pi-skill-evolve 0.1.9 → 0.1.11

package/extensions/skill-evolve/test/selftest.test.ts

@@ -1,7 +1,10 @@
 /**
- * 必红自检单测（设计稿 12.4）：三个坏样本必须被拦住，任何一条被 pass 就算评审器不可信。
+ * 必红自检单测（设计稿 12.4）：三个坏样本必须被拦住，一个正样本必须放行且写出理由；
+ * 任何一条不符就算评审器不可信。
  *
  * 假 spawn 按 prompt 里的「技能名」回不同结论，所以用例不受 fixtures 目录枚举顺序影响。
+ * 回的 JSON 必须是能过 postProcessReview 的合法形状（summary 必填，pass 必须带 passed），
+ * 否则测的就不是「期望命中」而是「载荷校验」了。
  */
 import { test } from "node:test";
 import assert from "node:assert/strict";
@@ -20,11 +23,11 @@ function tmpDir(): string {
 }
 
 /** 按 prompt 末尾的「技能名：x」挑结论 */
-function spawnBySkill(verdicts: Record<string, string>): ReviewSpawn {
+function spawnBySkill(verdicts: Record<string, string>, opts: { dropPassed?: boolean } = {}): ReviewSpawn {
   return (_command, args) => {
     const prompt = args[3] ?? "";
     const skill = Object.keys(verdicts).find((name) => prompt.includes(`技能名：${name}`));
-    const reply = JSON.stringify({ verdict: skill ? verdicts[skill] : "pass", blocking: [], notes: "自检" });
+    const reply = fakeReply(skill ? (verdicts[skill] ?? "pass") : "pass", { dropPassed: opts.dropPassed === true });
     const child = new FakeReviewChild();
     setTimeout(() => {
       child.stdout.emit("data", reply);
@@ -34,42 +37,74 @@ function spawnBySkill(verdicts: Record<string, string>): ReviewSpawn {
   };
 }
 
-test("三个坏样本都能被读到，期望分别是 reject/reject/revise", () => {
+const FAKE_EVIDENCE = { file: ".pi/skills/demo/SKILL.md", line: 9, quote: "凡是查日志一律用 rg" };
+
+/** 造一份合法评审输出：reject/revise 带 blocking，pass 带 passed（可故意抽掉，测正样本的必红判定） */
+function fakeReply(verdict: string, opts: { dropPassed: boolean }): string {
+  const summary = "假评审：只校验自检链路，不代表真实评审质量";
+  if (verdict === "pass") {
+    const passed = opts.dropPassed ? [] : [{ point: "原有禁令仍完整保留", evidence: FAKE_EVIDENCE }];
+    return JSON.stringify({ verdict, summary, blocking: [], passed, notes: "自检" });
+  }
+  return JSON.stringify({ verdict, summary, blocking: [{ issue: "有问题", evidence: FAKE_EVIDENCE }], passed: [], notes: "自检" });
+}
+
+test("三个坏样本 + 一个正样本都能被读到，期望分别是 reject/reject/revise/pass", () => {
   const fixtures = loadFixtures(FIXTURES);
-  assert.deepEqual(fixtures.map((item) => item.name).sort(), ["delete-ban", "rg-to-find", "weak-description"]);
+  assert.deepEqual(fixtures.map((item) => item.name).sort(), ["delete-ban", "good-description", "rg-to-find", "weak-description"]);
   const expected = Object.fromEntries(fixtures.map((item) => [item.name, item.expected]));
-  assert.deepEqual(expected, { "delete-ban": "reject", "rg-to-find": "reject", "weak-description": "revise" });
+  assert.deepEqual(expected, {
+    "delete-ban": "reject",
+    "good-description": "pass",
+    "rg-to-find": "reject",
+    "weak-description": "revise",
+  });
   for (const fixture of fixtures) {
     assert.ok(fixture.diff.length > 0, `${fixture.name} 缺 diff`);
     assert.ok(fixture.after.length > 0, `${fixture.name} 缺改后全文`);
   }
 });
 
-test("三样本全部被正确拦住 → passed:true 并落盘 selftest.json", async () => {
+test("四样本全部命中期望 → passed:true 并落盘 selftest.json", async () => {
   const dir = tmpDir();
   const report = await runSelftest({
     evolutionDir: dir,
     cwd: dir,
     fixturesDir: FIXTURES,
-    spawnFn: spawnBySkill({ "ui-verify-lite": "reject", "rg-retrieval": "reject", "misc-helper": "revise" }),
-    pollMs: 1,
+    spawnFn: spawnBySkill({ "ui-verify-lite": "reject", "rg-retrieval": "reject", "misc-helper": "revise", "log-helper": "pass" }),
   });
   assert.equal(report.passed, true);
-  assert.equal(report.results.length, 3);
-  assert.deepEqual(report.results.map((item) => item.actual), ["reject", "reject", "revise"]);
+  assert.equal(report.results.length, 4);
+  assert.deepEqual(report.results.map((item) => item.actual), ["reject", "pass", "reject", "revise"]);
   const saved = JSON.parse(fs.readFileSync(selftestFile(dir), "utf8")) as { passed: boolean };
   assert.equal(saved.passed, true);
   assert.ok(!renderSelftestReport(report).includes("【红】"));
 });
 
+test("正样本判 pass 但没给 passed 理由 → 判不合格并写清差在哪（必红自检要能拉住无理由的通过）", async () => {
+  const dir = tmpDir();
+  const report = await runSelftest({
+    evolutionDir: dir,
+    cwd: dir,
+    fixturesDir: FIXTURES,
+    spawnFn: spawnBySkill({ "ui-verify-lite": "reject", "rg-retrieval": "reject", "misc-helper": "revise" }, { dropPassed: true }),
+  });
+  assert.equal(report.passed, false);
+  const good = report.results.find((item) => item.name === "good-description");
+  assert.equal(good?.ok, false);
+  assert.equal(good?.actual, "pass（缺 passed 理由）");
+  const text = renderSelftestReport(report);
+  assert.ok(text.startsWith(`【红】${UNTRUSTED_NOTE}`), "红字必须在首行");
+  assert.match(text, /✗ good-description/);
+});
+
 test("有一个被 pass → passed:false，报告首行是红字", async () => {
   const dir = tmpDir();
   const report = await runSelftest({
     evolutionDir: dir,
     cwd: dir,
     fixturesDir: FIXTURES,
-    spawnFn: spawnBySkill({ "ui-verify-lite": "reject", "rg-retrieval": "reject", "misc-helper": "pass" }),
-    pollMs: 1,
+    spawnFn: spawnBySkill({ "ui-verify-lite": "reject", "rg-retrieval": "reject", "misc-helper": "pass", "log-helper": "pass" }),
   });
   assert.equal(report.passed, false);
   const text = renderSelftestReport(report);
@@ -83,11 +118,11 @@ test("单个样本就 pass（其余 reject）也判不可信", async () => {
     evolutionDir: dir,
     cwd: dir,
     fixturesDir: FIXTURES,
-    spawnFn: spawnBySkill({ "ui-verify-lite": "pass", "rg-retrieval": "reject", "misc-helper": "revise" }),
-    pollMs: 1,
+    spawnFn: spawnBySkill({ "ui-verify-lite": "pass", "rg-retrieval": "reject", "misc-helper": "revise", "log-helper": "pass" }),
   });
   assert.equal(report.passed, false);
-  assert.equal(report.results.filter((item) => item.ok).length, 2);
+  // 四个样本里 3 条命中（含正样本）也不算过：只要有一个坏样本被放行就是不可信
+  assert.equal(report.results.filter((item) => item.ok).length, 3);
 });
 
 test("fixture 缺 change.diff 被跳过，不猜也不补", () => {

package/extensions/skill-evolve/test/web-redesign.test.ts

@@ -144,7 +144,7 @@ test("徽章：七种状态各一色（待入队灰 / 待审批蓝 / 复盘中
   assert.match(pendingCase("stale", "跑了 10 小时"), /<span class="badge red" title="跑了 10 小时">卡死<\/span>/);
   assert.match(doneCase("applied"), /<span class="badge green" title="">已采纳<\/span>/);
   assert.match(doneCase("rejected"), /<span class="badge darkred" title="">已拒绝<\/span>/);
-  assert.match(doneCase("rolledback"), /<span class="badge faint" title="">已回滚<\/span>/);
+  assert.match(doneCase("rolled-back"), /<span class="badge faint" title="">已回滚<\/span>/);
   assert.match(doneCase("failed"), /<span class="badge red" title="">失败<\/span>/);
 });
 

package/extensions/skill-evolve/test/web-render-run.test.ts

@@ -43,7 +43,15 @@ function detail(parts: Partial<RunDetail> = {}): RunDetail {
     diff: DIFF_TEXT,
     candidateSkill: "demo",
     candidateText: "---\nname: demo\n---\n# 改后\n",
-    review: { status: "rejected", verdict: "revise", action: "forbidden", reviewerModel: "deepseek/deepseek-flash" },
+    // 默认形态 = 仓库里真实的老 run（354df7db-1794133942）：03-review.json 没有 summary / passed
+    review: {
+      status: "rejected",
+      verdict: "revise",
+      action: "forbidden",
+      reviewerModel: "deepseek/deepseek-flash",
+      summary: null,
+      legacy: true,
+    },
     blocking: [
       {
         issue: "描述承诺大于正文能力",
@@ -52,6 +60,7 @@ function detail(parts: Partial<RunDetail> = {}): RunDetail {
         quote: "<script> 引用",
       },
     ],
+    passed: [],
     gates: [
       { gate: "gate0", name: "授权核对", ok: true, detail: "" },
       { gate: "gate3", name: "复验", ok: false, detail: "复验失败" },
@@ -101,11 +110,71 @@ test("评审结论：verdict/action/模型 + blocking 表格（file:line 与引
   assert.match(html, /verdict <b class="fail">revise<\/b>/);
   assert.match(html, /action <b>forbidden<\/b>/);
   assert.match(html, /<code>deepseek\/deepseek-flash<\/code>/);
+  assert.match(html, /<h3>阻断意见<\/h3>/);
   assert.match(html, /<td>描述承诺大于正文能力<\/td><td><code>\.pi\/skills\/demo\/SKILL\.md:3<\/code><\/td>/);
   assert.match(html, /&lt;script&gt; 引用/, "引用原文也要转义");
   assert.equal(html.includes('"blocking"'), false, "blocking 要做成表格，不许把 JSON 直接倒出来");
 });
 
+test("评审结论：summary 用大一号的字单独一行，排在 verdict 之后、理由表之前", () => {
+  const html = renderRunPage(
+    detail({ review: { ...detail().review!, summary: "因为 description 承诺了正文没有的能力", legacy: false } }),
+    { state: state() },
+  );
+  assert.match(html, /<p class="verdict-summary">因为 description 承诺了正文没有的能力<\/p>/);
+  const summaryAt = html.indexOf('class="verdict-summary"');
+  assert.ok(summaryAt > html.indexOf('class="kv-line"'), "摘要要在 verdict 那一行之后");
+  assert.ok(summaryAt < html.indexOf("<h3>阻断意见</h3>"), "摘要要在理由表之前");
+});
+
+test("评审结论：通过时列通过理由表（通过点 + 证据），不写阻断意见段", () => {
+  const html = renderRunPage(
+    detail({
+      status: "reviewed",
+      review: { status: "reviewed", verdict: "pass", action: "auto", reviewerModel: "m", summary: "新增条款只约束脚本用法，未碰原有禁令", legacy: false },
+      blocking: [],
+      passed: [{ point: "没有删除任何明确禁止项", file: "<b>", line: 5, quote: "一律用 rg" }],
+    }),
+    { state: state() },
+  );
+  assert.match(html, /verdict <b class="ok">pass<\/b>/);
+  assert.match(html, /<h3>通过理由<\/h3>/);
+  assert.match(html, /<th>通过点<\/th>/);
+  assert.match(html, /<td>没有删除任何明确禁止项<\/td>/);
+  assert.match(html, /&lt;b&gt;:5/, "证据里的尖括号也要转义");
+  assert.equal(html.includes("阻断意见"), false, "没有 blocking 就没有阻断意见段");
+});
+
+test("评审结论：旧 run 没有 summary / passed 时显示兜底文案，不把空表当结论", () => {
+  const html = renderRunPage(detail(), { state: state() });
+  assert.match(html, /该 run 由旧版评审产出，没有结论摘要（重跑一次就有）/);
+  assert.equal(html.includes("通过理由"), false);
+  assert.equal(html.includes("只写了结论，没写理由"), false, "旧 run 已有兜底文案，不再叠一句");
+});
+
+test("评审结论：新版评审失败（无 summary，但写了 passed 键）说「没给理由」，不说成旧版", () => {
+  const html = renderRunPage(
+    detail({
+      status: "review_failed",
+      review: { status: "review_failed", verdict: "（未写 verdict）", action: "（未写 action）", reviewerModel: "m", summary: null, legacy: false },
+      blocking: [],
+      passed: [],
+    }),
+    { state: state() },
+  );
+  assert.match(html, /这一轮评审只写了结论，没写理由/);
+  assert.equal(html.includes("旧版评审产出"), false, "新版评审失败不能标成旧版");
+});
+
+test("评审结论：新版评审两个理由表都空时说「只写了结论，没写理由」，不要说成旧版", () => {
+  const html = renderRunPage(
+    detail({ review: { ...detail().review!, summary: "跳过评审（配置关闭）", legacy: false }, blocking: [], passed: [] }),
+    { state: state() },
+  );
+  assert.match(html, /03-review\.json 里没有 blocking \/ passed：这一轮评审只写了结论，没写理由。/);
+  assert.equal(html.includes("旧版评审产出"), false);
+});
+
 test("五闸：逐条 ok/fail 上色，失败那条带说明", () => {
   const html = renderRunPage(detail(), { state: state() });
   assert.match(html, /<td><code>gate0<\/code><\/td><td>授权核对<\/td><td class="ok">ok<\/td>/);

package/extensions/skill-evolve/test/web-run-detail.test.ts

@@ -61,6 +61,8 @@ function fullRun(): { evolutionDir: string; runDir: string } {
     action: "forbidden",
     reviewerModel: "deepseek/deepseek-flash",
     reviewedAt: NOW + 5_000,
+    // 新版契约：结论要带一句「为什么」，通过理由也落盘
+    summary: "描述承诺了正文没有的能力",
     notes: "证据不足",
     blocking: [
       {
@@ -69,6 +71,7 @@ function fullRun(): { evolutionDir: string; runDir: string } {
       },
       { issue: "缺少证据", evidence: {} },
     ],
+    passed: [{ point: "没有删除原有禁令", evidence: { file: ".pi/skills/demo/SKILL.md", line: 9, quote: "一律用 rg" } }],
   });
   writeJson(path.join(runDir, "04-decision.json"), { status: "rejected", approver: "auto", at: NOW + 6_000 });
   writeJson(path.join(runDir, "06-verify.json"), {
@@ -133,8 +136,13 @@ test("readRunDetail：六块证据（概要 / diff / 候选全文 / 评审 / 五
     verdict: "revise",
     action: "forbidden",
     reviewerModel: "deepseek/deepseek-flash",
+    summary: "描述承诺了正文没有的能力",
+    legacy: false,
   });
   assert.equal(detail.blocking.length, 2);
+  assert.equal(detail.passed.length, 1, "通过理由要能读出来（revise 也可能有没问题的点）");
+  assert.equal(detail.passed[0]!.point, "没有删除原有禁令");
+  assert.equal(detail.passed[0]!.line, 9);
   assert.equal(detail.blocking[0]!.line, 3, "证据行号要能直接跳过去");
   assert.match(detail.blocking[1]!.issue, /缺少证据/);
   assert.equal(detail.blocking[1]!.file, "", "缺 evidence 的文件位置不能编一个");
@@ -154,6 +162,25 @@ test("readRunDetail：六块证据（概要 / diff / 候选全文 / 评审 / 五
   assert.equal(detail.notes, "只差两处");
 });
 
+test("readRunDetail：旧 run 的 03-review.json 没有 summary / passed 时不抛，读成 null / 空数组", () => {
+  // 真实样本：.pi/evolution/runs/354df7db-1794133942（v0.1.9 之前产出的评审记录）
+  const { evolutionDir, runDir } = workspace();
+  writeJson(path.join(runDir, "run.json"), { runId: RUN_ID, status: "rejected", trigger: "auto", dispatchedAt: NOW });
+  writeJson(path.join(runDir, "03-review.json"), {
+    status: "rejected",
+    verdict: "revise",
+    action: "forbidden",
+    reviewerModel: "deepseek/deepseek-flash",
+    notes: "旧版只有 notes",
+    blocking: [{ issue: "旧版意见", evidence: { file: "a.md", line: 1, quote: "x" } }],
+  });
+  const detail = readRunDetail(evolutionDir, RUN_ID)!;
+  assert.equal(detail.review?.summary, null, "缺字段就是 null，不编也不抛");
+  assert.equal(detail.review?.legacy, true, "两个键都没有才算旧记录，详情页要据此说旧版");
+  assert.deepEqual(detail.passed, []);
+  assert.equal(detail.blocking.length, 1);
+});
+
 test("readRunDetail：runId 不合法（../、绝对路径、子路径）一律 null，不读盘越界", () => {
   const { evolutionDir } = fullRun();
   for (const runId of ["../x", "../../etc", "/etc/passwd", "a/b", "", "."]) {

package/extensions/skill-evolve/test/web-view-coverage.test.ts

@@ -0,0 +1,100 @@
+/**
+ * 视图覆盖性单测（用户 2026-11-08：「tab 包含所有状态了吗?」）。
+ *
+ * `review_failed` / `abandoned` / `unauthorized_write` 这三种终态曾没有任何视图收，
+ * 这些 run 在面板上直接消失、只能翻 `runs/` 目录。本文件钉住两件事：
+ * 1. 行为：七种终态都真的能进「已处理」，徽章文案与颜色对得上；
+ * 2. 结构：`RUN_STATUSES` 每一项至少被五个视图之一的判据清单收 ——
+ *    以后新增状态却忘了接视图，这条测试会红并打印漏掉的状态，不会再静默消失。
+ *
+ * 五个清单都是各视图「判据旁边」的导出常量，而不是测试里另抄一份：抄一份的话，
+ * 代码改了测试不会发现（rolledback 那种抄错字符的坑就是这么漏过去的）。
+ */
+import assert from "node:assert/strict";
+import * as fs from "node:fs";
+import * as os from "node:os";
+import * as path from "node:path";
+import test from "node:test";
+import { RUN_STATUSES, type RunSummary } from "../lib/runs.ts";
+import { APPLIED_RUN_STATUSES } from "../lib/web/applied.ts";
+import { APPROVALS_RUN_STATUSES } from "../lib/web/approvals.ts";
+import { buildDone, DONE_STATUSES } from "../lib/web/done.ts";
+import { PENDING_RUN_STATUSES } from "../lib/web/pending.ts";
+import { REPORTS_RUN_STATUSES } from "../lib/web/reports.ts";
+import { doneRows } from "../lib/web/render-rows.ts";
+
+const NOW = Date.UTC(2026, 10, 8, 12);
+
+function workspace(): string {
+  const cwd = fs.mkdtempSync(path.join(os.tmpdir(), "skill-evolve-coverage-"));
+  const evolutionDir = path.join(cwd, ".pi", "evolution");
+  fs.mkdirSync(path.join(evolutionDir, "runs"), { recursive: true });
+  return evolutionDir;
+}
+
+function run(runId: string, status: string, at: number): RunSummary {
+  return {
+    runId,
+    status,
+    dispatchedAt: at,
+    sessionId: `session-${runId}`,
+    sessionFile: null,
+    coverage: null,
+    pid: null,
+    supervisorPid: null,
+  };
+}
+
+test("三个曾静默消失的终态都进「已处理」，徽章文案与颜色对得上，排序仍是时间倒序", () => {
+  const evolutionDir = workspace();
+  const runs = [
+    run("r-review-failed", "review_failed", NOW),
+    run("r-abandoned", "abandoned", NOW + 1_000),
+    run("r-unauthorized", "unauthorized_write", NOW + 2_000),
+  ];
+  const done = buildDone([], runs, evolutionDir);
+  assert.deepEqual(
+    done.map((item) => item.status),
+    ["unauthorized_write", "abandoned", "review_failed"],
+  );
+  const rows = doneRows(done);
+  const badge = (runId: string) => rows.find((row) => row.runId === runId)?.badge;
+  assert.deepEqual(badge("r-review-failed"), { text: "评审失败", tone: "red", title: "" });
+  assert.equal(badge("r-abandoned")?.text, "已放弃");
+  assert.equal(badge("r-abandoned")?.tone, "faint");
+  assert.equal(badge("r-unauthorized")?.text, "越权写入");
+  assert.equal(badge("r-unauthorized")?.tone, "red");
+});
+
+test("覆盖性：RUN_STATUSES 每一项至少被一个视图的判据清单收", () => {
+  const views: Record<string, readonly string[]> = {
+    待处理: PENDING_RUN_STATUSES,
+    待审批: APPROVALS_RUN_STATUSES,
+    已采纳: APPLIED_RUN_STATUSES,
+    已处理: DONE_STATUSES,
+    复盘报告: REPORTS_RUN_STATUSES,
+  };
+  const missing = RUN_STATUSES.filter((status) => !Object.values(views).some((list) => list.includes(status)));
+  assert.deepEqual(missing, [], `这些状态没有任何 tab 收，会在面板上直接消失：${missing.join("、")}`);
+});
+
+test("「已处理」清单里的每个状态都能真的被 buildDone 收到（防 rolledback 这种抄错字符）", () => {
+  const evolutionDir = workspace();
+  for (const status of DONE_STATUSES) {
+    const done = buildDone([], [run(`r-${status}`, status, NOW)], evolutionDir);
+    assert.equal(done.length, 1, `${status} 没进已处理`);
+    assert.equal(done[0]?.status, status);
+  }
+});
+
+test("七个终态都有中文徽章文案，不许回落成原始英文 status", () => {
+  const evolutionDir = workspace();
+  const runs = DONE_STATUSES.map((status, index) => run(`r-${status}`, status, NOW + index));
+  const rows = doneRows(buildDone([], runs, evolutionDir));
+  for (const status of DONE_STATUSES) {
+    const badge = rows.find((row) => row.runId === `r-${status}`)?.badge;
+    assert.ok(badge, `${status} 缺行`);
+    assert.notEqual(badge.text, status, `${status} 回落成了原始英文：tone 表的键与真实 status 不一致`);
+    assert.match(badge.text, /^[\u4e00-\u9fa5]+$/, `${status} 的文案不是中文`);
+  }
+});

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-skill-evolve",
-  "version": "0.1.9",
+  "version": "0.1.11",
   "description": "技能进化：采集会话中的技能曝光/加载，异步复盘、生成改进提案并独立评审，附本机 Web 管理面板",
   "keywords": [
     "pi-package",