilyV1<Family extends AttemptObservabilityFamilySchemaIdV1>(input: AttemptObservabilityFamilyValidationInputV1<Family>): readonly ObservabilityRecordContractError[];
+export declare function validateRunObservabilityFamilyV1<Family extends RunObservabilityFamilySchemaIdV1>(input: RunObservabilityFamilyValidationInputV1<Family>): readonly ObservabilityRecordContractError[];
+export declare function makeAttemptObservabilityFamilyValidationV1<Family extends AttemptObservabilityFamilySchemaIdV1>(input: AttemptObservabilityFamilyValidationInputV1<Family>): ObservabilityFamilyValidationV1<"attempt">;
+export declare function makeRunObservabilityFamilyValidationV1<Family extends RunObservabilityFamilySchemaIdV1>(input: RunObservabilityFamilyValidationInputV1<Family>): ObservabilityFamilyValidationV1<"run">;
+export interface AttemptObservabilityContractInputV1 {
+    readonly families: readonly ObservabilityFamilyValidationV1<"attempt">[];
+}
+export interface RunObservabilityContractInputV1 {
+    readonly families: readonly ObservabilityFamilyValidationV1<"run">[];
+}
+export interface ObservabilityRecordContractValidationInputV1 {
+    readonly run: RunObservabilityContractInputV1;
+    readonly attempts: readonly AttemptObservabilityContractInputV1[];
+}
+/**
+ * Pure aggregate preflight for the official five Attempt and two Run families.
+ * It validates only durable official facts; opening a writer, emitting blobs,
+ * and publishing the complete marker remain Record's separate responsibilities.
+ */
+export declare function validateObservabilityRecordContractV1(input: ObservabilityRecordContractValidationInputV1): readonly ObservabilityRecordContractError[];
+export {};

package/dist/o11y/record/validation.d.ts

@@ -0,0 +1,62 @@
+import { type AttemptObservabilityFamilySchemaIdV1, type AttemptReferenceTargetV1, type AttemptReferencesForFamilyV1, type ObservabilityEntityIdV1, type ObservabilityOwnerV1, type ObservabilityReferenceTargetV1, type RunObservabilityFamilySchemaIdV1, type RunReferenceTargetV1, type RunReferencesForFamilyV1 } from "./model.js";
+import { type ObservabilityRecordContractError } from "./errors.js";
+declare const observabilityFamilyValidationTypeId: unique symbol;
+/**
+ * A package-created proof that one family has completed its local validation.
+ * Payloads remain in their owning family; this proof carries only entity and
+ * direct-reference facts needed by the whole-Run contract.
+ */
+export interface ObservabilityFamilyValidationV1<Owner extends ObservabilityOwnerV1> {
+    readonly [observabilityFamilyValidationTypeId]: () => Owner;
+}
+export interface AttemptReferenceSourceV1<Family extends AttemptObservabilityFamilySchemaIdV1> {
+    readonly sourceId: ObservabilityEntityIdV1;
+    readonly refs: readonly AttemptReferencesForFamilyV1<Family>[];
+}
+export interface RunReferenceSourceV1<Family extends RunObservabilityFamilySchemaIdV1> {
+    readonly sourceId: ObservabilityEntityIdV1;
+    readonly refs: readonly RunReferencesForFamilyV1<Family>[];
+}
+export interface AttemptObservabilityFamilyValidationInputV1<Family extends AttemptObservabilityFamilySchemaIdV1> {
+    readonly schemaId: Family;
+    readonly entities: readonly AttemptReferenceTargetV1[];
+    readonly references: readonly AttemptReferenceSourceV1<Family>[];
+    /** Family-local timing/source-frame errors use the stable aggregate union. */
+    readonly localErrors?: readonly ObservabilityRecordContractError[];
+}
+export interface RunObservabilityFamilyValidationInputV1<Family extends RunObservabilityFamilySchemaIdV1> {
+    readonly schemaId: Family;
+    readonly entities: readonly RunReferenceTargetV1[];
+    readonly references: readonly RunReferenceSourceV1<Family>[];
+    /** Family-local timing/source-frame errors use the stable aggregate union. */
+    readonly localErrors?: readonly ObservabilityRecordContractError[];
+}
+export declare const REQUIRED_ATTEMPT_OBSERVABILITY_FAMILIES_V1: readonly ["niceeval.conversation/v1", "niceeval.commands/v1", "niceeval.usage/v1", "niceeval.timing/v1", "niceeval.diagnostics/v1"];
+export declare const REQUIRED_RUN_OBSERVABILITY_FAMILIES_V1: readonly ["niceeval.timing/v1", "niceeval.diagnostics/v1"];
+export declare function compareObservabilityFamilySchemaIdV1(left: string, right: string): number;
+export declare function compareObservabilityEntityTargetV1(left: ObservabilityReferenceTargetV1, right: ObservabilityReferenceTargetV1): number;
+/** A pure convenience for family encoders that need a canonical entity order. */
+export declare function sortObservabilityEntityTargetsV1(targets: readonly ObservabilityReferenceTargetV1[]): readonly ObservabilityReferenceTargetV1[];
+export declare function compareObservabilityRecordContractErrorV1(left: ObservabilityRecordContractError, right: ObservabilityRecordContractError): number;
+export declare function sortObservabilityRecordContractErrorsV1(errors: readonly ObservabilityRecordContractError[]): readonly ObservabilityRecordContractError[];
+export declare function validateAttemptObservabilityFamilyV1<Family extends AttemptObservabilityFamilySchemaIdV1>(input: AttemptObservabilityFamilyValidationInputV1<Family>): readonly ObservabilityRecordContractError[];
+export declare function validateRunObservabilityFamilyV1<Family extends RunObservabilityFamilySchemaIdV1>(input: RunObservabilityFamilyValidationInputV1<Family>): readonly ObservabilityRecordContractError[];
+export declare function makeAttemptObservabilityFamilyValidationV1<Family extends AttemptObservabilityFamilySchemaIdV1>(input: AttemptObservabilityFamilyValidationInputV1<Family>): ObservabilityFamilyValidationV1<"attempt">;
+export declare function makeRunObservabilityFamilyValidationV1<Family extends RunObservabilityFamilySchemaIdV1>(input: RunObservabilityFamilyValidationInputV1<Family>): ObservabilityFamilyValidationV1<"run">;
+export interface AttemptObservabilityContractInputV1 {
+    readonly families: readonly ObservabilityFamilyValidationV1<"attempt">[];
+}
+export interface RunObservabilityContractInputV1 {
+    readonly families: readonly ObservabilityFamilyValidationV1<"run">[];
+}
+export interface ObservabilityRecordContractValidationInputV1 {
+    readonly run: RunObservabilityContractInputV1;
+    readonly attempts: readonly AttemptObservabilityContractInputV1[];
+}
+/**
+ * Pure aggregate preflight for the official five Attempt and two Run families.
+ * It validates only durable official facts; opening a writer, emitting blobs,
+ * and publishing the complete marker remain Record's separate responsibilities.
+ */
+export declare function validateObservabilityRecordContractV1(input: ObservabilityRecordContractValidationInputV1): readonly ObservabilityRecordContractError[];
+export {};

package/dist/o11y/record/validation.js

@@ -0,0 +1,28 @@
+import { createRequire as __niceevalCreateRequire } from "node:module";
+const __niceevalRequire = __niceevalCreateRequire(import.meta.url);
+const __niceevalCanonical = __niceevalRequire("./validation.cjs");
+const __niceevalExport0 = __niceevalCanonical["REQUIRED_ATTEMPT_OBSERVABILITY_FAMILIES_V1"];
+export { __niceevalExport0 as REQUIRED_ATTEMPT_OBSERVABILITY_FAMILIES_V1 };
+const __niceevalExport1 = __niceevalCanonical["REQUIRED_RUN_OBSERVABILITY_FAMILIES_V1"];
+export { __niceevalExport1 as REQUIRED_RUN_OBSERVABILITY_FAMILIES_V1 };
+const __niceevalExport2 = __niceevalCanonical["compareObservabilityEntityTargetV1"];
+export { __niceevalExport2 as compareObservabilityEntityTargetV1 };
+const __niceevalExport3 = __niceevalCanonical["compareObservabilityFamilySchemaIdV1"];
+export { __niceevalExport3 as compareObservabilityFamilySchemaIdV1 };
+const __niceevalExport4 = __niceevalCanonical["compareObservabilityRecordContractErrorV1"];
+export { __niceevalExport4 as compareObservabilityRecordContractErrorV1 };
+const __niceevalExport5 = __niceevalCanonical["makeAttemptObservabilityFamilyValidationV1"];
+export { __niceevalExport5 as makeAttemptObservabilityFamilyValidationV1 };
+const __niceevalExport6 = __niceevalCanonical["makeRunObservabilityFamilyValidationV1"];
+export { __niceevalExport6 as makeRunObservabilityFamilyValidationV1 };
+const __niceevalExport7 = __niceevalCanonical["sortObservabilityEntityTargetsV1"];
+export { __niceevalExport7 as sortObservabilityEntityTargetsV1 };
+const __niceevalExport8 = __niceevalCanonical["sortObservabilityRecordContractErrorsV1"];
+export { __niceevalExport8 as sortObservabilityRecordContractErrorsV1 };
+const __niceevalExport9 = __niceevalCanonical["validateAttemptObservabilityFamilyV1"];
+export { __niceevalExport9 as validateAttemptObservabilityFamilyV1 };
+const __niceevalExport10 = __niceevalCanonical["validateObservabilityRecordContractV1"];
+export { __niceevalExport10 as validateObservabilityRecordContractV1 };
+const __niceevalExport11 = __niceevalCanonical["validateRunObservabilityFamilyV1"];
+export { __niceevalExport11 as validateRunObservabilityFamilyV1 };
+//# sourceMappingURL=validation.js.map

package/dist/o11y/record/validation.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"validation.js","sources":[],"names":[],"mappings":""}

package/dist/o11y/tool-names.cjs

@@ -0,0 +1,68 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.GENERIC_VERB_ALIASES = void 0;
+exports.normalizeToolName = normalizeToolName;
+/**
+ * 通用别名基表(键一律小写;normalizeToolName 匹配前先 toLowerCase)。
+ * 只收「不会撞上被测应用域内工具名」的保守别名(多为带下划线的复合名);
+ * `search`/`run`/`fetch` 这类裸动词在 GENERIC_VERB_ALIASES,由确知自己 transcript
+ * 词汇的 parser 显式 opt-in —— 否则 AI SDK 应用里一个叫 `search` 的商品搜索工具
+ * 会被误归成 web_search,负断言语义被悄悄改写。
+ */
+const BASE_TOOL_ALIASES = {
+    // 文件
+    read_file: "file_read",
+    write_file: "file_write",
+    create_file: "file_write",
+    delete_file: "file_write",
+    edit_file: "file_edit",
+    patch_file: "file_edit",
+    apply_patch: "file_edit",
+    str_replace_editor: "file_edit",
+    // shell
+    shell: "shell",
+    bash: "shell",
+    command_execution: "shell",
+    local_shell: "shell",
+    execute_command: "shell",
+    run_command: "shell",
+    // web
+    web_fetch: "web_fetch",
+    fetch_url: "web_fetch",
+    http_request: "web_fetch",
+    web_search: "web_search",
+    // 检索 / 导航
+    glob: "glob",
+    find_files: "glob",
+    list_files: "glob",
+    grep: "grep",
+    search_files: "grep",
+    ripgrep: "grep",
+    list_dir: "list_dir",
+    list_directory: "list_dir",
+};
+/**
+ * 裸动词别名:coding-agent CLI 自己的 transcript 词汇(codex/bub/claude 的工具名
+ * 由 CLI 控制,不会和用户域名冲突)。对被测应用(如 AI SDK 的自定义工具)不适用。
+ */
+exports.GENERIC_VERB_ALIASES = {
+    exec: "shell",
+    execute: "shell",
+    run: "shell",
+    terminal: "shell",
+    fetch: "web_fetch",
+    curl: "web_fetch",
+    search: "web_search",
+    ls: "list_dir",
+    dir: "list_dir",
+    task: "agent_task",
+};
+/**
+ * 原始工具名 → 规范 ToolName。先查 agent 特有别名(overrides),再查通用基表,
+ * 都没有则 "unknown"。键匹配不区分大小写(Claude Code 的 PascalCase 名靠这里兜住)。
+ */
+function normalizeToolName(name, overrides) {
+    const key = name.toLowerCase();
+    return overrides?.[key] ?? BASE_TOOL_ALIASES[key] ?? "unknown";
+}
+//# sourceMappingURL=tool-names.cjs.map

package/dist/o11y/tool-names.cjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"tool-names.js","sourceRoot":"","sources":["tool-names.ts"],"names":[],"mappings":";;;AA+DA,8CAGC;AA9DD;;;;;;GAMG;AACH,MAAM,iBAAiB,GAAwC;IAC3D,KAAK;IACL,SAAS,EAAE,WAAW;IACtB,UAAU,EAAE,YAAY;IACxB,WAAW,EAAE,YAAY;IACzB,WAAW,EAAE,YAAY;IACzB,SAAS,EAAE,WAAW;IACtB,UAAU,EAAE,WAAW;IACvB,WAAW,EAAE,WAAW;IACxB,kBAAkB,EAAE,WAAW;IAC/B,QAAQ;IACR,KAAK,EAAE,OAAO;IACd,IAAI,EAAE,OAAO;IACb,iBAAiB,EAAE,OAAO;IAC1B,WAAW,EAAE,OAAO;IACpB,eAAe,EAAE,OAAO;IACxB,WAAW,EAAE,OAAO;IACpB,MAAM;IACN,SAAS,EAAE,WAAW;IACtB,SAAS,EAAE,WAAW;IACtB,YAAY,EAAE,WAAW;IACzB,UAAU,EAAE,YAAY;IACxB,UAAU;IACV,IAAI,EAAE,MAAM;IACZ,UAAU,EAAE,MAAM;IAClB,UAAU,EAAE,MAAM;IAClB,IAAI,EAAE,MAAM;IACZ,YAAY,EAAE,MAAM;IACpB,OAAO,EAAE,MAAM;IACf,QAAQ,EAAE,UAAU;IACpB,cAAc,EAAE,UAAU;CAC7B,CAAC;AACF;;;GAGG;AACU,QAAA,oBAAoB,GAAwC;IACrE,IAAI,EAAE,OAAO;IACb,OAAO,EAAE,OAAO;IAChB,GAAG,EAAE,OAAO;IACZ,QAAQ,EAAE,OAAO;IACjB,KAAK,EAAE,WAAW;IAClB,IAAI,EAAE,WAAW;IACjB,MAAM,EAAE,YAAY;IACpB,EAAE,EAAE,UAAU;IACd,GAAG,EAAE,UAAU;IACf,IAAI,EAAE,YAAY;CACrB,CAAC;AACF;;;GAGG;AACH,SAAgB,iBAAiB,CAAC,IAAY,EAAE,SAA+C;IAC3F,MAAM,GAAG,GAAG,IAAI,CAAC,WAAW,EAAE,CAAC;IAC/B,OAAO,SAAS,EAAE,CAAC,GAAG,CAAC,IAAI,iBAAiB,CAAC,GAAG,CAAC,IAAI,SAAS,CAAC;AACnE,CAAC","sourcesContent":["// 规范工具名归一的唯一出处。canonical ToolName 是跨 agent 断言(calledTool(\"file_read\"))\n// 的命中基础:基表收「跨 agent 一致」的通用别名,各 parser 只声明自己特有的别名叠加在\n// 基表上——新增一个 canonical 名只改这里,不必同步 4 张漂移的表。\nimport type { ToolName } from \"../types.ts\";\n/**\n * 通用别名基表(键一律小写;normalizeToolName 匹配前先 toLowerCase)。\n * 只收「不会撞上被测应用域内工具名」的保守别名(多为带下划线的复合名);\n * `search`/`run`/`fetch` 这类裸动词在 GENERIC_VERB_ALIASES,由确知自己 transcript\n * 词汇的 parser 显式 opt-in —— 否则 AI SDK 应用里一个叫 `search` 的商品搜索工具\n * 会被误归成 web_search,负断言语义被悄悄改写。\n */\nconst BASE_TOOL_ALIASES: globalThis.Record<string, ToolName> = {\n    // 文件\n    read_file: \"file_read\",\n    write_file: \"file_write\",\n    create_file: \"file_write\",\n    delete_file: \"file_write\",\n    edit_file: \"file_edit\",\n    patch_file: \"file_edit\",\n    apply_patch: \"file_edit\",\n    str_replace_editor: \"file_edit\",\n    // shell\n    shell: \"shell\",\n    bash: \"shell\",\n    command_execution: \"shell\",\n    local_shell: \"shell\",\n    execute_command: \"shell\",\n    run_command: \"shell\",\n    // web\n    web_fetch: \"web_fetch\",\n    fetch_url: \"web_fetch\",\n    http_request: \"web_fetch\",\n    web_search: \"web_search\",\n    // 检索 / 导航\n    glob: \"glob\",\n    find_files: \"glob\",\n    list_files: \"glob\",\n    grep: \"grep\",\n    search_files: \"grep\",\n    ripgrep: \"grep\",\n    list_dir: \"list_dir\",\n    list_directory: \"list_dir\",\n};\n/**\n * 裸动词别名:coding-agent CLI 自己的 transcript 词汇(codex/bub/claude 的工具名\n * 由 CLI 控制,不会和用户域名冲突)。对被测应用(如 AI SDK 的自定义工具)不适用。\n */\nexport const GENERIC_VERB_ALIASES: globalThis.Record<string, ToolName> = {\n    exec: \"shell\",\n    execute: \"shell\",\n    run: \"shell\",\n    terminal: \"shell\",\n    fetch: \"web_fetch\",\n    curl: \"web_fetch\",\n    search: \"web_search\",\n    ls: \"list_dir\",\n    dir: \"list_dir\",\n    task: \"agent_task\",\n};\n/**\n * 原始工具名 → 规范 ToolName。先查 agent 特有别名(overrides),再查通用基表,\n * 都没有则 \"unknown\"。键匹配不区分大小写(Claude Code 的 PascalCase 名靠这里兜住)。\n */\nexport function normalizeToolName(name: string, overrides?: globalThis.Record<string, ToolName>): ToolName {\n    const key = name.toLowerCase();\n    return overrides?.[key] ?? BASE_TOOL_ALIASES[key] ?? \"unknown\";\n}\n"]}

package/dist/o11y/tool-names.d.cts

@@ -0,0 +1,11 @@
+import type { ToolName } from "../types.cjs";
+/**
+ * 裸动词别名:coding-agent CLI 自己的 transcript 词汇(codex/bub/claude 的工具名
+ * 由 CLI 控制,不会和用户域名冲突)。对被测应用(如 AI SDK 的自定义工具)不适用。
+ */
+export declare const GENERIC_VERB_ALIASES: globalThis.Record<string, ToolName>;
+/**
+ * 原始工具名 → 规范 ToolName。先查 agent 特有别名(overrides),再查通用基表,
+ * 都没有则 "unknown"。键匹配不区分大小写(Claude Code 的 PascalCase 名靠这里兜住)。
+ */
+export declare function normalizeToolName(name: string, overrides?: globalThis.Record<string, ToolName>): ToolName;

package/dist/o11y/tool-names.d.mts

@@ -0,0 +1,11 @@
+import type { ToolName } from "../types.mjs";
+/**
+ * 裸动词别名:coding-agent CLI 自己的 transcript 词汇(codex/bub/claude 的工具名
+ * 由 CLI 控制,不会和用户域名冲突)。对被测应用(如 AI SDK 的自定义工具)不适用。
+ */
+export declare const GENERIC_VERB_ALIASES: globalThis.Record<string, ToolName>;
+/**
+ * 原始工具名 → 规范 ToolName。先查 agent 特有别名(overrides),再查通用基表,
+ * 都没有则 "unknown"。键匹配不区分大小写(Claude Code 的 PascalCase 名靠这里兜住)。
+ */
+export declare function normalizeToolName(name: string, overrides?: globalThis.Record<string, ToolName>): ToolName;

package/dist/o11y/tool-names.d.ts

@@ -0,0 +1,11 @@
+import type { ToolName } from "../types.js";
+/**
+ * 裸动词别名:coding-agent CLI 自己的 transcript 词汇(codex/bub/claude 的工具名
+ * 由 CLI 控制,不会和用户域名冲突)。对被测应用(如 AI SDK 的自定义工具)不适用。
+ */
+export declare const GENERIC_VERB_ALIASES: globalThis.Record<string, ToolName>;
+/**
+ * 原始工具名 → 规范 ToolName。先查 agent 特有别名(overrides),再查通用基表,
+ * 都没有则 "unknown"。键匹配不区分大小写(Claude Code 的 PascalCase 名靠这里兜住)。
+ */
+export declare function normalizeToolName(name: string, overrides?: globalThis.Record<string, ToolName>): ToolName;

package/dist/o11y/tool-names.js

@@ -0,0 +1,8 @@
+import { createRequire as __niceevalCreateRequire } from "node:module";
+const __niceevalRequire = __niceevalCreateRequire(import.meta.url);
+const __niceevalCanonical = __niceevalRequire("./tool-names.cjs");
+const __niceevalExport0 = __niceevalCanonical["GENERIC_VERB_ALIASES"];
+export { __niceevalExport0 as GENERIC_VERB_ALIASES };
+const __niceevalExport1 = __niceevalCanonical["normalizeToolName"];
+export { __niceevalExport1 as normalizeToolName };
+//# sourceMappingURL=tool-names.js.map

package/dist/o11y/tool-names.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"tool-names.js","sources":[],"names":[],"mappings":""}

package/dist/o11y/types.cjs

@@ -0,0 +1,3 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+//# sourceMappingURL=types.cjs.map

package/dist/o11y/types.cjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.js","sourceRoot":"","sources":["types.ts"],"names":[],"mappings":"","sourcesContent":["// o11y 域类型:标准事件流(StreamEvent)、折叠事实(DerivedFacts)、\n// OTLP trace(TraceSpan / SpanKind)与用量 / 摘要。\nimport type { JsonValue, SourceLoc } from \"../shared/types.ts\";\n/**\n * 一次运行的 token 用量(沙箱型从 transcript/OTel span 的 `gen_ai.usage.*` 属性抠,remote 由\n * send 的 `Turn.usage` 直接返回)。每个字段只在协议真实提供该值时存在——原始协议没有 usage 时\n * 省略,不编造数值;不存在「默认 0」或「默认 1」的字段。三个输入侧桶恒互斥:相加才是送进模型的\n * 完整上下文量;把协议原生口径归一到互斥是 adapter 的落值义务——OpenAI 系报「含缓存的输入总量 +\n * 缓存命中子集」,落 inputTokens 前先扣掉子集(docs/feature/record/architecture.md#usage,\n * 各协议明细见 docs/feature/adapters/sdk/<name>/cost.md)。\n */\nexport interface Usage {\n    /** 未命中缓存、按全价计费的输入 token;与两个 cache 桶互斥。 */\n    inputTokens?: number;\n    /** 输出(completion)token 数。 */\n    outputTokens?: number;\n    /** 从提示缓存命中的输入 token;独立计价桶,不包含在 inputTokens 里(省略表示该 agent 不上报此项)。 */\n    cacheReadTokens?: number;\n    /** 写入提示缓存的输入 token;独立计价桶,不包含在 inputTokens 里(省略表示该 agent 不上报此项)。 */\n    cacheCreationTokens?: number;\n    /** 推理(thinking)token 数,outputTokens 的已含明细,单列展示用;只在协议真实提供时存在。 */\n    reasoningTokens?: number;\n    /** 真实发生的模型请求数。协议不提供请求计数就省略,绝不写 1 凑数。 */\n    requests?: number;\n    /**\n     * 网关/adapter 实测的真实美元成本(只能由 `Turn.usage.costUSD` 显式带回,从不从\n     * token 用量或 OTel span 反推得到)。与顶层 `estimatedCostUSD`(价目表估算)是两个事实:\n     * 存在时优先于按价格表(`defineConfig({ pricing })`)估算的成本——见 `estimateCost` 的\n     * `usage.costUSD ?? estimateCost(...)` 兜底顺序。\n     */\n    costUSD?: number;\n}\n/** 跨 agent 归一化后的规范工具名。 */\nexport type ToolName = \"file_read\" | \"file_write\" | \"file_edit\" | \"shell\" | \"web_fetch\" | \"web_search\" | \"glob\" | \"grep\" | \"list_dir\" | \"agent_task\" | \"unknown\";\n/** Adapter 确认原始 command 不能安全还原为 argv 的原因。 */\nexport type OriginalCommandOpaqueReason = \"redacted\" | \"truncated\" | \"compound-shell\" | \"dynamic-shell\" | \"unsupported-protocol\";\n/** Adapter 已确认的原始 command invocation；token 保留原生执行边界的形状。 */\nexport type OriginalCommandInvocation = {\n    readonly state: \"available\";\n    readonly executable: string;\n    readonly args: readonly string[];\n} | {\n    readonly state: \"opaque\";\n    readonly reason: OriginalCommandOpaqueReason;\n};\n/** Observation Protocol 唯一的 logical-command normalizer profile。 */\nexport type LogicalCommandNormalizer = \"logical-command/v1\";\n/** 可供 command matcher 消费的、已经归一的逻辑 invocation。 */\nexport type LogicalCommandInvocation = {\n    readonly state: \"available\";\n    readonly executable: string;\n    readonly args: readonly string[];\n    readonly normalizer: LogicalCommandNormalizer;\n    readonly normalization: \"identity\" | \"pnpm-exec\" | \"npx\";\n} | {\n    readonly state: \"opaque\";\n    readonly normalizer: LogicalCommandNormalizer;\n    readonly reason: \"original-opaque\";\n    readonly originalReason: OriginalCommandOpaqueReason;\n} | {\n    readonly state: \"opaque\";\n    readonly normalizer: LogicalCommandNormalizer;\n    readonly reason: \"unsupported-wrapper-form\" | \"ambiguous-wrapper-target\" | \"multiple-executions\";\n};\n/** 一笔 tool operation 的穷尽 command 分类；不从 tool name 或 input 推导。 */\nexport type CommandProjection = {\n    readonly kind: \"not-command\";\n} | {\n    readonly kind: \"command\";\n    readonly original: OriginalCommandInvocation;\n    readonly logical: LogicalCommandInvocation;\n};\n/** 一条 Agent session 内的稳定事件位置。 */\nexport interface EventPosition {\n    readonly turnOrdinal: number;\n    readonly eventOrdinal: number;\n}\n/** 单笔 tool occurrence 的可用 lifecycle 证据。 */\nexport type LogicalToolLifecycle = {\n    readonly state: \"available\";\n    readonly status: \"pending\";\n} | {\n    readonly state: \"available\";\n    readonly status: \"completed\" | \"failed\" | \"rejected\";\n    readonly finish: EventPosition;\n} | {\n    readonly state: \"opaque\";\n    readonly reason: \"partial-stream\" | \"missing-lifecycle-evidence\";\n};\n/**\n * A tool field is independently observable.  Input and output deliberately\n * use the same envelope so a matcher only becomes uncertain when it actually\n * asks for the field whose material is incomplete.\n */\nexport type LogicalToolEvidence = {\n    readonly state: \"complete\";\n    readonly value: JsonValue;\n} | {\n    readonly state: \"partial\";\n    readonly value: JsonValue;\n    readonly opaquePointers: readonly string[];\n    readonly reason: \"truncated\" | \"redacted\";\n} | {\n    readonly state: \"unavailable\";\n    readonly reason: string;\n};\n/** Fact matcher 消费的一笔、按 started/finished 关联后的 tool operation。 */\nexport interface LogicalToolOccurrence {\n    /** 由 session、turn 与 started event position 组成的 occurrence identity；不是 operationId。 */\n    readonly id: string;\n    readonly session: string;\n    readonly turn: string;\n    readonly name: {\n        readonly original: string;\n        readonly canonical?: ToolName;\n    };\n    readonly input: LogicalToolEvidence;\n    /**\n     * A completed lifecycle alone does not prove that output was collected.\n     * Missing, pending, and orphan output are explicit unavailable evidence.\n     */\n    readonly output: LogicalToolEvidence;\n    /**\n     * 旧 Adapter 尚未升级时允许缺失，调用方必须把它视作 command evidence unavailable，\n     * 不能回退为 not-command。新 Adapter 对每笔 tool operation 都必须提供分类。\n     */\n    readonly command?: CommandProjection;\n    readonly start: EventPosition;\n    readonly lifecycle: LogicalToolLifecycle;\n}\n/** `operation.finished` 找不到可信 started 时留下的协议诊断输入，不构造 occurrence。 */\nexport interface OrphanToolOperationFinish {\n    readonly session: string;\n    readonly turn: string;\n    readonly operationId: string;\n    readonly status: \"completed\" | \"failed\" | \"rejected\";\n    readonly position: EventPosition;\n}\n/** `deriveLogicalToolOccurrences` 的 scope 与可信 Turn outcome。 */\nexport interface LogicalToolOccurrenceDeriveOptions {\n    readonly session: string;\n    readonly turn: string;\n    readonly turnOrdinal: number;\n    /** events[0] 对应的 event ordinal；完整 Turn.events 默认从 core user message 的 0 开始。 */\n    readonly firstEventOrdinal?: number;\n    /** 没有可信 Outcome 时保持省略，未关闭 operation 因而只能是 partial-stream。 */\n    readonly outcome?: \"completed\" | \"failed\" | \"waiting\";\n}\n/** One immutable Turn segment in a scope projection. */\nexport interface LogicalToolOccurrenceScopeTurn {\n    readonly session: string;\n    readonly turn: string;\n    readonly turnOrdinal: number;\n    readonly events: readonly StreamEvent[];\n    /** The scope's latest Turn controls whether an unclosed call is pending. */\n    readonly outcome?: \"completed\" | \"failed\" | \"waiting\";\n    readonly firstEventOrdinal?: number;\n}\n/** occurrence 投影与其不能成为匹配候选的 orphan finished。 */\nexport interface LogicalToolOccurrenceDerivation {\n    readonly occurrences: readonly LogicalToolOccurrence[];\n    readonly orphanFinishes: readonly OrphanToolOperationFinish[];\n}\n/** HITL 停轮请求的结构化描述,供 `t.requireInputRequest(filter)` / `t.respondAll` 按条件匹配。 */\nexport interface InputRequest {\n    /** 请求的唯一标识;多个请求并停时,`InputResponse.requestId` 靠它对位。 */\n    readonly id?: string;\n    /** 模型提出的原始问题/文本(用于 `stringMatches` 过滤)。 */\n    readonly prompt?: string;\n    /** adapter 自定义的人类可读展示文案,与 `prompt` 二选一或并存,用于 UI/日志展示及过滤匹配。 */\n    readonly display?: string;\n    /** 请求关联的动作类型(如某个待批准的工具调用名),供过滤用。 */\n    readonly action?: string;\n    /** 请求携带的结构化输入(如待批准命令的参数)。 */\n    readonly input?: JsonValue;\n    /** 若请求提供了预设选项(如批准/拒绝),逐项列出;`id` 对应 `InputResponse.optionId`。 */\n    readonly options?: readonly {\n        id: string;\n        label?: string;\n    }[];\n}\n/**\n * 落盘截断的结构化标记(见 docs/feature/record/architecture.md「大值截断」):\n * `path` 是被截断的位置(事件里是字段名,如 \"output\";span 里是 attribute key,如 \"output.value\"),\n * `originalBytes` 是截断前的 UTF-8 字节数。view 显示「输出过大,已截断」靠它,不正则匹配 marker。\n */\nexport interface Truncation {\n    path: string;\n    originalBytes: number;\n}\n/**\n * 标准事件流的词汇(对标 docs/feature/adapters/architecture/events.md)。adapter 唯一的硬活就是把\n * 各 agent 五花八门的原始 transcript 映射成 StreamEvent[];映射完,整套断言免费。\n * `truncated` 只由 results writer 在落盘时刻写入(运行时全量,落盘截断);adapter 不产出它。\n */\nexport type StreamEvent = {\n    truncated?: Truncation[];\n    /**\n     * Field paths intentionally withheld by the producing boundary.  Runtime\n     * matching sees the original event shape plus this evidence limitation;\n     * persistence can retain the same signal without replacing values with a\n     * magic string.\n     */\n    redacted?: readonly string[];\n} & (\n/** assistant 回复；它不是 eval 源码上的作者事实，不参与 sourceOrder 序列。 */\n{\n    type: \"message\";\n    role: \"assistant\";\n    text: string;\n    loc?: SourceLoc;\n}\n/**\n * eval 作者发出的用户输入；`loc` 把消息叠回源码，`sourceOrder` 与断言、直接给分共用同一条\n * attempt 级序列。历史事件可省略，当前 SessionManager 产出的 user message 必写。\n */\n | {\n    type: \"message\";\n    role: \"user\";\n    text: string;\n    loc?: SourceLoc;\n    sourceOrder?: number;\n}\n/** 发起一次工具或子 agent 操作；嵌套 `kind` 让两种 started 形状保持闭合。 */\n | {\n    type: \"operation.started\";\n    operationId: string;\n    operation: {\n        kind: \"tool\";\n        name: string;\n        input: JsonValue;\n        tool?: ToolName;\n        /**\n         * Adapter 对 command / not-command 的协议级分类。暂时可选以兼容尚未迁移的\n         * 第三方 Adapter；缺失不得在 core 侧由 name、input 或 shell text 补造。\n         */\n        command?: CommandProjection;\n    } | {\n        kind: \"subagent\";\n        name: string;\n        remoteUrl?: string;\n    };\n}\n/** 工具操作完成；人工拒绝只属于工具操作。 */\n | {\n    type: \"operation.finished\";\n    operationId: string;\n    kind: \"tool\";\n    output?: JsonValue;\n    status: \"completed\" | \"failed\" | \"rejected\";\n}\n/** 子 agent 操作完成；子 agent 没有人工拒绝状态。 */\n | {\n    type: \"operation.finished\";\n    operationId: string;\n    kind: \"subagent\";\n    output?: JsonValue;\n    status: \"completed\" | \"failed\";\n}\n/**\n * 一次 Skill 加载:一等事件,由 adapter 从原生协议里识别出「这是加载 Skill」并直接产出,\n * 不靠 renderer 按工具名/文本猜。`operationId` 仅当原生协议把 Skill 加载表达成可关联的工具调用时才有。\n */\n | {\n    type: \"skill.loaded\";\n    skill: string;\n    operationId?: string;\n}\n/** 模型停下来向人请求输入(HITL);具体请求内容见 InputRequest。 */\n | {\n    type: \"input.requested\";\n    request: InputRequest;\n}\n/** 模型的思考/推理文本(非最终回复)。 */\n | {\n    type: \"thinking\";\n    text: string;\n}\n/**\n * 被测系统内部注入的、不披着 `message` 外衣的上下文文本(如 Claude Code 的 SessionStart /\n * UserPromptSubmit hook 在下一轮开始前前置进模型上下文的文本)。只承载带实际文本内容的注入;\n * `source` 是可选的原始来源标记(如 hook 名),adapter 按各自协议原样透传,不强行归一到封闭枚举。\n */\n | {\n    type: \"context.injected\";\n    text: string;\n    source?: string;\n}\n/** 上下文被压缩/摘要(如超长会话截断历史);`reason` 是可选的压缩原因说明。 */\n | {\n    type: \"compaction\";\n    reason?: string;\n}\n/** 运行中出现的错误。 */\n | {\n    type: \"error\";\n    message: string;\n});\n/**\n * core 从事件流折叠出的结构化事实(deriveRunFacts)。折叠按 operationId 把 started 与 finished 对成一条\n * 调用:配上 finished 的取其状态;只有 started、尚未等到 finished 的调用状态是 `pending`——\n * HITL 停在审批上的调用就以这个状态被断言,不是容错分支(见 docs/feature/adapters/architecture/events.md)。\n */\nexport interface ToolCall {\n    operationId: string;\n    name: ToolName;\n    originalName?: string;\n    input: JsonValue;\n    output?: JsonValue;\n    status: \"pending\" | \"completed\" | \"failed\" | \"rejected\";\n}\nexport interface SubagentCall {\n    operationId: string;\n    name: string;\n    remoteUrl?: string;\n    output?: JsonValue;\n    /** 子 agent 委派没有 rejected 状态(operation.finished 只报 completed / failed)。 */\n    status: \"pending\" | \"completed\" | \"failed\";\n}\nexport interface DerivedFacts {\n    readonly toolCalls: readonly ToolCall[];\n    readonly subagentCalls: readonly SubagentCall[];\n    readonly inputRequests: readonly InputRequest[];\n    readonly parked: boolean;\n    readonly messageCount: number;\n    readonly compactions: number;\n    /** 事件流里 `context.injected` 事件的次数;只回答存在性问题,不替代逐条读取原文。 */\n    readonly contextInjections: number;\n}\n/**\n * span 的【语义角色】,从 OTel GenAI 语义约定的 gen_ai.operation.name 归一而来\n * (见 o11y/otlp/canonical.ts)。view 据此着色 / 分组 / 跨 agent 对比,**只认这个,\n * 不读原生 span 名**。未识别的 span 落 \"other\",view 折叠。\n */\nexport type SpanKind = \"turn\" | \"model\" | \"tool\" | \"agent\" | \"other\";\n/**\n * 一条分布式追踪的 span(从 agent 经 OpenTelemetry 导出的 OTLP traces 归一而来)。\n * 与 StreamEvent 不同:它带【时间】(起止 epoch 毫秒)与【父子】(parentSpanId),\n * 所以 view 能画成瀑布图。事件流回答「做了什么」,trace 回答「各花了多久、谁套谁」。\n *\n * 两层归一:线格式层(OTLP/JSON|protobuf → 本结构,见 otlp/parse.ts,通用);\n * 语义层(原生 span 名/属性 → canonical GenAI semconv,见 otlp/mappers/<agent>.ts,每 agent 一个薄 mapper)。\n */\nexport interface TraceSpan {\n    traceId: string;\n    spanId: string;\n    parentSpanId?: string;\n    name: string;\n    /** span 起点 / 终点(epoch 毫秒)。 */\n    startMs: number;\n    endMs: number;\n    status?: \"ok\" | \"error\" | \"unset\";\n    /**\n     * 归一后的语义角色(每-agent mapper 据 canonical GenAI semconv 定;view/select 只认它)。\n     * 未经 mapper 或未识别时为 undefined / \"other\"。\n     */\n    kind?: SpanKind;\n    /** OTLP span 属性(gen_ai.* / tool 名 / token 等),按 key 摊平。raw 属性始终保留供下钻。 */\n    attributes?: globalThis.Record<string, JsonValue>;\n    /** 落盘截断标记(只由 results writer 在序列化时写入,运行时全量;见 Truncation)。 */\n    truncated?: Truncation[];\n}\n/**\n * 给人与宿主侧行为断言(`t.o11y`)看的 o11y 摘要。从 `events.json` 派生的\n * 行为计数缓存,不是权威——同一 niceeval 版本写读,删除后可从 `events.json` 重算。token 用量、\n * 成本与耗时不在这里:权威唯一在 `result.json` 的 `Usage` / `estimatedCostUSD` 与 `durationMs`\n * / `phases`(见 docs/feature/record/architecture.md「o11y.json」)。\n */\nexport interface O11ySummary {\n    totalTurns: number;\n    /** 只有跑之前调用过的 canonical 工具名才有 key,不是全量出现;省略的 ToolName 视为 0。 */\n    toolCalls: Partial<globalThis.Record<ToolName, number>>;\n    totalToolCalls: number;\n    filesRead: string[];\n    filesModified: string[];\n    shellCommands: {\n        command: string;\n        exitCode?: number;\n        success?: boolean;\n    }[];\n    webFetches: {\n        url: string;\n        status?: number;\n        success?: boolean;\n    }[];\n    errors: string[];\n    thinkingBlocks: number;\n    compactions: number;\n}\n"]}