niceeval 0.13.0 → 0.13.1

package/dist/report/runtime/resolved-page.d.cts

@@ -2,6 +2,7 @@ import { Effect } from "effect";
 import type * as Scope from "effect/Scope";
 import { type JsonValue, type Sample, type SampleIdentity } from "../../analysis/contracts.cjs";
 import type { LocalizedText } from "../../shared/types.cjs";
+import type { PanelMode } from "../model/panel.cjs";
 /** A closed page has no route back to an author callback or a live Sample. */
 export declare const RESOLVED_PAGE_FORMAT = "niceeval.report.resolved-page/v1";
 export interface ResolvedPageTarget {
@@ -13,6 +14,7 @@ export interface ResolvedPageTarget {
 export interface ResolvedPageTextProjection {
     readonly locale: string;
     readonly width: number;
+    readonly panelMode: PanelMode;
     readonly text: string;
 }
 export interface ResolvedPageWebProjection {
@@ -124,6 +126,7 @@ export declare function resolvePage<Rendered, Error, Requirements>(input: Resolv
 export declare function resolvedPageTextProjection(page: ResolvedPage, input: {
     readonly locale: string;
     readonly width: number;
+    readonly panelMode: PanelMode;
 }): ResolvedPageTextProjection | undefined;
 /** Finds an already-built browser projection. It never invokes a renderer. */
 export declare function resolvedPageWebProjection(page: ResolvedPage, input: {

package/dist/report/runtime/resolved-page.d.mts

@@ -2,6 +2,7 @@ import { Effect } from "effect";
 import type * as Scope from "effect/Scope";
 import { type JsonValue, type Sample, type SampleIdentity } from "../../analysis/contracts.mjs";
 import type { LocalizedText } from "../../shared/types.mjs";
+import type { PanelMode } from "../model/panel.mjs";
 /** A closed page has no route back to an author callback or a live Sample. */
 export declare const RESOLVED_PAGE_FORMAT = "niceeval.report.resolved-page/v1";
 export interface ResolvedPageTarget {
@@ -13,6 +14,7 @@ export interface ResolvedPageTarget {
 export interface ResolvedPageTextProjection {
     readonly locale: string;
     readonly width: number;
+    readonly panelMode: PanelMode;
     readonly text: string;
 }
 export interface ResolvedPageWebProjection {
@@ -124,6 +126,7 @@ export declare function resolvePage<Rendered, Error, Requirements>(input: Resolv
 export declare function resolvedPageTextProjection(page: ResolvedPage, input: {
     readonly locale: string;
     readonly width: number;
+    readonly panelMode: PanelMode;
 }): ResolvedPageTextProjection | undefined;
 /** Finds an already-built browser projection. It never invokes a renderer. */
 export declare function resolvedPageWebProjection(page: ResolvedPage, input: {

package/dist/report/runtime/resolved-page.d.ts

@@ -2,6 +2,7 @@ import { Effect } from "effect";
 import type * as Scope from "effect/Scope";
 import { type JsonValue, type Sample, type SampleIdentity } from "../../analysis/contracts.js";
 import type { LocalizedText } from "../../shared/types.js";
+import type { PanelMode } from "../model/panel.js";
 /** A closed page has no route back to an author callback or a live Sample. */
 export declare const RESOLVED_PAGE_FORMAT = "niceeval.report.resolved-page/v1";
 export interface ResolvedPageTarget {
@@ -13,6 +14,7 @@ export interface ResolvedPageTarget {
 export interface ResolvedPageTextProjection {
     readonly locale: string;
     readonly width: number;
+    readonly panelMode: PanelMode;
     readonly text: string;
 }
 export interface ResolvedPageWebProjection {
@@ -124,6 +126,7 @@ export declare function resolvePage<Rendered, Error, Requirements>(input: Resolv
 export declare function resolvedPageTextProjection(page: ResolvedPage, input: {
     readonly locale: string;
     readonly width: number;
+    readonly panelMode: PanelMode;
 }): ResolvedPageTextProjection | undefined;
 /** Finds an already-built browser projection. It never invokes a renderer. */
 export declare function resolvedPageWebProjection(page: ResolvedPage, input: {

package/dist/report/runtime/text.cjs

@@ -27,6 +27,7 @@ function textProjectionMissing(page, options) {
         route: page.target.route,
         locale: options.locale,
         width: options.width,
+        panelMode: options.panelMode,
     });
 }
 //# sourceMappingURL=text.cjs.map

package/dist/report/runtime/text.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"text.js","sourceRoot":"","sources":["text.ts"],"names":[],"mappings":";;AAkBA,wDAKC;AAED,4CAQC;AAjCD,yDAAoF;AAcpF;;;GAGG;AACH,SAAgB,sBAAsB,CAAC,IAAkB,EAAE,OAAsC;IAC7F,MAAM,UAAU,GAAG,IAAA,6CAA0B,EAAC,IAAI,EAAE,OAAO,CAAC,CAAC;IAC7D,IAAI,UAAU,KAAK,SAAS;QACxB,OAAO,UAAU,CAAC,IAAI,CAAC;IAC3B,MAAM,qBAAqB,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;AAC/C,CAAC;AACD,2EAA2E;AAC3E,SAAgB,gBAAgB,CAAC,IAAkB,EAAE,OAAsC;IAIvF,MAAM,UAAU,GAAG,IAAA,6CAA0B,EAAC,IAAI,EAAE,OAAO,CAAC,CAAC;IAC7D,IAAI,UAAU,KAAK,SAAS;QACxB,OAAO,MAAM,CAAC,MAAM,CAAC,EAAE,KAAK,EAAE,UAAmB,EAAE,IAAI,EAAE,UAAU,CAAC,IAAI,EAAE,CAAC,CAAC;IAChF,OAAO,qBAAqB,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;AAChD,CAAC;AACD,SAAS,qBAAqB,CAAC,IAAkB,EAAE,OAAsC;IACrF,OAAO,MAAM,CAAC,MAAM,CAAC;QACjB,IAAI,EAAE,gCAAyC;QAC/C,MAAM,EAAE,IAAI,CAAC,MAAM,CAAC,MAAM;QAC1B,KAAK,EAAE,IAAI,CAAC,MAAM,CAAC,KAAK;QACxB,MAAM,EAAE,OAAO,CAAC,MAAM;QACtB,KAAK,EAAE,OAAO,CAAC,KAAK;KACvB,CAAC,CAAC;AACP,CAAC","sourcesContent":["import { resolvedPageTextProjection, type ResolvedPage, } from \"./resolved-page.ts\";\nexport interface RenderResolvedPageTextOptions {\n    /** Explicit: a machine or terminal caller must never inherit process locale. */\n    readonly locale: string;\n    /** Explicit: closed terminal output is only reusable at the requested width. */\n    readonly width: number;\n}\nexport interface ResolvedPageTextProjectionMissing {\n    readonly code: \"report-text-projection-missing\";\n    readonly pageId: string;\n    readonly route: string;\n    readonly locale: string;\n    readonly width: number;\n}\n/**\n * Reads one pre-closed terminal projection. This function deliberately has no\n * Sample, callback, renderer, or Effect dependency.\n */\nexport function renderResolvedPageText(page: ResolvedPage, options: RenderResolvedPageTextOptions): string {\n    const projection = resolvedPageTextProjection(page, options);\n    if (projection !== undefined)\n        return projection.text;\n    throw textProjectionMissing(page, options);\n}\n/** A total variant for Host error mapping without exception inspection. */\nexport function resolvedPageText(page: ResolvedPage, options: RenderResolvedPageTextOptions): {\n    readonly state: \"rendered\";\n    readonly text: string;\n} | ResolvedPageTextProjectionMissing {\n    const projection = resolvedPageTextProjection(page, options);\n    if (projection !== undefined)\n        return Object.freeze({ state: \"rendered\" as const, text: projection.text });\n    return textProjectionMissing(page, options);\n}\nfunction textProjectionMissing(page: ResolvedPage, options: RenderResolvedPageTextOptions): ResolvedPageTextProjectionMissing {\n    return Object.freeze({\n        code: \"report-text-projection-missing\" as const,\n        pageId: page.target.pageId,\n        route: page.target.route,\n        locale: options.locale,\n        width: options.width,\n    });\n}\n"]}
+{"version":3,"file":"text.js","sourceRoot":"","sources":["text.ts"],"names":[],"mappings":";;AAsBA,wDAKC;AAED,4CAQC;AArCD,yDAAoF;AAkBpF;;;GAGG;AACH,SAAgB,sBAAsB,CAAC,IAAkB,EAAE,OAAsC;IAC7F,MAAM,UAAU,GAAG,IAAA,6CAA0B,EAAC,IAAI,EAAE,OAAO,CAAC,CAAC;IAC7D,IAAI,UAAU,KAAK,SAAS;QACxB,OAAO,UAAU,CAAC,IAAI,CAAC;IAC3B,MAAM,qBAAqB,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;AAC/C,CAAC;AACD,2EAA2E;AAC3E,SAAgB,gBAAgB,CAAC,IAAkB,EAAE,OAAsC;IAIvF,MAAM,UAAU,GAAG,IAAA,6CAA0B,EAAC,IAAI,EAAE,OAAO,CAAC,CAAC;IAC7D,IAAI,UAAU,KAAK,SAAS;QACxB,OAAO,MAAM,CAAC,MAAM,CAAC,EAAE,KAAK,EAAE,UAAmB,EAAE,IAAI,EAAE,UAAU,CAAC,IAAI,EAAE,CAAC,CAAC;IAChF,OAAO,qBAAqB,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;AAChD,CAAC;AACD,SAAS,qBAAqB,CAAC,IAAkB,EAAE,OAAsC;IACrF,OAAO,MAAM,CAAC,MAAM,CAAC;QACjB,IAAI,EAAE,gCAAyC;QAC/C,MAAM,EAAE,IAAI,CAAC,MAAM,CAAC,MAAM;QAC1B,KAAK,EAAE,IAAI,CAAC,MAAM,CAAC,KAAK;QACxB,MAAM,EAAE,OAAO,CAAC,MAAM;QACtB,KAAK,EAAE,OAAO,CAAC,KAAK;QACpB,SAAS,EAAE,OAAO,CAAC,SAAS;KAC/B,CAAC,CAAC;AACP,CAAC","sourcesContent":["import { resolvedPageTextProjection, type ResolvedPage, } from \"./resolved-page.ts\";\nimport type { PanelMode } from \"../model/panel.ts\";\nexport interface RenderResolvedPageTextOptions {\n    /** Explicit: a machine or terminal caller must never inherit process locale. */\n    readonly locale: string;\n    /** Explicit: closed terminal output is only reusable at the requested width. */\n    readonly width: number;\n    /** Explicit: callers select a pre-closed terminal or plain-text projection. */\n    readonly panelMode: PanelMode;\n}\nexport interface ResolvedPageTextProjectionMissing {\n    readonly code: \"report-text-projection-missing\";\n    readonly pageId: string;\n    readonly route: string;\n    readonly locale: string;\n    readonly width: number;\n    readonly panelMode: PanelMode;\n}\n/**\n * Reads one pre-closed terminal projection. This function deliberately has no\n * Sample, callback, renderer, or Effect dependency.\n */\nexport function renderResolvedPageText(page: ResolvedPage, options: RenderResolvedPageTextOptions): string {\n    const projection = resolvedPageTextProjection(page, options);\n    if (projection !== undefined)\n        return projection.text;\n    throw textProjectionMissing(page, options);\n}\n/** A total variant for Host error mapping without exception inspection. */\nexport function resolvedPageText(page: ResolvedPage, options: RenderResolvedPageTextOptions): {\n    readonly state: \"rendered\";\n    readonly text: string;\n} | ResolvedPageTextProjectionMissing {\n    const projection = resolvedPageTextProjection(page, options);\n    if (projection !== undefined)\n        return Object.freeze({ state: \"rendered\" as const, text: projection.text });\n    return textProjectionMissing(page, options);\n}\nfunction textProjectionMissing(page: ResolvedPage, options: RenderResolvedPageTextOptions): ResolvedPageTextProjectionMissing {\n    return Object.freeze({\n        code: \"report-text-projection-missing\" as const,\n        pageId: page.target.pageId,\n        route: page.target.route,\n        locale: options.locale,\n        width: options.width,\n        panelMode: options.panelMode,\n    });\n}\n"]}

package/dist/report/runtime/text.d.cts

@@ -1,9 +1,12 @@
 import { type ResolvedPage } from "./resolved-page.cjs";
+import type { PanelMode } from "../model/panel.cjs";
 export interface RenderResolvedPageTextOptions {
     /** Explicit: a machine or terminal caller must never inherit process locale. */
     readonly locale: string;
     /** Explicit: closed terminal output is only reusable at the requested width. */
     readonly width: number;
+    /** Explicit: callers select a pre-closed terminal or plain-text projection. */
+    readonly panelMode: PanelMode;
 }
 export interface ResolvedPageTextProjectionMissing {
     readonly code: "report-text-projection-missing";
@@ -11,6 +14,7 @@ export interface ResolvedPageTextProjectionMissing {
     readonly route: string;
     readonly locale: string;
     readonly width: number;
+    readonly panelMode: PanelMode;
 }
 /**
  * Reads one pre-closed terminal projection. This function deliberately has no

package/dist/report/runtime/text.d.mts

@@ -1,9 +1,12 @@
 import { type ResolvedPage } from "./resolved-page.mjs";
+import type { PanelMode } from "../model/panel.mjs";
 export interface RenderResolvedPageTextOptions {
     /** Explicit: a machine or terminal caller must never inherit process locale. */
     readonly locale: string;
     /** Explicit: closed terminal output is only reusable at the requested width. */
     readonly width: number;
+    /** Explicit: callers select a pre-closed terminal or plain-text projection. */
+    readonly panelMode: PanelMode;
 }
 export interface ResolvedPageTextProjectionMissing {
     readonly code: "report-text-projection-missing";
@@ -11,6 +14,7 @@ export interface ResolvedPageTextProjectionMissing {
     readonly route: string;
     readonly locale: string;
     readonly width: number;
+    readonly panelMode: PanelMode;
 }
 /**
  * Reads one pre-closed terminal projection. This function deliberately has no

package/dist/report/runtime/text.d.ts

@@ -1,9 +1,12 @@
 import { type ResolvedPage } from "./resolved-page.js";
+import type { PanelMode } from "../model/panel.js";
 export interface RenderResolvedPageTextOptions {
     /** Explicit: a machine or terminal caller must never inherit process locale. */
     readonly locale: string;
     /** Explicit: closed terminal output is only reusable at the requested width. */
     readonly width: number;
+    /** Explicit: callers select a pre-closed terminal or plain-text projection. */
+    readonly panelMode: PanelMode;
 }
 export interface ResolvedPageTextProjectionMissing {
     readonly code: "report-text-projection-missing";
@@ -11,6 +14,7 @@ export interface ResolvedPageTextProjectionMissing {
     readonly route: string;
     readonly locale: string;
     readonly width: number;
+    readonly panelMode: PanelMode;
 }
 /**
  * Reads one pre-closed terminal projection. This function deliberately has no

package/docs-site/zh/explanation/runner.mdx

@@ -66,8 +66,8 @@ Reporter 只接入当前 Invocation。它可以把当前进程已知的结果交
 Runner 结束后，用明确 Run 选择查看结果：
 
 ```sh
-npx niceeval show --run 01J9ZK3M6P4T7V9X2C5N8QW0RY
-npx niceeval view --run 01J9ZK3M6P4T7V9X2C5N8QW0RY --no-open
+pnpm exec niceeval show --run 01J9ZK3M6P4T7V9X2C5N8QW0RY
+pnpm exec niceeval view --run 01J9ZK3M6P4T7V9X2C5N8QW0RY --no-open
 ```
 
 不带 locator 或 `--run` 的 `show` 和 `view` 读取身份仍匹配当前项目的全部结果；locator 与 `--run` 读取指定历史事实。NiceEval 形成 `Sample`，再按选中 Page 或完整站点所需的 Measure 与领域视图取得闭合值。

package/docs-site/zh/index.mdx

@@ -107,8 +107,8 @@ export default defineExperiment({
 
 ```sh
 npx niceeval exp local weather-tool
-npx niceeval show       # 终端读取结果，失败项可继续下钻
-npx niceeval view       # 网页交互浏览
+pnpm exec niceeval show       # 终端读取结果，失败项可继续下钻
+pnpm exec niceeval view       # 网页交互浏览
 ```
 
 如果被测对象是需要隔离工作区的 coding agent（Codex、Claude Code 插件/Skill），评估用例里改用 `t.sandbox.uploadDirectory()` 铺工作区。检查文件变化用 `t.sandbox.fileChanged()` 等归因断言。`t.sandbox.runCommand()` 跑测试。参考独立仓库 [coding-agent-skill](https://github.com/CorrectRoadH/coding-agent-skill)。

package/docs-site/zh/reference/cli.mdx

@@ -24,10 +24,10 @@ description: "NiceEval CLI 参考：exp、debug、show、view、init、list 和
   <Card title="npx niceeval clean" icon="trash">
     删除明确 writer 留下的 Record 临时目录，不删除正式 Run 或 Attempt。
   </Card>
-  <Card title="npx niceeval show [--run]" icon="terminal">
+  <Card title="pnpm exec niceeval show [--run]" icon="terminal">
     在终端执行报告。不带 locator 或 `--run` 的命令读取当前项目结果；精确 Attempt 定位符或 Run ID 用于审计历史。
   </Card>
-  <Card title="npx niceeval view" icon="eye">
+  <Card title="pnpm exec niceeval view" icon="eye">
     构建完整报告站并在浏览器中托管。它与 `show` 共享选择规则，但会枚举全部页面。
   </Card>
 </CardGroup>
@@ -205,7 +205,7 @@ npx niceeval exp compare-models weather-tool
 
 运行命名 experiment，用矩阵比较 agents、models 或 flags。第二个参数开始是评估用例 ID 前缀过滤。
 
-`show` 与 `view` 是顶层命令，应写成 `npx niceeval show`、`npx niceeval view`。误写成 `npx niceeval exp show` / `exp view` 且没有同名 experiment 时，CLI 会在“不存在的实验”错误后提示正确命令。仓库确实存在同名 experiment 时仍按合法 id 执行。
+`show` 与 `view` 是顶层命令，应写成 `pnpm exec niceeval show`、`pnpm exec niceeval view`。误写成 `pnpm exec niceeval exp show` / `exp view` 且没有同名 experiment 时，CLI 会在“不存在的实验”错误后提示正确命令。仓库确实存在同名 experiment 时仍按合法 id 执行。
 
 ## 两种输出：人读文本与 `--json`
 
@@ -227,10 +227,10 @@ npx niceeval exp compare --junit ./artifacts/niceeval-junit.xml
 ## `view`
 
 ```bash
-npx niceeval view
-npx niceeval view --run 01H...
-npx niceeval view @01H...
-npx niceeval view --host 192.168.0.199
+pnpm exec niceeval view
+pnpm exec niceeval view --run 01H...
+pnpm exec niceeval view @01H...
+pnpm exec niceeval view --host 192.168.0.199
 ```
 
 打开本地结果查看器。它和 `show` 共用默认 Report 与选择规则。不带 locator 或 `--run` 的命令读取所有身份仍匹配当前项目的结果；精确 `@<AttemptLocator>` 显示该 Attempt 的默认概览；`--run` 读取指定历史 Run。NiceEval 随后形成固定 Sample、枚举全部 Page 与参数实例，并构建完整站点。
@@ -246,13 +246,13 @@ npx niceeval view --host 192.168.0.199
 ## `show`
 
 ```bash
-npx niceeval show
-npx niceeval show --run 01H...
-npx niceeval show --run 01H... --run 01J...
-npx niceeval show @01H...
-npx niceeval show --run 01H... --page /
-npx niceeval show --run 01H... --report ./reports/summary.tsx --page /overview
-npx niceeval show --run 01H... --report ./reports/summary.tsx --json
+pnpm exec niceeval show
+pnpm exec niceeval show --run 01H...
+pnpm exec niceeval show --run 01H... --run 01J...
+pnpm exec niceeval show @01H...
+pnpm exec niceeval show --run 01H... --page /
+pnpm exec niceeval show --run 01H... --report ./reports/summary.tsx --page /overview
+pnpm exec niceeval show --run 01H... --report ./reports/summary.tsx --json
 ```
 
 `show` 是终端报告入口，适合人直接阅读，也适合 coding agent 在上下文窗口里逐级下钻。它不接受模糊的位置选择。省略 `--page` 时选择默认 Page；提供 `--page <route>` 时只执行这个精确 route。

package/docs-site/zh/reference/events.mdx

@@ -267,7 +267,7 @@ events.push({
 });
 ```
 
-`commandProjection()` 保留 Adapter 已确认的 original tokens，并调用同一份 `normalizeLogicalCommand()` 生成 `logical-command/v1` 投影。`pnpm exec niceeval show` 和 `npx niceeval show` 因此都能由 `commandMatch("niceeval", { argsStart: ["show"] })` 精确匹配。
+`commandProjection()` 保留 Adapter 已确认的 original tokens，并调用同一份 `normalizeLogicalCommand()` 生成 `logical-command/v1` 投影。`pnpm exec niceeval show` 能由 `commandMatch("niceeval", { argsStart: ["show"] })` 精确匹配。
 
 只有原生协议已经给出 argv，或协议 grammar 能无歧义地产生单一 invocation，才能把 original 标为 available。协议只给 shell source、内容已截断或脱敏时使用 `opaqueCommandProjection(reason)`；能确认不是 command 时使用 `notCommandProjection()`。无法确认 command / not-command 时降低 actions coverage，不能从 tool name、input 或 shell 文本猜测。
 

package/docs-site/zh/troubleshooting/debugging.mdx

@@ -7,8 +7,8 @@ description: "根据可见错误、Sample 状态和 rebuild 反馈，定位查
 先用明确 Run 读取问题，不要从目录名或时间猜测目标。
 
 ```sh
-npx niceeval show --run <run-id>
-npx niceeval view --run <run-id> --no-open
+pnpm exec niceeval show --run <run-id>
+pnpm exec niceeval view --run <run-id> --no-open
 ```
 
 这两条命令从同一范围形成 Sample，并执行同一份 Report。先确定选择是否正确，再检查页面中的度量和问题。

package/docs-site/zh/tutorials/agent-feedback-loop.mdx

@@ -49,8 +49,8 @@ Agent 应把最后一条 receipt 当作本次调用的交接信息。它不是
 运行结束后，先读取 Run 的 Report：
 
 ```sh
-npx niceeval show --run 01J9ZK3M6P4T7V9X2C5N8QW0RY
-npx niceeval view --run 01J9ZK3M6P4T7V9X2C5N8QW0RY --no-open
+pnpm exec niceeval show --run 01J9ZK3M6P4T7V9X2C5N8QW0RY
+pnpm exec niceeval view --run 01J9ZK3M6P4T7V9X2C5N8QW0RY --no-open
 ```
 
 `show` 与 `view` 选择 Sample，再取得当前目标需要的闭合分析值。`show` 只执行一个 Page；`view` 构建完整站点。Agent 应区分以下状态：

package/docs-site/zh/tutorials/ci-integration.mdx

@@ -50,7 +50,7 @@ with:
 成功建立 Invocation 后，NDJSON 的最后一行包含 receipt。后续脚本可以读取它的 `runIds`，用明确 Run 选择查看或导出报告：
 
 ```sh
-npx niceeval view --run 01J9ZK3M6P4T7V9X2C5N8QW0RY \
+pnpm exec niceeval view --run 01J9ZK3M6P4T7V9X2C5N8QW0RY \
   --out ./report-site \
   --no-open
 ```

package/docs-site/zh/tutorials/connect-your-agent.mdx

@@ -101,10 +101,10 @@ export default defineEval({
 ```bash
 npx niceeval exp my-agent        # 跑这个 experiment 下的全部 eval
 npx niceeval exp my-agent refund # 只跑 ID 以 refund 开头的
-npx niceeval view --experiment my-agent  # 本地查看器里看结果
+pnpm exec niceeval view --experiment my-agent  # 本地查看器里看结果
 ```
 
-**验证运行结果**：终端会显示动态 dashboard，完成和排队数量在原位更新。失败、错误和 warning 会保留在输出中。运行结束后会打印摘要和 receipt；用其中的 Run ID 也可以执行 `npx niceeval view --run <runId>`，查看每条评估用例逐轮的输入、事件和评分明细。
+**验证运行结果**：终端会显示动态 dashboard，完成和排队数量在原位更新。失败、错误和 warning 会保留在输出中。运行结束后会打印摘要和 receipt；用其中的 Run ID 也可以执行 `pnpm exec niceeval view --run <runId>`，查看每条评估用例逐轮的输入、事件和评分明细。
 
 没跑通时，按报错的位置分三类排查：
 

package/docs-site/zh/tutorials/custom-reports.mdx

@@ -173,10 +173,10 @@ export default defineConfig({
 ## 运行、读取和发布
 
 ```sh
-npx niceeval show --run <run-id> --report ./reports/quality.tsx
-npx niceeval show --run <run-id> --report ./reports/quality.tsx --json
-npx niceeval view --run <run-id> --report ./reports/quality.tsx --no-open
-npx niceeval view --run <run-id> --report ./reports/quality.tsx --out ./quality-site
+pnpm exec niceeval show --run <run-id> --report ./reports/quality.tsx
+pnpm exec niceeval show --run <run-id> --report ./reports/quality.tsx --json
+pnpm exec niceeval view --run <run-id> --report ./reports/quality.tsx --no-open
+pnpm exec niceeval view --run <run-id> --report ./reports/quality.tsx --out ./quality-site
 ```
 
 前三条命令从相同选择形成 Sample，但范围不同：`show` 和 `show --json` 只执行一个 Page，并只捕获该页的 projection closure；`view` 与 `view --out` 枚举全部 Page 后形成完整站点及 `_niceeval/data/projections.json`。该文件的 bytes 进入 revision identity。view 的成功 rebuild 原子替换当前站点；失败时保留上一份可用站点。`--out` 必须指向不存在的目录。

package/docs-site/zh/tutorials/deploy-report-site.mdx

@@ -7,7 +7,7 @@ description: "将 view --out 生成的完整自包含报告目录部署到任意
 先生成完整报告站，再部署输出目录。托管平台不读取源 Record，也不需要安装或运行 NiceEval。
 
 ```sh
-npx niceeval view --experiment checkout \
+pnpm exec niceeval view --experiment checkout \
   --out ./report-site \
   --no-open
 ```
@@ -15,7 +15,7 @@ npx niceeval view --experiment checkout \
 要部署一次明确运行，请从 receipt 读取 Run ID：
 
 ```sh
-npx niceeval view --run <run-id> --out ./report-site --no-open
+pnpm exec niceeval view --run <run-id> --out ./report-site --no-open
 ```
 
 ## CI 流程
@@ -30,7 +30,7 @@ GitHub Actions 的构建任务示例：
 
 ```yaml
 - name: 生成报告站
-  run: npx niceeval view --experiment checkout --out ./report-site --no-open
+  run: pnpm exec niceeval view --experiment checkout --out ./report-site --no-open
 
 - name: 上传静态站点目录
   uses: actions/upload-pages-artifact@v3

package/docs-site/zh/tutorials/experiments.mdx

@@ -75,8 +75,8 @@ api-validation   claude-code             pass@3 = 1/3 (33%)    mean 41s
 除了 pass rate，还应该看平均耗时、token、成本和失败类型。
 
 ```bash
-npx niceeval show --experiment models
-npx niceeval view --experiment models
+pnpm exec niceeval show --experiment models
+pnpm exec niceeval view --experiment models
 ```
 
 每个 experiment 用自己的 `evals` 选择评估用例。函数形式会遍历所有已发现的评估用例：

package/docs-site/zh/tutorials/handle-execution-failures.mdx

@@ -108,7 +108,7 @@ npx niceeval exp compare evals/coding
 需要查看某个 Attempt 的原始错误和重试摘要时，使用终端给出的定位符：
 
 ```shell
-npx niceeval show @<attempt-locator>
+pnpm exec niceeval show @<attempt-locator>
 ```
 
 自定义 Adapter 如果有专用的受理前拒绝信号，可以提供 `classifySendFailure`。该字段的类型和边界见 [`defineSandboxAgent` 参考](/zh/reference/define-agent#definesandboxagent)。

package/docs-site/zh/tutorials/install-custom-sandbox-agent.mdx

@@ -87,7 +87,7 @@ Sandbox 不需要 Node.js 或 npm。`createNpmCliInstaller` 在宿主机上下
 安装或复检失败时，Attempt 会在 `agent.ensure` 阶段变成 `errored`。终端会给出 Attempt 定位符。
 
 ```shell
-npx niceeval show @<attempt-locator>
+pnpm exec niceeval show @<attempt-locator>
 ```
 
 先检查错误里的包名、版本和目标平台。修正 `identity`、npm 包或安装步骤后，重新运行原命令。

package/docs-site/zh/tutorials/publish-report.mdx

@@ -7,7 +7,7 @@ description: "用 view --out 导出可离线打开、与源 Record 脱离且包
 `niceeval view --out <目录>` 从已发布结果形成固定 Sample，再构建完整报告站。它不是把当前浏览器页面保存为文件：`pages` 中显式声明的所有普通 Page、参数 Page 实例、资源和下载都必须先关闭并通过校验；Host 不补建详情页。
 
 ```sh
-npx niceeval view --experiment checkout \
+pnpm exec niceeval view --experiment checkout \
   --out ./report-site \
   --no-open
 ```
@@ -15,7 +15,7 @@ npx niceeval view --experiment checkout \
 需要固定到本次运行时，从 receipt 的 `runIds` 取得 Run ID：
 
 ```sh
-npx niceeval view --run <run-id> \
+pnpm exec niceeval view --run <run-id> \
   --out ./report-site \
   --no-open
 ```
@@ -47,14 +47,14 @@ Analysis 的 `partial`、`empty`、`unsupported` 与 `failed` 是可显示的数
 
 ```sh
 # 成功时写入全新目录
-npx niceeval view --run <run-id> --out ./report-site-2026-08-15
+pnpm exec niceeval view --run <run-id> --out ./report-site-2026-08-15
 ```
 
 ## 在 CI 中生成
 
 ```sh
 npx niceeval exp ci --json
-npx niceeval view --experiment ci --out ./report-site --no-open
+pnpm exec niceeval view --experiment ci --out ./report-site --no-open
 ```
 
 若 CI 必须发布刚完成的那次运行，读取第一条命令最后一行 receipt 的 `runIds`，并使用 `view --run <run-id>`。不要让托管平台读取源 Record；它只应部署生成目录。

package/docs-site/zh/tutorials/reporters.mdx

@@ -47,7 +47,7 @@ npx niceeval exp checkout --json
 `--json` 输出当前进程的 NDJSON。最后一条 receipt 包含 `invocationId`、`runIds`、开始时间、结束时间和 completion。外部自动化可以保留 receipt，再根据 `runIds` 选择要查看的 Report。
 
 ```sh
-npx niceeval show --run 01J9ZK3M6P4T7V9X2C5N8QW0RY
+pnpm exec niceeval show --run 01J9ZK3M6P4T7V9X2C5N8QW0RY
 ```
 
 不要把 progress 或 diagnostic 行当作长期结果格式。它们服务当前 Invocation，完整业务事实属于停稳 Record。

package/docs-site/zh/tutorials/viewing-results.mdx

@@ -7,8 +7,8 @@ description: "用 show 或 view 从已发布的 Record 选择固定 Sample，并
 运行结束后，从下面两条命令开始：
 
 ```sh
-npx niceeval show
-npx niceeval view
+pnpm exec niceeval show
+pnpm exec niceeval view
 ```
 
 两条命令用相同选择规则形成固定 `Sample`，但执行范围不同。`show` 只读取一个目标 Page；`view` 会先构建完整站点，再把它托管在浏览器中。
@@ -18,7 +18,7 @@ npx niceeval view
 选择一个明确 Run：
 
 ```sh
-npx niceeval show --run 7b8d2ea4-b840-4870-9840-f85a436a5527
+pnpm exec niceeval show --run 7b8d2ea4-b840-4870-9840-f85a436a5527
 ```
 
 普通运行可从 `exp --json` 最后一条 receipt 的 `runIds` 读取 Run ID：
@@ -43,10 +43,10 @@ Sample 记录每个位置的状态：
 ## 在终端读取一个 Page
 
 ```sh
-npx niceeval show --run <run-id>
-npx niceeval show --run <run-id> --page /
-npx niceeval show @<attempt-locator>
-npx niceeval show --run <run-id> --report ./reports/quality.tsx --json
+pnpm exec niceeval show --run <run-id>
+pnpm exec niceeval show --run <run-id> --page /
+pnpm exec niceeval show @<attempt-locator>
+pnpm exec niceeval show --run <run-id> --report ./reports/quality.tsx --json
 ```
 
 省略 `--page` 时，`show` 使用 Report 的默认可导航页面。提供 `--page` 时，它只 decode、load、render 和关闭这一个精确 route。参数 Page 不会因此调用 `enumerate()`；其它 Page、Analysis 查询和作者 callback 也不会执行。
@@ -69,8 +69,8 @@ JSON 不包含其它 Page、HTML、React tree、站点 identity 或 Record 读
 ## 在浏览器读取完整站点
 
 ```sh
-npx niceeval view --experiment checkout --no-open
-npx niceeval view --run <run-id> --page / --no-open
+pnpm exec niceeval view --experiment checkout --no-open
+pnpm exec niceeval view --run <run-id> --page / --no-open
 ```
 
 view 在启动 server 前枚举所有显式声明的普通 Page 与参数 Page 的全部实例，并验证路径、链接、资源、下载、`_niceeval/data/projections.json` 和预算。`pages` 是唯一页面集合，Host 不会补建详情 route。`--page` 只决定初始打开的已存在 route，不能缩小构建范围。全站 projections 文件的 bytes 属于 revision identity。

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "niceeval",
-  "version": "0.13.0",
+  "version": "0.13.1",
   "description": "Agent-native eval tool — eval agents, services, functions, and coding-agent fixtures",
   "type": "module",
   "license": "MIT",