@theseus-whale/dsh-continuity 0.0.0-stage → 4.12.2

package/bin/public-install-check.mjs

@@ -0,0 +1,222 @@
+#!/usr/bin/env node
+/**
+ * Public install read gate.
+ *
+ * The package manager cannot force a human to read a file. This script is the
+ * supported path's gate: it asks five comprehension questions, and only writes
+ * an install acknowledgement when all answers are correct. `--verify` checks
+ * that the ack still matches the docs and package version.
+ */
+import { createHash } from 'node:crypto';
+import { existsSync, mkdirSync, readFileSync, writeFileSync } from 'node:fs';
+import { homedir } from 'node:os';
+import { dirname, join } from 'node:path';
+import { fileURLToPath } from 'node:url';
+import { createInterface } from 'node:readline';
+
+const root = dirname(dirname(fileURLToPath(import.meta.url)));
+const pkg = JSON.parse(readFileSync(join(root, 'package.json'), 'utf8'));
+const docs = {
+  baseRules: join(root, 'BASE-RULES.md'),
+  install: join(root, 'INSTALL.md'),
+  agentsTemplate: join(root, 'templates', 'AGENTS-continuity.md'),
+  rulesTemplate: join(root, 'templates', 'rules.md'),
+  habitsTemplate: join(root, 'templates', 'habits.md'),
+};
+const home = process.env.DSH_CONTINUITY_HOME
+  ? process.env.DSH_CONTINUITY_HOME
+  : join(homedir(), '.dsh', 'continuity');
+const agentsFile = join(homedir(), '.dsh', 'AGENTS.md');
+const ackFile = join(home, 'public-install-ack.json');
+const sha256 = (file) => createHash('sha256').update(readFileSync(file)).digest('hex');
+
+function readDocs() {
+  for (const file of Object.values(docs)) {
+    if (!existsSync(file)) {
+      console.error(`[read-gate] 缺少必读/模板文件：${file}`);
+      process.exit(1);
+    }
+  }
+  return {
+    baseRules: sha256(docs.baseRules),
+    install: sha256(docs.install),
+    agentsTemplate: sha256(docs.agentsTemplate),
+    rulesTemplate: sha256(docs.rulesTemplate),
+    habitsTemplate: sha256(docs.habitsTemplate),
+  };
+}
+
+function verify() {
+  const expected = readDocs();
+  if (!existsSync(ackFile)) {
+    console.error(`[read-gate] 没有 ack：${ackFile}`);
+    console.error('[read-gate] 先运行 node bin/public-install-check.mjs，读完并通过 9 题。');
+    process.exit(1);
+  }
+  let ack;
+  try { ack = JSON.parse(readFileSync(ackFile, 'utf8')); } catch (error) {
+    console.error(`[read-gate] ack 无法解析：${error.message}`);
+    process.exit(1);
+  }
+  const problems = [];
+  if (ack.version !== pkg.version) problems.push(`version ${ack.version} != ${pkg.version}`);
+  for (const key of ['baseRules', 'install', 'agentsTemplate', 'rulesTemplate', 'habitsTemplate']) {
+    if (ack.docs?.[key] !== expected[key]) problems.push(`${key} 已变化`);
+  }
+  if (!existsSync(agentsFile)) problems.push('AGENTS.md 不存在');
+  else {
+    try {
+      const agentsText = readFileSync(agentsFile, 'utf8');
+      if (!agentsText.includes('## Continuity 的 rules.md')
+        && !agentsText.includes('## Continuity 的规则层')) {
+        problems.push('AGENTS.md 缺少 Continuity 指针');
+      }
+    } catch { problems.push('AGENTS.md 不可读'); }
+  }
+  for (const file of ['rules.md', 'habits.md']) {
+    if (!existsSync(join(home, file))) problems.push(`${file} 不存在（先跑 install-templates）`);
+  }
+  if (problems.length > 0) {
+    console.error('[read-gate] ack/安装失效：' + problems.join('；'));
+    console.error('[read-gate] 先跑 node bin/install-templates.mjs --apply，再重新运行本脚本。');
+    process.exit(1);
+  }
+  console.log(`[read-gate] ack 有效：${ackFile}`);
+  console.log('[read-gate] AGENTS 指针、rules/habits 模板均存在。');
+  process.exit(0);
+}
+
+if (process.argv.includes('--verify')) verify();
+
+const hashes = readDocs();
+console.log('');
+console.log('Continuity public read gate');
+console.log(`version: ${pkg.version}`);
+console.log('');
+console.log('安装前必须完整阅读：');
+console.log(`  1. ${docs.baseRules}`);
+console.log(`  2. ${docs.install}`);
+console.log('');
+console.log('这不是形式：下面 9 题全对才会写 ack。答错不会写任何东西。');
+console.log('');
+
+const rl = createInterface({ input: process.stdin, output: process.stdout });
+let inputClosed = false;
+const lineQueue = [];
+const waiters = [];
+rl.on('line', (line) => {
+  const waiter = waiters.shift();
+  if (waiter !== undefined) waiter(line);
+  else lineQueue.push(line);
+});
+rl.on('close', () => {
+  inputClosed = true;
+  while (waiters.length > 0) waiters.shift()('');
+});
+// A line queue (not rl.question) so piped input and TTY input behave the same.
+const ask = (prompt) => {
+  process.stdout.write(prompt);
+  if (lineQueue.length > 0) return Promise.resolve(lineQueue.shift());
+  if (inputClosed) return Promise.resolve('');
+  return new Promise((resolve) => waiters.push(resolve));
+};
+let exitCode = 0;
+try {
+  await ask('读完后按 Enter 开始：');
+  const questions = [
+    {
+      text: '折叠/压缩之后，原文在哪里？',
+      options: ['A. 被删除', 'B. 存档里，可用取回读取', 'C. 只在摘要里'],
+      correct: 'B',
+    },
+    {
+      text: '这个 public 包含跨会话投递 / peer_message 吗？',
+      options: ['A. 包含', 'B. 不包含'],
+      correct: 'B',
+    },
+    {
+      text: '安装前第一件事是什么？',
+      options: ['A. 直接装', 'B. 备份、读完、跑 read gate', 'C. 清空 ~/.dsh'],
+      correct: 'B',
+    },
+    {
+      text: '线程/标签默认跨会话共享吗？',
+      options: ['A. 共享', 'B. 按会话本地存储'],
+      correct: 'B',
+    },
+    {
+      text: '装坏之后正确做法是什么？',
+      options: ['A. 删掉 ~/.dsh/continuity', 'B. 移除插件并恢复备份', 'C. 继续装直到好'],
+      correct: 'B',
+    },
+    {
+      text: '子代理能看到完整记忆/线头吗？',
+      options: ['A. 能', 'B. 不能，只看到被给的一小块'],
+      correct: 'B',
+    },
+    {
+      text: '看到 [候选]/[话题?] 里当前 T 以外的标签，最符合规则的做法是？',
+      options: [
+        'A. 忽略，自动归位已经完成',
+        'B. 当场判；更像另一条线就 load 过去',
+        'C. 只把末行 TAG 改成它，不切',
+      ],
+      correct: 'B',
+    },
+    {
+      text: '想开一个新话题，正确的 TAG 写法是？',
+      options: ['A. TAG: 新词', 'B. TAG: NEW:新词', 'C. TAG: "新词"'],
+      correct: 'B',
+    },
+    {
+      text: '安装后还必须做什么，否则规则层是断的？',
+      options: [
+        'A. 直接开始聊天',
+        'B. 合并 AGENTS 指针，并复制 rules/habits 模板',
+        'C. 删除 AGENTS.md',
+      ],
+      correct: 'B',
+    },
+  ];
+  const answers = [];
+  for (let index = 0; index < questions.length; index += 1) {
+    const item = questions[index];
+    console.log(`Q${index + 1}. ${item.text}`);
+    for (const option of item.options) console.log(`   ${option}`);
+    const answer = (await ask('你的答案（A/B/C）：')).trim().toUpperCase();
+    if (!['A', 'B', 'C'].includes(answer)) {
+      console.error(`Q${index + 1}: 只接受 A/B/C。`);
+      exitCode = 1;
+      break;
+    }
+    answers.push(answer);
+    if (answer !== item.correct) {
+      console.error('');
+      console.error(`Q${index + 1} 不对。没有写 ack。`);
+      console.error('回去读 BASE-RULES.md 的“机制事实”和 INSTALL.md 的“安装前/回滚”部分，再运行。');
+      exitCode = 1;
+      break;
+    }
+    console.log('');
+  }
+  if (exitCode === 0) {
+    mkdirSync(home, { recursive: true });
+    const ack = {
+      version: pkg.version,
+      docs: hashes,
+      answers,
+      at: new Date().toISOString(),
+    };
+    writeFileSync(ackFile, `${JSON.stringify(ack, null, 2)}\n`, 'utf8');
+    console.log(`[read-gate] 通过，已写：${ackFile}`);
+    console.log('');
+    console.log('下一步（确认路径后执行）：');
+    console.log(`  node "${join(root, 'bin', 'install-templates.mjs')}" --apply`);
+    console.log(`  dsh plugin --profile <profile> add link:${root}`);
+    console.log('');
+    console.log(`安装后复核：node "${join(root, 'bin', 'public-install-check.mjs')}" --verify`);
+  }
+} finally {
+  rl.close();
+}
+process.exitCode = exitCode;

package/bin/relabel-apply.mjs

@@ -0,0 +1,272 @@
+#!/usr/bin/env node
+/**
+ * relabel-apply.mjs — 历史重打（**只改 topic 字段**）的写盘版。
+ *
+ * 默认是演练：不带 --apply 时只算 diff、不写任何文件。
+ * 带 --apply 时必须先自动备份 bc-index.jsonl（--no-backup 才能关，**不建议**）。
+ *
+ * 只改这四个字段：
+ *   genericTopic / topicScore / topicConfidence / topicMinScore
+ * 其余字段（generic、genericWorkflow、terms、keywords、conceptScores、vec、seq…）**原样保留**。
+ * workflow 侧不动（等 baseline 够 100+ 再说）。
+ *
+ * 用法：
+ *   node bin/relabel-apply.mjs --index <bc-index.jsonl> --raw <session.jsonl[.zstd]>
+ *     [--concepts <concepts.md>] [--baseline-file <file>] [--apply] [--no-backup]
+ *     [--center-min 0.02] [--lead 0.02] [--topic-min-score 0.33] [--json]
+ */
+import fs from 'node:fs';
+import path from 'node:path';
+import zlib from 'node:zlib';
+import { fileURLToPath } from 'node:url';
+
+const here = path.dirname(fileURLToPath(import.meta.url));
+const pkgRoot = path.resolve(here, '..');
+
+const argv = process.argv.slice(2);
+const opt = (name, fallback = null) => {
+  const i = argv.indexOf(`--${name}`);
+  if (i < 0) return fallback;
+  const next = argv[i + 1];
+  return next === undefined || next.startsWith('--') ? true : next;
+};
+const num = (name, fallback) => {
+  const v = opt(name);
+  if (v === null || v === true) return fallback;
+  const n = Number(v);
+  return Number.isFinite(n) ? n : fallback;
+};
+const flag = (name) => argv.includes(`--${name}`);
+
+const indexFile = opt('index');
+const rawFile = opt('raw');
+if (!indexFile || indexFile === true || !rawFile || rawFile === true) {
+  console.error('用法：node bin/relabel-apply.mjs --index <bc-index.jsonl> --raw <session.jsonl[.zstd]> [--apply]');
+  process.exit(2);
+}
+const apply = flag('apply');
+const noBackup = flag('no-backup');
+const asJson = flag('json');
+
+const conceptsPath = (() => {
+  const explicit = opt('concepts');
+  if (explicit && explicit !== true) return explicit;
+  const candidate = path.join(pkgRoot, 'templates', 'concepts.md');
+  return fs.existsSync(candidate) ? candidate : null;
+})();
+if (!conceptsPath || !fs.existsSync(conceptsPath)) {
+  console.error('找不到 A 表：请用 --concepts 指定');
+  process.exit(2);
+}
+
+const centerMin = num('center-min', 0.02);
+const lead = num('lead', 0.02);
+const topicMinScore = num('topic-min-score', 0.33);
+const minScore = num('min-score', 0.25);
+const baselineFile = opt('baseline-file');
+
+const { createBLabeler } = await import('file://' + path.join(pkgRoot, 'src/memory/bc-label.mjs'));
+const { loadConcepts } = await import('file://' + path.join(pkgRoot, 'src/memory/concepts-file.mjs'));
+const { embedTexts } = await import('file://' + path.join(pkgRoot, 'src/facets/embedding.mjs'));
+const { eventText } = await import('file://' + path.join(pkgRoot, 'src/session-log.mjs'));
+
+const INJECTED = /^\[(?:抽取|工作线|标签|规则|习惯|技能|归位|候选|存\?|闸|回正|学习|追忆|记忆|continuity:[a-z-]+)\]/;
+const MACHINE = /^(?:Current runtime context|Current working directory|Current DSH file policy|Approval prompts|Time sampled)/;
+
+const conceptRows = loadConcepts(conceptsPath);
+const labels = Object.fromEntries(conceptRows.map((r) => [r.label, r.description]));
+const kinds = Object.fromEntries(conceptRows.map((r) => [r.label, r.kind]));
+const concepts = conceptRows.map((r) => r.label);
+const topics = conceptRows.filter((r) => r.kind === 'topic').map((r) => r.label);
+
+// ---------- raw ----------
+function readRawLines(file) {
+  if (!/\.zstd$/i.test(file)) return fs.readFileSync(file, 'utf8').replace(/^\uFEFF/, '').split(/\r?\n/).filter(Boolean);
+  const buf = fs.readFileSync(file);
+  const MAGIC = Buffer.from([0x28, 0xb5, 0x2f, 0xfd]);
+  const starts = [];
+  for (let i = 0; ; ) {
+    const k = buf.indexOf(MAGIC, i);
+    if (k < 0) break;
+    starts.push(k);
+    i = k + 4;
+  }
+  const lines = [];
+  for (const start of starts) {
+    let out;
+    try { out = zlib.zstdDecompressSync(buf.subarray(start)); } catch { continue; }
+    for (const line of out.toString('utf8').split(/\r?\n/)) if (line) lines.push(line);
+  }
+  return lines;
+}
+
+const events = [];
+const seenSeq = new Set();
+for (const line of readRawLines(rawFile)) {
+  let event;
+  try { event = JSON.parse(line.replace(/^\uFEFF/, '')); } catch { continue; }
+  const seq = Number(event?.seq);
+  if (Number.isInteger(seq)) { if (seenSeq.has(seq)) continue; seenSeq.add(seq); }
+  events.push(event);
+}
+events.sort((a, b) => Number(a.seq) - Number(b.seq));
+
+const markers = events
+  .map((event) => ({ seq: Number(event?.seq), turn: Number(event?.data?.turn) }))
+  .filter((m) => Number.isInteger(m.seq) && Number.isInteger(m.turn))
+  .sort((a, b) => a.seq - b.seq);
+const turnAt = (seq) => {
+  let found = null;
+  for (const marker of markers) { if (marker.seq > seq) break; found = marker.turn; }
+  return found ?? markers.find((m) => m.seq >= seq)?.turn ?? null;
+};
+const isHuman = (event) => {
+  const source = event?.data?.source ?? event?.source;
+  if (source && typeof source.kind === 'string' && source.kind !== 'user') return false;
+  if (source && source.peer === true) return false;
+  const text = eventText(event, { includeReasoning: false }).trim();
+  return Boolean(text) && !INJECTED.test(text) && !MACHINE.test(text);
+};
+
+const turns = new Map();
+for (const event of events) {
+  const type = event?.type;
+  if (type !== 'user/message' && type !== 'assistant/message') continue;
+  if (type === 'user/message' && !isHuman(event)) continue;
+  const text = eventText(event, { includeReasoning: false }).trim();
+  if (!text || INJECTED.test(text) || MACHINE.test(text)) continue;
+  const seq = Number(event?.seq);
+  const turn = Number.isInteger(Number(event?.data?.turn)) ? Number(event.data.turn) : turnAt(seq);
+  if (!Number.isInteger(turn)) continue;
+  const item = turns.get(turn) ?? { turn, userTexts: [], assistantTexts: [] };
+  if (type === 'user/message') item.userTexts.push(text);
+  else item.assistantTexts.push(text);
+  turns.set(turn, item);
+}
+
+// ---------- baseline（v2 / v1 / flat） ----------
+let baseline = null;
+let baselineSource = 'none(raw-bootstrap)';
+if (baselineFile && baselineFile !== true) {
+  const raw = JSON.parse(fs.readFileSync(baselineFile, 'utf8').replace(/^\uFEFF/, ''));
+  const source = raw?.topics && typeof raw.topics === 'object'
+    ? raw.topics
+    : (raw?.samples && typeof raw.samples === 'object' ? raw.samples : raw);
+  const map = Object.fromEntries(topics.map((t) => {
+    const value = source?.[t];
+    if (Array.isArray(value)) {
+      const nums = value.map((x) => Number(x)).filter((x) => Number.isFinite(x));
+      return [t, nums.length > 0 ? nums.reduce((a, b) => a + b, 0) / nums.length : Number.NaN];
+    }
+    if (value === undefined || value === null || value === '') return [t, Number.NaN];
+    const n = Number(value);
+    return [t, Number.isFinite(n) ? n : Number.NaN];
+  }));
+  const known = Object.values(map).filter((v) => Number.isFinite(v));
+  if (known.length > 0) {
+    const fallback = known.reduce((a, b) => a + b, 0) / known.length;
+    for (const t of topics) if (!Number.isFinite(map[t])) map[t] = fallback;
+    baseline = map;
+    baselineSource = `file:${baselineFile}`;
+  }
+}
+
+// ---------- 逐行重打 topic ----------
+const rows = fs.readFileSync(indexFile, 'utf8').replace(/^\uFEFF/, '').split(/\r?\n/).filter((l) => l.trim() !== '');
+const labeler = createBLabeler({ embed: embedTexts, labels, concepts, kinds });
+const updated = [];
+const stats = { rows: rows.length, relabelled: 0, unresolved: 0, topicChanged: 0, toEmpty: 0, fromEmpty: 0, same: 0 };
+const samples = [];
+
+for (const line of rows) {
+  const record = JSON.parse(line);
+  const turn = Number(record.turn);
+  const entry = Number.isInteger(turn) ? turns.get(turn) : null;
+  if (!entry) {
+    stats.unresolved += 1;
+    updated.push(line); // 原样保留
+    continue;
+  }
+  const text = [...entry.userTexts, ...entry.assistantTexts].join('\n');
+  const res = await labeler.labelChunked(text, {
+    topK: 3,
+    minScore,
+    topicMinScore,
+    topicCenterMin: centerMin,
+    topicLead: lead,
+    topicTopK: 1,
+    topicBaseline: baseline ?? undefined,
+  });
+  const newTopic = (res.topic ?? [])[0]?.concept ?? null;
+  const newScore = newTopic ? Number(res.topic?.[0]?.score ?? res.topicScore ?? 0) : null;
+  const newConfidence = newTopic ? (res.topicConfidence ?? null) : null;
+  const oldTopic = Array.isArray(record.genericTopic) && record.genericTopic.length > 0 ? record.genericTopic[0] : null;
+
+  // 按旧字段顺序重建，只替换四个打标字段
+  const next = {};
+  for (const [key, value] of Object.entries(record)) {
+    if (key === 'genericTopic' || key === 'topicScore' || key === 'topicConfidence' || key === 'topicMinScore') continue;
+    next[key] = value;
+  }
+  next.genericTopic = newTopic ? [newTopic] : [];
+  if (newTopic) {
+    next.topicScore = Number(newScore.toFixed(6));
+    if (newConfidence) next.topicConfidence = newConfidence;
+  }
+  next.topicMinScore = topicMinScore;
+
+  if (oldTopic !== newTopic) {
+    stats.topicChanged += 1;
+    if (oldTopic && !newTopic) stats.toEmpty += 1;
+    if (!oldTopic && newTopic) stats.fromEmpty += 1;
+    samples.push({ turn, oldTopic, newTopic });
+  } else {
+    stats.same += 1;
+  }
+  stats.relabelled += 1;
+  updated.push(JSON.stringify(next));
+}
+
+const result = { apply, baselineSource, stats, samples: samples.slice(0, 40) };
+
+if (!apply) {
+  if (asJson) console.log(JSON.stringify(result, null, 2));
+  else {
+    console.log(`  演练（不写盘）· index ${stats.rows} 行 · 可重打 ${stats.relabelled} · unresolved ${stats.unresolved}`);
+    console.log(`  换 topic ${stats.topicChanged}（其中 topic→空 ${stats.toEmpty} · 空→topic ${stats.fromEmpty}）· 不变 ${stats.same}`);
+    console.log(`  baseline ${baselineSource}`);
+    for (const s of result.samples.slice(0, 12)) console.log(`    turn ${s.turn}: ${s.oldTopic ?? '(空)'} → ${s.newTopic ?? '(空)'}`);
+  }
+  process.exit(0);
+}
+
+// ---------- 写盘 ----------
+const backup = `${indexFile}.bak-${new Date().toISOString().replace(/[-:.TZ]/g, '').slice(0, 14)}`;
+if (!noBackup) {
+  fs.copyFileSync(indexFile, backup);
+  console.log(`  已备份 → ${backup}（${fs.statSync(backup).size} B）`);
+} else {
+  console.log('  ⚠ --no-backup：未备份');
+}
+const tmp = `${indexFile}.tmp-${process.pid}`;
+fs.writeFileSync(tmp, updated.join('\n') + '\n', 'utf8');
+
+// 校验：行数一致、每行可解析、字段完整、非目标字段未变
+const before = rows.map((l) => JSON.parse(l));
+const after = fs.readFileSync(tmp, 'utf8').split(/\r?\n/).filter(Boolean).map((l) => JSON.parse(l));
+if (after.length !== before.length) throw new Error(`行数不一致 ${before.length} → ${after.length}`);
+const target = new Set(['genericTopic', 'topicScore', 'topicConfidence', 'topicMinScore']);
+let touchedOther = 0;
+for (let i = 0; i < before.length; i += 1) {
+  const a = before[i];
+  const b = after[i];
+  for (const key of Object.keys(a)) {
+    if (target.has(key)) continue;
+    if (JSON.stringify(a[key]) !== JSON.stringify(b[key])) touchedOther += 1;
+  }
+}
+if (touchedOther > 0) throw new Error(`非目标字段被改动 ${touchedOther} 处，已中止（原文件未动）`);
+fs.renameSync(tmp, indexFile);
+console.log(`  已写盘：${indexFile}`);
+console.log(`  重打 ${stats.relabelled} · unresolved ${stats.unresolved} · 换 topic ${stats.topicChanged}（→空 ${stats.toEmpty} · 空→topic ${stats.fromEmpty}）· 不变 ${stats.same}`);
+console.log('  校验通过：行数一致、非目标字段零改动');