@kenz1117/dsh-engram 0.2.0

package/package.json

@@ -0,0 +1,68 @@
+{
+  "name": "@kenz1117/dsh-engram",
+  "description": "Cross-session long-term memory for DeepSeek Harness: dual-scope SQLite memory graph, hybrid retrieval, provenance audit, and a knowledge flywheel.",
+  "version": "0.2.0",
+  "publishConfig": {
+    "access": "public"
+  },
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/kenz1117/dsh-engram.git"
+  },
+  "type": "module",
+  "main": "./lib/index.mjs",
+  "types": "./lib/index.d.mts",
+  "exports": {
+    ".": {
+      "types": "./lib/index.d.mts",
+      "default": "./lib/index.mjs"
+    },
+    "./src/*": "./src/*",
+    "./package.json": "./package.json"
+  },
+  "files": [
+    "lib/",
+    "scripts/",
+    "cordis.patch.yml"
+  ],
+  "dsh": {
+    "bundle": {
+      "patch": "./cordis.patch.yml"
+    }
+  },
+  "scripts": {
+    "postinstall": "node scripts/link-peers.mjs",
+    "bundle": "tsdown",
+    "watch": "tsdown --watch",
+    "test": "vitest run",
+    "typecheck": "tsc --noEmit",
+    "backfill": "node scripts/backfill-embeddings.mjs"
+  },
+  "license": "MIT",
+  "engines": {
+    "node": "^22.19.0 || >=24.0.0"
+  },
+  "peerDependencies": {
+    "@deepseek-ai/cordis": "*",
+    "@deepseek-ai/dsh-agent": "*",
+    "@deepseek-ai/dsh-llm": "*",
+    "@deepseek-ai/dsh-tools": "*",
+    "@deepseek-ai/schemastery": "*"
+  },
+  "dependencies": {
+    "@huggingface/transformers": "^4.2.0"
+  },
+  "devDependencies": {
+    "@types/node": "^26.4.0",
+    "tsdown": "^0.22.14",
+    "typescript": "^5.9.3",
+    "vitest": "^4.1.11"
+  },
+  "pnpm": {
+    "onlyBuiltDependencies": [
+      "onnxruntime-node",
+      "sharp",
+      "protobufjs"
+    ]
+  }
+}

package/scripts/backfill-embeddings.mjs

@@ -0,0 +1,63 @@
+#!/usr/bin/env node
+/**
+ * 嵌入回填：给已有但缺向量的记忆补算 embedding。
+ * 场景：嵌入器曾降级（模型未就绪）时摄取/保存的记忆没有向量，
+ * 语义检索上线后运行本脚本一次，让存量记忆参与向量道。
+ *
+ * 用法：node scripts/backfill-embeddings.mjs [dbDir]
+ * 默认 dbDir = ~/.dsh/engram；只处理 active 且 embedding IS NULL 的条目。
+ */
+
+import { DatabaseSync } from 'node:sqlite'
+import { join } from 'node:path'
+import { homedir } from 'node:os'
+import { pipeline, env } from '@huggingface/transformers'
+
+const dbDir = process.argv[2] ?? join(homedir(), '.dsh', 'engram')
+const modelCacheDir = join(dbDir, 'models')
+const MODEL_ID = 'Xenova/bge-small-zh-v1.5'
+
+const dbs = ['user.db', 'project-db'].map(name => join(dbDir, name)).filter(p => {
+  try { return new DatabaseSync(p, { readOnly: true }) && true } catch { return false }
+})
+if (dbs.length === 0) {
+  console.log('[backfill] 未找到任何记忆分库，无需回填')
+  process.exit(0)
+}
+
+env.cacheDir = modelCacheDir
+// 下载端点可经 HF_ENDPOINT 覆盖（镜像）；模型已在缓存时不会访问网络。
+if (process.env.HF_ENDPOINT) env.remoteHost = process.env.HF_ENDPOINT
+const extractor = await pipeline('feature-extraction', MODEL_ID, { dtype: 'q8' })
+const embed = async text => {
+  const output = await extractor([text], { pooling: 'mean', normalize: true })
+  return new Float32Array(output.tolist()[0])
+}
+
+let patched = 0
+for (const path of dbs) {
+  const db = new DatabaseSync(path)
+  const rows = db.prepare("SELECT id, content FROM nodes WHERE status = 'active' AND embedding IS NULL").all()
+  if (rows.length === 0) {
+    console.log(`[backfill] ${path}: 无缺向量条目`)
+    db.close()
+    continue
+  }
+  db.exec('BEGIN')
+  try {
+    for (const row of rows) {
+      const vector = await embed(row.content)
+      db.prepare('UPDATE nodes SET embedding = ? WHERE id = ?')
+        .run(new Uint8Array(vector.buffer, vector.byteOffset, vector.byteLength), row.id)
+      patched += 1
+      console.log(`[backfill] ${path}: ${row.content.slice(0, 24)}… ✓`)
+    }
+    db.exec('COMMIT')
+  } catch (error) {
+    db.exec('ROLLBACK')
+    db.close()
+    throw error
+  }
+  db.close()
+}
+console.log(`[backfill] 完成：共补算 ${patched} 条向量`)

package/scripts/link-peers.mjs

@@ -0,0 +1,57 @@
+#!/usr/bin/env node
+/**
+ * 把本插件的 @deepseek-ai/* peer 依赖从 deepseek-harness 工作区 symlink 进 node_modules。
+ *
+ * 为什么不用 file:/npm 安装：harness 内部包的 dependencies 使用 workspace:^ 协议，
+ * 脱离其工作区无法经 pnpm 解析；npm 上的 @deepseek-ai 依赖链也不完整（dsh-type-meta 未发布）。
+ * symlink 后，被链接包内部对其他 @deepseek-ai/* 的 import 会沿真实路径（harness 树内）
+ * 向上解析到 harness 自己的 workspace node_modules，传递链自动闭合。
+ *
+ * 前置：deepseek-harness 已执行 pnpm install && pnpm run build（存在 lib/ 产物）。
+ * 幂等：已存在的链接跳过。发布形态下这些包是 peerDependencies，由 dsh profile 提供。
+ */
+
+import { existsSync, mkdirSync, readdirSync, readFileSync, symlinkSync, statSync } from 'node:fs'
+import { dirname, join, resolve } from 'node:path'
+import { fileURLToPath } from 'node:url'
+
+const repoRoot = resolve(dirname(fileURLToPath(import.meta.url)), '..')
+const harnessRoot = resolve(repoRoot, '..', 'deepseek-harness')
+
+/** 直接 import 的包：类型与运行时的真实入口；其传递依赖由 harness 树内解析闭合。 */
+const LINKED_PACKAGES = [
+  ['@deepseek-ai/cordis', 'vendor/cordis'],
+  ['@deepseek-ai/cosmokit', 'vendor/cosmokit'],
+  ['@deepseek-ai/schemastery', 'vendor/schemastery'],
+  ['@deepseek-ai/cordis-plugin-loader', 'vendor/loader'],
+  ['@deepseek-ai/cordis-plugin-include', 'vendor/include'],
+  ['@deepseek-ai/dsh-tools', 'packages/core/tools'],
+  ['@deepseek-ai/dsh-system-prompt', 'packages/core/system-prompt'],
+  ['@deepseek-ai/dsh-session', 'packages/core/session'],
+  ['@deepseek-ai/dsh-llm', 'packages/llm/llm'],
+  ['@deepseek-ai/dsh-agent', 'packages/core/agent'],
+  ['@deepseek-ai/dsh-scope', 'packages/core/scope'],
+  ['@deepseek-ai/dsh-util-values', 'packages/util/values'],
+  ['@deepseek-ai/dsh-brand', 'packages/util/brand'],
+  ['@deepseek-ai/dsh-invariants', 'packages/runtime-diagnostics/invariants'],
+]
+
+const scopeDir = join(repoRoot, 'node_modules', '@deepseek-ai')
+mkdirSync(scopeDir, { recursive: true })
+
+let created = 0
+let skipped = 0
+for (const [name, rel] of LINKED_PACKAGES) {
+  const target = join(harnessRoot, rel)
+  const link = join(scopeDir, name.split('/')[1])
+  if (existsSync(link)) continue
+  // 温和跳过：发布形态或 profile 安装环境没有 harness 源码树，peer 由
+  // profile 的 node_modules 提供；链接缺失绝不使安装失败。
+  if (!existsSync(target)) {
+    skipped += 1
+    continue
+  }
+  symlinkSync(target, link, 'dir')
+  created += 1
+}
+console.log(`[link-peers] ok（新建 ${created} 个链接，跳过 ${skipped} 个无源码环境的 peer，共 ${LINKED_PACKAGES.length} 个）`)