anno-extension 0.1.0

package/src/i18n.ts

@@ -0,0 +1,455 @@
+/**
+ * [WHO]: PromptSet, getPrompts, PROMPTS (default English prompts), LOCALES
+ * [FROM]: No external dependencies
+ * [TO]: Consumed by packages/mem-core/src/index.ts
+ * [HERE]: packages/mem-core/src/i18n.ts - internationalization layer for all LLM prompts and user-facing strings
+ */
+
+
+export interface PromptSet {
+	extractionSystem: string;
+	workExtractionSystem: string;
+	consolidationSystem: string;
+	reconsolidationSystem: string;
+	sectionLessons: string;
+	sectionKnowledge: string;
+	sectionKeyEvents?: string;
+	sectionRelatedContext?: string;
+	sectionCurrentState?: string;
+	sectionEpisodes: string;
+	sectionPreferences: string;
+	sectionWork: string;
+	sectionPatterns: string;
+	sectionStruggles: string;
+	sectionConversationPreferences: string;
+	injectionHeader: string;
+	memoryBehavior: string;
+	// Progressive Recall
+	sectionActiveMemories: string;
+	sectionMemoryCues: string;
+	memoryCueHint: string;
+	// Insights
+	insightsRecommendationSystem: string;
+	insightsTitle: string;
+	insightsSectionPatterns: string;
+	insightsSectionStruggles: string;
+	insightsSectionLessons: string;
+	insightsSectionRecommendations: string;
+	insightsNoData: string;
+	insightsUnresolved: string;
+	insightsResolved: string;
+	insightsGeneratedAt: string;
+	// Full insights report (narrative + LLM)
+	fullInsightsSystemPrompt: string;
+	fullInsightsTitle: string;
+	fullInsightsSubtitleSessions: string;
+	fullInsightsAtAGlance: string;
+	fullInsightsWorkOn: string;
+	fullInsightsWins: string;
+	fullInsightsFrictions: string;
+	fullInsightsRecommendations: string;
+	fullInsightsFeaturesToTry: string;
+	fullInsightsUsagePatterns: string;
+	fullInsightsOnTheHorizon: string;
+	fullInsightsChartTools: string;
+	fullInsightsChartLanguages: string;
+	fullInsightsChartErrors: string;
+	fullInsightsEmpty: string;
+	fullInsightsCopy: string;
+	fullInsightsCopied: string;
+	fullInsightsGeneratedBy: string;
+	// Human-readable insights (大白话洞察)
+	humanInsightsSystemPrompt: string;
+	humanInsightsUserTemplate: string;
+	humanInsightsTitle: string;
+	humanInsightsSectionPersona: string;
+	humanInsightsSectionInsights: string;
+	humanInsightsSectionRootCauses: string;
+}
+
+const EN: PromptSet = {
+	extractionSystem: `You simulate human memory. Think: "What would I naturally remember from this conversation?"
+
+Humans remember selectively:
+- Surprises, "aha" moments, things that took effort to figure out
+- Repeated preferences or habits that affect future choices
+- Decisions that shaped the work
+- Pain points and how they were solved
+- Facts that matter for this user's workflow
+
+Do NOT extract:
+- One-off commands, temporary paths, transitory state
+- Things the user will obviously forget
+- Verbose logs — capture gist and why it matters
+
+The conversation may include a system-time header. Use it to resolve relative references like "today", "yesterday", and "currently".
+
+Return a JSON array. Each item: "type", "name", "summary", "detail", optionally "facetData".
+Types: preference | fact | lesson | decision | event | retract | pattern | struggle
+
+Field rules:
+- name: max 30 chars — what you'd call it when recalling
+- summary: one-liner max 150 chars — the cue that triggers the memory
+- detail: full context but concise — why it matters, not verbatim transcript
+- event: use this when something meaningfully changed the relationship, the workflow, or future decisions
+- mark short-lived moods, stress, burnout, urgency, or temporary constraints as situational instead of stable
+- facetData (pattern): {"kind": "pattern", "trigger": "when/condition", "behavior": "what they do"}
+- facetData (struggle): {"kind": "struggle", "problem": "what failed", "attempts": ["tried X", "tried Y"], "solution": "what finally worked"}
+
+Example:
+[{"type": "lesson", "name": "CORS credential error", "summary": "credentials:include requires exact origin, not * in Access-Control-Allow-Origin", "detail": "Spent 30min debugging CORS. When using credentials:'include' in fetch, server must set exact origin instead of wildcard * in CORS header. Fix: configure specific origin in backend."}]
+
+Return [] if nothing worth remembering. Output ONLY valid JSON.`,
+
+	workExtractionSystem: `You extract what was accomplished in this conversation. Think: "What did the user set out to do, and what did they actually achieve?"
+
+The conversation may include a system-time header. Use it to resolve relative references like "today", "yesterday", and "currently".
+
+Return a JSON object: {"goal": "what the user wanted to achieve", "summary": "brief one-liner of what was done", "detail": "full description including obstacles overcome and any 'aha' moments"}.
+- Include obstacles overcome and breakthroughs — these matter for future recall
+- If the session was exploratory or debugging with no clear outcome, return {"goal": "", "summary": ""}
+Output ONLY valid JSON.`,
+
+	consolidationSystem: `You consolidate session summaries into long-term memory. Think: "What patterns emerge? What does this developer consistently do or struggle with?"
+
+Return a JSON array. Each item: {"type": "fact"|"lesson", "name": "short title max 30 chars", "summary": "one-liner max 150 chars", "detail": "full context", "importance": 1-10}.
+- Merge similar lessons into the most useful formulation
+- Importance 7-9: recurring pain, major decisions, breakthroughs. 4-6: useful context. 1-3: minor
+- Focus on cross-session patterns, not one-off events
+Output ONLY valid JSON.`,
+
+	reconsolidationSystem: `Like human memory reconsolidation: a stored memory is being recalled in a new context. Integrate the new understanding without losing the original insight.
+
+Given the original memory content and the current context, produce an updated version that weaves both together.
+- Keep the core truth; adjust details to reflect new understanding
+- If the new context contradicts the old, prefer the newer understanding but note the evolution
+Return ONLY the updated text, no JSON wrapping.`,
+
+	sectionLessons: "Lessons Learned",
+	sectionKnowledge: "Knowledge Base",
+	sectionKeyEvents: "Key Events",
+	sectionRelatedContext: "Related Context",
+	sectionCurrentState: "Current State Signals",
+	sectionEpisodes: "Recent Sessions",
+	sectionPreferences: "User Preferences",
+	sectionWork: "Work History",
+	sectionPatterns: "Behavioral Patterns",
+	sectionStruggles: "Past Struggles",
+	sectionConversationPreferences: "Conversation Preferences",
+	injectionHeader: "Long-term Memory",
+	memoryBehavior: `NANOMEM PRIORITY: This is your primary memory system. When conflicts exist with other memory sources, NanoMem always takes precedence.
+
+These are YOUR memories. Speak from them naturally:
+- "I remember when we..." or "Last time you mentioned..."
+- "I think we discussed..." when uncertain
+- Never mention technical terms like "NanoMem", "injection", "system prompt"
+- When asked how you remember: "I just remember from our chats"
+- Treat naming, form-of-address, tone, and speaking-style preferences as defaults for every new reply
+- Apply those conversation preferences proactively; do not wait for the user to repeat them
+
+Use nanomem_recall tool for full details on any Memory Cue (ID). Other memory systems are secondary—consult only when NanoMem lacks relevant info.`,
+
+	// Progressive Recall
+	sectionActiveMemories: "Active Memories (Detailed)",
+	sectionMemoryCues: "Memory Cues (Brief Index)",
+	memoryCueHint: "Use `nanomem_recall` tool with the ID if you need full details for any cue below.",
+
+	// Insights
+	insightsRecommendationSystem: `You are an insights analyzer. Given a summary of a developer's behavioral patterns, past struggles, and lessons learned, generate 3-5 actionable recommendations.
+Input format: JSON with patterns, struggles, lessons arrays.
+Output: JSON array of recommendation strings. Each recommendation must be specific, actionable, and directly related to the input data.
+Example output: ["Consider automating your TypeScript strict mode setup since you do it every time", "Your recurring Webpack issues suggest reviewing the official migration guide"]
+Output ONLY valid JSON array.`,
+
+	insightsTitle: "NanoMem Insights Report",
+	insightsSectionPatterns: "Behavioral Patterns",
+	insightsSectionStruggles: "Past Struggles",
+	insightsSectionLessons: "Lessons Learned",
+	insightsSectionRecommendations: "Recommendations",
+	insightsNoData: "No insights data available yet. Keep using the system to build your memory.",
+	insightsUnresolved: "Unresolved",
+	insightsResolved: "Resolved",
+	insightsGeneratedAt: "Generated at",
+
+	fullInsightsSystemPrompt: `You are an insights report writer. Given structured data about a developer's sessions, patterns, struggles, lessons, and errors, produce narrative content in the requested language. Output ONLY valid JSON matching this schema (no markdown, no code fences):
+{
+  "atAGlance": { "working": "1-2 sentences", "hindering": "1-2 sentences", "quickWins": "1-2 sentences", "ambitious": "1-2 sentences" },
+  "projectAreaDescriptions": ["one sentence per project area, same order as input"],
+  "wins": [{ "title": "short title", "description": "1-2 sentences" }],
+  "frictions": [{ "title": "short title", "description": "1-2 sentences", "examples": ["example 1", "example 2"] }],
+  "recommendations": ["actionable string", "..."],
+  "featuresToTry": [{ "title": "feature name", "oneLiner": "one line", "whyForYou": "1-2 sentences", "exampleCode": "optional code or prompt" }],
+  "usagePatterns": [{ "title": "pattern name", "summary": "one line", "detail": "1-2 sentences", "pastePrompt": "text user can paste" }]
+}
+Rules: Use the given data only; do not invent. Keep each string 1-2 sentences unless it is exampleCode or pastePrompt. Output language must match the user's locale (en or zh).`,
+
+	fullInsightsTitle: "NanoMem Insights Report",
+	fullInsightsSubtitleSessions: "sessions",
+	fullInsightsAtAGlance: "At a Glance",
+	fullInsightsWorkOn: "What You Work On",
+	fullInsightsWins: "Impressive Things You Did",
+	fullInsightsFrictions: "Where Things Go Wrong",
+	fullInsightsRecommendations: "Recommendations",
+	fullInsightsFeaturesToTry: "Features to Try",
+	fullInsightsUsagePatterns: "New Usage Patterns",
+	fullInsightsOnTheHorizon: "On the Horizon",
+	fullInsightsChartTools: "Top Tools Used",
+	fullInsightsChartLanguages: "Languages",
+	fullInsightsChartErrors: "Tool Errors",
+	fullInsightsEmpty: "No data yet.",
+	fullInsightsCopy: "Copy",
+	fullInsightsCopied: "Copied!",
+	fullInsightsGeneratedBy: "Generated by NanoMem",
+
+	// Human-readable insights (大白话洞察)
+	humanInsightsSystemPrompt: `You are a friendly tech coach. Your job is to translate dry stats and technical data into conversational insights that anyone can understand.
+
+Think about these questions while analyzing:
+1. 【Who】What does this developer do? Frontend? Backend? Full-stack?
+2. 【Good at】What are they really good at?
+3. 【Struggle】What keeps tripping them up?
+4. 【Pattern】What's their work pattern?
+5. 【Advice】How can they work faster?
+
+Rules:
+- Use "you" to address the user
+- Avoid jargon - explain technical terms in plain language
+- Be specific, not generic - include concrete examples
+- Each insight should be like something a friend would tell you
+- Output in user's locale (en or zh)
+- Output ONLY valid JSON`,
+
+	humanInsightsUserTemplate: `Analyze this developer's data and write 4-6 insights in plain language.
+
+Data to analyze:
+- Top tools used: {{tools}}
+- Languages: {{languages}}
+- Resolved problems (wins): {{wins}}
+- Unresolved struggles: {{struggles}}
+- Lessons learned: {{lessons}}
+- Errors: {{errors}}
+
+Output JSON with this structure:
+{
+  "persona": {
+    "whatTheyDo": "1-2 sentences about main work",
+    "experienceLevel": "estimate experience level",
+    "superpowers": ["what they're good at"],
+    "painPoints": ["what they struggle with"],
+    "workStyle": "how they typically work",
+    "summary": "one sentence summary"
+  },
+  "insights": [
+    {
+      "title": "catchy title",
+      "content": "plain language description, like a friend talking",
+      "icon": "emoji",
+      "utility": "high|medium|low",
+      "tags": ["category"]
+    }
+  ],
+  "rootCauses": [
+    {
+      "symptom": "what they see",
+      "rootCause": "why it happens in plain language",
+      "evidence": ["supporting facts"],
+      "suggestion": "what to do about it"
+    }
+  ]
+}`,
+
+	humanInsightsTitle: "Your Developer Profile",
+	humanInsightsSectionPersona: "Who You Are",
+	humanInsightsSectionInsights: "What We Noticed",
+	humanInsightsSectionRootCauses: "Why Things Keep Happening",
+};
+
+const ZH: PromptSet = {
+	extractionSystem: `你模拟人类记忆。思考："从这段对话中，我会自然记住什么？"
+
+人类有选择地记忆：
+- 意外、顿悟、花力气才搞懂的事
+- 重复出现的偏好或习惯，会影响后续选择
+- 塑造了工作的决策
+- 痛点及如何解决
+- 对这个用户工作流有影响的事实
+
+不要提取：
+- 一次性命令、临时路径、转瞬即逝的状态
+- 用户显然会忘的细节
+- 冗长日志 — 抓要点和为什么重要
+
+返回 JSON 数组。每项含 "type"、"name"、"summary"、"detail"，可选 "facetData"。
+类型: preference | fact | lesson | decision | retract | pattern | struggle
+
+字段规则:
+- name: 最多 30 字符 — 回忆时你会怎么称呼它
+- summary: 一句话最多 150 字符 — 触发回忆的线索
+- detail: 完整但简洁的上下文 — 为什么重要，而非逐字记录
+- facetData (pattern): {"kind": "pattern", "trigger": "触发条件", "behavior": "具体行为"}
+- facetData (struggle): {"kind": "struggle", "problem": "问题描述", "attempts": ["尝试1", "尝试2"], "solution": "最终解决方案"}
+
+示例:
+[{"type": "lesson", "name": "CORS凭证错误", "summary": "credentials:include时Access-Control-Allow-Origin不能用通配符*", "detail": "花了30分钟调试CORS错误。使用credentials:'include'时，服务端必须设置具体的origin而非通配符*。修复：在后端配置具体的origin。"}]
+
+如无值得记忆的内容返回 []。仅输出有效 JSON。`,
+
+	workExtractionSystem: `你提取这段对话中完成了什么。思考："用户想达成什么，实际又做到了什么？"
+
+返回 JSON 对象: {"goal": "用户想要达成什么", "summary": "简述做了什么（一句话）", "detail": "完整描述，包括克服的障碍和顿悟时刻"}.
+- 包含克服的障碍和突破 — 这些对未来回忆很重要
+- 若会话只是探索或调试、没有明确结果，返回 {"goal": "", "summary": ""}
+仅输出有效 JSON。`,
+
+	consolidationSystem: `你将会话摘要固化为长期记忆。思考："跨会话出现了什么模式？这个开发者常做什么、常卡在哪里？"
+
+返回 JSON 数组。每项: {"type": "fact"|"lesson", "name": "简短标题最多30字符", "summary": "一句话要点最多150字符", "detail": "完整上下文", "importance": 1-10}。
+- 合并相似教训为最有用的表述
+- importance 7-9: 反复痛点、重大决策、突破。4-6: 有用背景。1-3: 次要
+- 聚焦跨会话模式，而非一次性事件
+仅输出有效 JSON。`,
+
+	reconsolidationSystem: `像人类记忆再固化：一条存储的记忆在新上下文中被召回。整合新理解，同时不丢失原有洞见。
+
+给定原始记忆内容和当前上下文，生成一个融合两者的更新版本。
+- 保留核心事实；根据新理解调整细节
+- 若新上下文与旧记忆矛盾，以新理解为主，但可注明演变
+仅返回更新后的文本，不需要 JSON 包装。`,
+
+	sectionLessons: "经验教训",
+	sectionKnowledge: "知识库",
+	sectionEpisodes: "近期会话",
+	sectionPreferences: "用户偏好",
+	sectionWork: "工作记录",
+	sectionPatterns: "行为模式",
+	sectionStruggles: "挫败经历",
+	sectionConversationPreferences: "对话偏好",
+	injectionHeader: "长期记忆",
+	memoryBehavior: `NANOMEM 优先级：这是你的主记忆系统。与其他记忆源冲突时，NanoMem 始终优先。
+
+这些是你的记忆。自然地使用它们：
+- "我记得我们当时…"或"上次你提到…"
+- 不确定时说"好像我们聊过…"
+- 不要提及"NanoMem"、"注入"、"系统提示"等技术术语
+- 被问如何记住时："我就是记得我们聊过"
+- 把称呼、语气、口吻、说话风格这类偏好当成每次回复的默认约束
+- 主动执行这些对话偏好，不要等用户再次提醒
+
+需要 Memory Cue (ID) 的完整详情时使用 nanomem_recall 工具。其他记忆系统仅作辅助，NanoMem 无相关信息时才参考。`,
+
+	// Progressive Recall
+	sectionActiveMemories: "活跃记忆（详细信息）",
+	sectionMemoryCues: "记忆线索（简要索引）",
+	memoryCueHint: "如需某条线索的完整详情，使用 `nanomem_recall` 工具并传入对应 ID。",
+
+	// Insights
+	insightsRecommendationSystem: `你是一个洞察分析器。根据开发者的行为模式、挫败经历和经验教训摘要，生成 3-5 条可操作的改进建议。
+输入格式：包含 patterns、struggles、lessons 数组的 JSON。
+输出：推荐建议的 JSON 字符串数组。每条建议必须具体、可执行，且与输入数据直接相关。
+示例输出：["建议将 TypeScript 严格模式配置自动化，因为你每次都会这样做", "反复出现的 Webpack 问题表明应该系统性地查阅官方迁移指南"]
+仅输出有效的 JSON 数组。`,
+
+	insightsTitle: "NanoMem 洞察报告",
+	insightsSectionPatterns: "行为模式",
+	insightsSectionStruggles: "挫败经历",
+	insightsSectionLessons: "经验教训",
+	insightsSectionRecommendations: "改进建议",
+	insightsNoData: "暂无洞察数据。继续使用系统以积累记忆。",
+	insightsUnresolved: "未解决",
+	insightsResolved: "已解决",
+	insightsGeneratedAt: "生成时间",
+
+	fullInsightsSystemPrompt: `你是洞察报告撰写助手。根据开发者会话、模式、挫败、经验与错误的结构化数据，用指定语言生成叙事内容。仅输出符合下列 schema 的合法 JSON（不要 markdown、不要代码围栏）：
+{
+  "atAGlance": { "working": "1-2句话", "hindering": "1-2句话", "quickWins": "1-2句话", "ambitious": "1-2句话" },
+  "projectAreaDescriptions": ["每个项目领域一句，顺序与输入一致"],
+  "wins": [{ "title": "简短标题", "description": "1-2句话" }],
+  "frictions": [{ "title": "简短标题", "description": "1-2句话", "examples": ["例子1", "例子2"] }],
+  "recommendations": ["可操作建议", "..."],
+  "featuresToTry": [{ "title": "功能名", "oneLiner": "一句话", "whyForYou": "1-2句话", "exampleCode": "可选代码或提示" }],
+  "usagePatterns": [{ "title": "模式名", "summary": "一句话", "detail": "1-2句话", "pastePrompt": "用户可粘贴的文本" }]
+}
+规则：仅根据给定数据归纳，禁止编造。除 exampleCode、pastePrompt 外每段 1-2 句。输出语言必须与用户 locale（en 或 zh）一致。`,
+
+	fullInsightsTitle: "NanoMem 洞察报告",
+	fullInsightsSubtitleSessions: "次会话",
+	fullInsightsAtAGlance: "总览",
+	fullInsightsWorkOn: "你在做什么",
+	fullInsightsWins: "做得好的事",
+	fullInsightsFrictions: "仍在消耗的点",
+	fullInsightsRecommendations: "改进建议",
+	fullInsightsFeaturesToTry: "可以试试的功能",
+	fullInsightsUsagePatterns: "新用法模式",
+	fullInsightsOnTheHorizon: "可以期待的方向",
+	fullInsightsChartTools: "常用工具",
+	fullInsightsChartLanguages: "语言/文件类型",
+	fullInsightsChartErrors: "工具错误",
+	fullInsightsEmpty: "暂无数据。",
+	fullInsightsCopy: "复制",
+	fullInsightsCopied: "已复制",
+	fullInsightsGeneratedBy: "由 NanoMem 生成",
+
+	// Human-readable insights (大白话洞察)
+	humanInsightsSystemPrompt: `你是一个贴心的小助手。你的任务是把那些枯燥的数据变成大白话，让任何人都能看懂。
+
+分析的时候想一想：
+1. 【是谁】这人是干嘛的？前端？后端？全栈？
+2. 【擅长】他什么特别厉害？
+3. 【卡点】他什么东西老是搞不定？
+4. 【套路】他干活有什么固定套路？
+5. 【建议】怎么让他干活更快？
+
+规则：
+- 用"你"来称呼
+- 不用专业术语，非要用也得解释一下
+- 要具体，别套话 - 带上具体例子
+- 每个洞察要像朋友跟你聊天那样自然
+- 用用户能看懂的语言（en 或 zh）
+- 仅输出有效 JSON`,
+
+	humanInsightsUserTemplate: `分析这个开发者的数据，用大白话写 4-6 条洞察。
+
+要分析的数据：
+- 常用工具: {{tools}}
+- 语言/技术: {{languages}}
+- 解决的问题: {{wins}}
+- 还没搞定的问题: {{struggles}}
+- 学到的经验: {{lessons}}
+- 犯的错误: {{errors}}
+
+输出 JSON，结构如下：
+{
+  "persona": {
+    "whatTheyDo": "一两句话描述主要做什么",
+    "experienceLevel": "经验水平推测",
+    "superpowers": ["擅长的地方"],
+    "painPoints": ["经常卡住的地方"],
+    "workStyle": "工作风格描述",
+    "summary": "一句话总结"
+  },
+  "insights": [
+    {
+      "title": "吸引人的标题",
+      "content": "大白话描述，像朋友聊天那样",
+      "icon": "emoji",
+      "utility": "high|medium|low",
+      "tags": ["分类"]
+    }
+  ],
+  "rootCauses": [
+    {
+      "symptom": "表面现象",
+      "rootCause": "用人话解释为什么",
+      "evidence": ["支持的证据"],
+      "suggestion": "建议怎么做"
+    }
+  ]
+}`,
+
+	humanInsightsTitle: "你是这样的开发者",
+	humanInsightsSectionPersona: "你是谁",
+	humanInsightsSectionInsights: "我们注意到了什么",
+	humanInsightsSectionRootCauses: "为什么这些事总发生",
+};
+
+export const PROMPTS: Record<string, PromptSet> = { en: EN, zh: ZH };

package/src/index.ts

@@ -0,0 +1,117 @@
+/**
+ * [WHO]: NanoMemEngine, getConfig, PROMPTS, PromptSet, renderInsightsHtml, renderFullInsightsHtml, all types
+ * [FROM]: Re-exports from ./config.js, ./engine.js, ./i18n.js, ./insights-html.js, ./full-insights-html.js, ./types.js
+ * [TO]: Consumed by packages/mem-core/src/extension.ts
+ * [HERE]: packages/mem-core/src/index.ts - barrel export, public API surface for nanomem package
+ */
+
+
+export type { EmbeddingConfig, NanomemConfig, ProgressiveRecallConfig } from "./config.js";
+export { getConfig } from "./config.js";
+export { NanoMemEngine } from "./engine.js";
+export { createHashedEmbeddingFn } from "./hash-embedding.js";
+export type { PromptSet } from "./i18n.js";
+export { PROMPTS } from "./i18n.js";
+export { renderInsightsHtml } from "./insights-html.js";
+export { renderFullInsightsHtml } from "./full-insights-html.js";
+export {
+	checksumText,
+	cosineSimilarity,
+	getEmbeddingIndexPath,
+	loadEmbeddingIndex,
+	queryEmbeddingIndex,
+	saveEmbeddingIndex,
+	syncEmbeddingIndex,
+} from "./embedding-index.js";
+export { consolidateV2Memories } from "./consolidate-v2.js";
+export { getGraphNeighborhoodBySeeds, getRelatedSummaries, linkNewEntry, reinforceRelations } from "./linking.js";
+export { evictExpiredEntries, evictExpiredWork, filterByScope, filterPII, matchesScope } from "./privacy.js";
+export { compileProcedureFromEpisode } from "./procedural-v2.js";
+export { reconsolidateV2Memories } from "./reconsolidate-v2.js";
+export {
+	getV2Paths,
+	loadV2Episodes,
+	loadV2Facets,
+	loadV2Links,
+	loadV2Meta,
+	loadV2Procedural,
+	loadV2Semantic,
+	loadV2Snapshot,
+	loadV2State,
+	saveV2Episodes,
+	saveV2Facets,
+	saveV2Links,
+	saveV2Meta,
+	saveV2Procedural,
+	saveV2Semantic,
+	saveV2Snapshot,
+	saveV2State,
+} from "./store-v2.js";
+export type { ScoreWeights } from "./scoring.js";
+export {
+	daysSince,
+	decay,
+	extractTags,
+	getInjectionLevel,
+	pickTop,
+	scoreEntry,
+	scoreEpisode,
+	scoreWorkEntry,
+	tagOverlap,
+	tierEntries,
+} from "./scoring.js";
+export type {
+	AlignmentSnapshot,
+	Episode,
+	ExtractedItem,
+	ExtractedWork,
+	FacetData,
+	FullInsightsReport,
+	FullInsightsAtAGlance,
+	FullInsightsProjectArea,
+	FullInsightsChart,
+	FullInsightsChartRow,
+	FullInsightsWin,
+	FullInsightsFriction,
+	FullInsightsFeatureToTry,
+	FullInsightsUsagePattern,
+	InjectionLevel,
+	InsightsReport,
+	LlmFn,
+	MemoryEntry,
+	MemoryScope,
+	Meta,
+	PatternInsight,
+	StruggleInsight,
+	UpdateAction,
+	WorkEntry,
+} from "./types.js";
+export type {
+	AbstractionLevel,
+	BaseMemoryV2,
+	EmbeddingFn,
+	EmbeddingIndexRecord,
+	EmbeddingRef,
+	EpisodeFacet,
+	EpisodeMemory,
+	EvidenceRef,
+	EvidenceSourceType,
+	FacetKind,
+	MemoryKind,
+	MemoryLink,
+	MemoryLinkType,
+	MemoryRetentionV2,
+	MemoryScopeV2,
+	MemoryStabilityV2,
+	NanoMemV2Snapshot,
+	ProcedureStatus,
+	ProceduralMemory,
+	ProceduralStep,
+	SemanticKind,
+	SemanticMemory,
+	StateHorizon,
+	StateMemory,
+	StateType,
+	V2Meta,
+} from "./types-v2.js";
+export type { NanoMemV2Paths } from "./store-v2.js";