newtype-profile 1.0.47 → 1.0.49

package/README.md

@@ -329,18 +329,23 @@ newtype-profile includes an automatic memory system for cross-session knowledge
 ### How It Works
 
 1. **Auto-save**: When a conversation ends (session.idle), key information is extracted and saved to `.opencode/memory/YYYY-MM-DD.md`
-2. **Auto-archive**: Logs older than 7 days are automatically consolidated into `.opencode/MEMORY.md`
-3. **AI Awareness**: Chief knows about the memory system and can query it when needed
+2. **Full transcript**: A complete conversation log is stored in `.opencode/memory/full/<sessionID>.md` (overwrites per session)
+3. **Auto-archive**: Logs older than 7 days are consolidated into `.opencode/MEMORY.md`
+4. **Deep summaries**: If a daily entry includes Decisions/TODOs or tags (#project/#preference/#policy/#important), the archivist agent summarizes the full transcript in background
+5. **AI Awareness**: Chief knows about the memory system and can query it when needed
 
 ### File Structure
 
 ```
 your-project/
 └── .opencode/
-    ├── MEMORY.md              # Long-term memory (archived)
+    ├── MEMORY.md              # Long-term memory (archived + deep summaries)
     └── memory/
-        ├── 2026-01-29.md      # Today's conversation log
-        ├── 2026-01-28.md      # Yesterday's log
+        ├── 2026-01-29.md      # Daily summaries
+        ├── 2026-01-28.md      # Daily summaries
+        ├── full/
+        │   ├── ses_xxxx.md    # Full transcript per session
+        │   └── ...
         └── ...
 ```
 

package/README.zh-cn.md

@@ -329,18 +329,23 @@ newtype-profile 内置了跨会话记忆系统，自动保存重要信息：
 ### 工作原理
 
 1. **自动保存**：对话结束时（session.idle），关键信息被提取并保存到 `.opencode/memory/YYYY-MM-DD.md`
-2. **自动归档**：超过 7 天的日志自动合并到 `.opencode/MEMORY.md`
-3. **AI 感知**：Chief 知道记忆系统的存在，需要时会主动查询
+2. **完整对话**：每个 session 的全文日志保存到 `.opencode/memory/full/<sessionID>.md`（覆盖写）
+3. **自动归档**：超过 7 天的日志自动合并到 `.opencode/MEMORY.md`
+4. **深度摘要**：当日记含 Decisions/TODOs 或标签（#project/#preference/#policy/#important）时，archivist 会后台读取全文生成深度摘要
+5. **AI 感知**：Chief 知道记忆系统的存在，需要时会主动查询
 
 ### 文件结构
 
 ```
 你的项目/
 └── .opencode/
-    ├── MEMORY.md              # 长期记忆（归档后）
+    ├── MEMORY.md              # 长期记忆（含深度摘要）
     └── memory/
-        ├── 2026-01-29.md      # 今天的对话记录
-        ├── 2026-01-28.md      # 昨天的记录
+        ├── 2026-01-29.md      # 每日摘要
+        ├── 2026-01-28.md      # 每日摘要
+        ├── full/
+        │   ├── ses_xxxx.md    # 每个 session 的完整记录
+        │   └── ...
         └── ...
 ```
 

package/dist/cli/index.js

@@ -2253,7 +2253,7 @@ var require_picocolors = __commonJS((exports, module) => {
 var require_package = __commonJS((exports, module) => {
   module.exports = {
     name: "newtype-profile",
-    version: "1.0.47",
+    version: "1.0.49",
     description: "AI Agent Collaboration System for Content Creation - Based on oh-my-opencode",
     main: "dist/index.js",
     types: "dist/index.d.ts",

package/dist/features/builtin-commands/templates/memory-consolidate.d.ts

@@ -1 +1 @@
-export declare const MEMORY_CONSOLIDATE_TEMPLATE = "Consolidate daily memory logs into long-term memory.\n\n## USAGE\n\n`/memory-consolidate`\n\n## WHAT TO DO\n\nRead all files in `.opencode/memory/` directory and consolidate important information into `.opencode/MEMORY.md`.\n\n### Step 1: Read Daily Logs\n\nUse `glob` to find all memory files:\n```\nglob(\".opencode/memory/*.md\")\n```\n\nThen `read` each file to understand the contents.\n\n### Step 2: Extract Key Information\n\nFrom the daily logs, identify:\n1. **User Preferences** - Recurring patterns in how user likes things done\n2. **Important Decisions** - Choices made that affect future work\n3. **Project Milestones** - Significant completions or achievements\n4. **Recurring Problems** - Issues that came up multiple times and their solutions\n5. **Key Insights** - Valuable learnings worth preserving\n\n### Step 3: Update MEMORY.md\n\nRead existing `.opencode/MEMORY.md` (create if doesn't exist).\n\nAppend new consolidated information using this structure:\n\n```markdown\n## Consolidated: YYYY-MM-DD\n\n### User Preferences\n- [preference 1]\n- [preference 2]\n\n### Decisions Made\n- [decision with context]\n\n### Lessons Learned\n- [insight or lesson]\n\n---\n```\n\n### Step 4: Archive Processed Logs\n\nAfter consolidation, you may suggest to the user whether to:\n- Keep the daily logs for reference\n- Delete processed daily logs to save space\n\n## OUTPUT FORMAT\n\nAfter consolidation:\n```\n\u2705 Memory consolidated\n\n**Processed:** X daily log files\n**Extracted:**\n- Y user preferences\n- Z decisions\n- W insights\n\n**Updated:** .opencode/MEMORY.md\n\n\uD83D\uDCA1 Tip: Daily logs in .opencode/memory/ are preserved. \n   Delete them manually if you want to save space.\n```\n\nIf no memory files found:\n```\n\uD83D\uDCED No memory files to consolidate\n\nMemory files are automatically created after conversations.\nCheck back after a few chat sessions.\n```\n\n## IMPORTANT\n\n- Never delete information from MEMORY.md, only append\n- Preserve existing structure and content\n- Be selective - only consolidate truly important information\n- Use user's language (Chinese/English) based on existing content\n";
+export declare const MEMORY_CONSOLIDATE_TEMPLATE = "Consolidate daily memory logs into long-term memory.\n\n## USAGE\n\n`/memory-consolidate`\n\n## WHAT TO DO\n\nRead all files in `.opencode/memory/` directory and consolidate important information into `.opencode/MEMORY.md`.\nIf a daily log entry contains Decisions, TODOs, or critical tags (#project, #preference, #policy, #important),\nalso pull the full transcript from `.opencode/memory/full/<sessionID>.md` for deeper summarization.\n\n### Step 1: Read Daily Logs\n\nUse `glob` to find all memory files:\n```\nglob(\".opencode/memory/*.md\")\n```\n\nThen `read` each file to understand the contents.\n\n### Step 2: Extract Key Information\n\nFrom the daily logs, identify:\n1. **User Preferences** - Recurring patterns in how user likes things done\n2. **Important Decisions** - Choices made that affect future work\n3. **Project Milestones** - Significant completions or achievements\n4. **Recurring Problems** - Issues that came up multiple times and their solutions\n5. **Key Insights** - Valuable learnings worth preserving\n\nIf deep-summary trigger conditions are met, read the full transcript for that session and extract additional\npreferences, decisions, and lessons.\n\n### Step 3: Update MEMORY.md\n\nRead existing `.opencode/MEMORY.md` (create if doesn't exist).\n\nAppend new consolidated information using this structure:\n\n```markdown\n## Consolidated: YYYY-MM-DD\n\n### User Preferences\n- [preference 1]\n- [preference 2]\n\n### Decisions Made\n- [decision with context]\n\n### Lessons Learned\n- [insight or lesson]\n\n### Deep Summaries (when triggered)\n- [sessionID] preference/decision/lesson from full transcript\n\n---\n```\n\n### Step 4: Archive Processed Logs\n\nAfter consolidation, you may suggest to the user whether to:\n- Keep the daily logs for reference\n- Delete processed daily logs to save space\n\n## OUTPUT FORMAT\n\nAfter consolidation:\n```\n\u2705 Memory consolidated\n\n**Processed:** X daily log files\n**Extracted:**\n- Y user preferences\n- Z decisions\n- W insights\n\n**Updated:** .opencode/MEMORY.md\n\n\uD83D\uDCA1 Tip: Daily logs in .opencode/memory/ are preserved. \n   Delete them manually if you want to save space.\n```\n\nIf no memory files found:\n```\n\uD83D\uDCED No memory files to consolidate\n\nMemory files are automatically created after conversations.\nCheck back after a few chat sessions.\n```\n\n## IMPORTANT\n\n- Never delete information from MEMORY.md, only append\n- Preserve existing structure and content\n- Be selective - only consolidate truly important information\n- Use user's language (Chinese/English) based on existing content\n";

package/dist/hooks/memory-system/constants.d.ts

@@ -1,8 +1,8 @@
 export declare const HOOK_NAME = "memory-system";
-/** Minimum number of messages to trigger memory save */
-export declare const MIN_MESSAGES_TO_SAVE = 4;
 /** Memory storage directory relative to project root */
 export declare const MEMORY_DIR = ".opencode/memory";
+/** Full transcript storage directory relative to project root */
+export declare const FULL_MEMORY_DIR = ".opencode/memory/full";
 /** Long-term memory file */
 export declare const MEMORY_FILE = ".opencode/MEMORY.md";
 /** Grace period before saving (ms) - allows for quick follow-ups */
@@ -11,3 +11,5 @@ export declare const SAVE_GRACE_PERIOD_MS = 5000;
 export declare const MAX_SUMMARY_LENGTH = 2000;
 /** Days after which daily logs are auto-archived to MEMORY.md */
 export declare const ARCHIVE_AFTER_DAYS = 7;
+/** Tags that trigger deep summary from full transcripts */
+export declare const DEEP_SUMMARY_TAGS: readonly ["#project", "#preference", "#policy", "#important"];

package/dist/hooks/memory-system/storage.d.ts

@@ -1,10 +1,34 @@
-import type { MemoryEntry } from "./types";
+import type { MemoryEntry, MemoryEntryMessage } from "./types";
 export interface ArchiveResult {
     archived: string[];
     totalFiles: number;
     needsArchive: boolean;
 }
+export interface FullTranscriptBlock {
+    role: string;
+    content: string;
+}
+export interface MemorySummary {
+    userPreferences: string[];
+    decisions: string[];
+    lessons: string[];
+}
+export interface DailyLogSession {
+    sessionID?: string;
+    raw: string;
+    tags: string[];
+    decisions: string[];
+    todos: string[];
+}
 export declare function checkArchiveNeeded(projectDir: string): ArchiveResult;
-export declare function archiveOldMemories(projectDir: string): ArchiveResult;
+export declare function archiveOldMemories(projectDir: string, options?: {
+    deepSummarizer?: (session: DailyLogSession, fullContent: string) => Promise<string | null>;
+}): Promise<ArchiveResult>;
+export declare function parseDailyLogSessions(content: string): DailyLogSession[];
+export declare function shouldDeepSummarize(session: DailyLogSession): boolean;
+export declare function getFullTranscriptPath(projectDir: string, sessionID: string): string;
+export declare function parseFullTranscript(content: string): FullTranscriptBlock[];
+export declare function summarizeFullTranscript(content: string): MemorySummary;
 export declare function appendMemoryEntry(projectDir: string, entry: MemoryEntry): boolean;
 export declare function hasMemoryForSession(projectDir: string, sessionID: string): boolean;
+export declare function saveFullTranscript(projectDir: string, sessionID: string, messages: MemoryEntryMessage[]): boolean;

package/dist/hooks/memory-system/types.d.ts

@@ -5,6 +5,12 @@ export interface MemoryEntry {
     keyPoints: string[];
     decisions?: string[];
     todos?: string[];
+    tags?: string[];
+}
+export interface MemoryEntryMessage {
+    role: string;
+    text: string;
+    timestamp?: string;
 }
 export interface SessionState {
     saved: boolean;

package/dist/index.js

@@ -24326,15 +24326,24 @@ ${buildStandaloneVerificationReminder(subagentSessionId)}
 }
 // src/hooks/memory-system/constants.ts
 var HOOK_NAME6 = "memory-system";
-var MIN_MESSAGES_TO_SAVE = 4;
 var MEMORY_DIR = ".opencode/memory";
+var FULL_MEMORY_DIR = ".opencode/memory/full";
 var MEMORY_FILE = ".opencode/MEMORY.md";
 var SAVE_GRACE_PERIOD_MS = 5000;
 var MAX_SUMMARY_LENGTH = 2000;
 var ARCHIVE_AFTER_DAYS = 7;
+var DEEP_SUMMARY_TAGS = ["#project", "#preference", "#policy", "#important"];
 
 // src/hooks/memory-system/storage.ts
-import { existsSync as existsSync40, mkdirSync as mkdirSync12, appendFileSync as appendFileSync6, readFileSync as readFileSync26, readdirSync as readdirSync15, unlinkSync as unlinkSync10 } from "fs";
+import {
+  existsSync as existsSync40,
+  mkdirSync as mkdirSync12,
+  appendFileSync as appendFileSync6,
+  readFileSync as readFileSync26,
+  readdirSync as readdirSync15,
+  unlinkSync as unlinkSync10,
+  writeFileSync as writeFileSync16
+} from "fs";
 import { join as join48 } from "path";
 function ensureMemoryDir(projectDir) {
   const memoryPath = join48(projectDir, MEMORY_DIR);
@@ -24343,6 +24352,13 @@ function ensureMemoryDir(projectDir) {
   }
   return memoryPath;
 }
+function ensureFullMemoryDir(projectDir) {
+  const memoryPath = join48(projectDir, FULL_MEMORY_DIR);
+  if (!existsSync40(memoryPath)) {
+    mkdirSync12(memoryPath, { recursive: true });
+  }
+  return memoryPath;
+}
 function getDateFileName() {
   const now = new Date;
   return `${now.toISOString().split("T")[0]}.md`;
@@ -24367,6 +24383,14 @@ function getDaysDiff(date) {
   date.setHours(0, 0, 0, 0);
   return Math.floor((now.getTime() - date.getTime()) / (1000 * 60 * 60 * 24));
 }
+function extractSectionItems(block, title) {
+  const pattern = new RegExp(`\\*\\*${title}:\\*\\*\\n([\\s\\S]*?)(?:\\n\\*\\*|$)`);
+  const match = block.match(pattern);
+  if (!match)
+    return [];
+  return match[1].split(`
+`).map((line) => line.trim()).filter((line) => line.startsWith("- ")).map((line) => line.replace(/^\-\s+/, "")).filter(Boolean);
+}
 function checkArchiveNeeded(projectDir) {
   const memoryDir = join48(projectDir, MEMORY_DIR);
   if (!existsSync40(memoryDir)) {
@@ -24386,7 +24410,7 @@ function checkArchiveNeeded(projectDir) {
     needsArchive: oldFiles.length > 0
   };
 }
-function archiveOldMemories(projectDir) {
+async function archiveOldMemories(projectDir, options) {
   const checkResult = checkArchiveNeeded(projectDir);
   if (!checkResult.needsArchive) {
     return checkResult;
@@ -24403,13 +24427,65 @@ function archiveOldMemories(projectDir) {
     try {
       const filePath = join48(memoryDir, file);
       const content = readFileSync26(filePath, "utf-8");
+      const sessions = parseDailyLogSessions(content);
       const dateMatch = file.match(/^(\d{4}-\d{2}-\d{2})/);
       const dateStr = dateMatch ? dateMatch[1] : file;
       archivedContent.push(`### From ${dateStr}
 
 `);
-      const contentWithoutHeader = content.replace(/^# Memory Log.*\n\n?/, "");
-      archivedContent.push(contentWithoutHeader);
+      const summaryBlocks = [];
+      const deepBlocks = [];
+      for (const session of sessions) {
+        summaryBlocks.push(session.raw, "", "---", "");
+        if (session.sessionID && shouldDeepSummarize(session)) {
+          const fullPath = getFullTranscriptPath(projectDir, session.sessionID);
+          if (existsSync40(fullPath)) {
+            const fullContent = readFileSync26(fullPath, "utf-8");
+            let deepSummary = null;
+            if (options?.deepSummarizer) {
+              deepSummary = await options.deepSummarizer(session, fullContent);
+            }
+            if (!deepSummary) {
+              const summary = summarizeFullTranscript(fullContent);
+              if (summary.userPreferences.length > 0 || summary.decisions.length > 0 || summary.lessons.length > 0) {
+                deepSummary = [
+                  summary.userPreferences.length > 0 ? [
+                    "**User Preferences:**",
+                    ...summary.userPreferences.map((item) => `- ${item}`),
+                    ""
+                  ].join(`
+`) : "",
+                  summary.decisions.length > 0 ? [
+                    "**Decisions Made:**",
+                    ...summary.decisions.map((item) => `- ${item}`),
+                    ""
+                  ].join(`
+`) : "",
+                  summary.lessons.length > 0 ? [
+                    "**Lessons Learned:**",
+                    ...summary.lessons.map((item) => `- ${item}`),
+                    ""
+                  ].join(`
+`) : ""
+                ].filter(Boolean).join(`
+`);
+              }
+            }
+            if (deepSummary) {
+              deepBlocks.push(`#### Deep Summary (${session.sessionID.slice(0, 12)})`, "", deepSummary.trim(), "", "---", "");
+            }
+          }
+        }
+      }
+      archivedContent.push(summaryBlocks.join(`
+`));
+      if (deepBlocks.length > 0) {
+        archivedContent.push(`#### Deep Summaries
+
+`);
+        archivedContent.push(deepBlocks.join(`
+`));
+      }
       unlinkSync10(filePath);
     } catch {
       continue;
@@ -24430,12 +24506,108 @@ This file contains archived conversation memories.
     needsArchive: false
   };
 }
+function parseDailyLogSessions(content) {
+  const cleaned = content.replace(/^# Memory Log.*\n\n?/, "");
+  const blocks = cleaned.split(/\n---\n/).map((block) => block.trim()).filter(Boolean);
+  return blocks.map((block) => {
+    const sessionMatch = block.match(/SessionID:\s*(.+)/);
+    const tags = extractSectionItems(block, "Tags");
+    const decisions = extractSectionItems(block, "Decisions");
+    const todos = extractSectionItems(block, "TODOs").map((item) => item.replace(/^\[ \]\s+/, ""));
+    return {
+      sessionID: sessionMatch?.[1]?.trim(),
+      raw: block,
+      tags,
+      decisions,
+      todos
+    };
+  });
+}
+function shouldDeepSummarize(session) {
+  if (session.decisions.length > 0 || session.todos.length > 0)
+    return true;
+  return session.tags.some((tag) => DEEP_SUMMARY_TAGS.includes(tag.toLowerCase()));
+}
+function getFullTranscriptPath(projectDir, sessionID) {
+  const safeSessionID = sessionID.replace(/[^a-zA-Z0-9_-]/g, "_");
+  return join48(projectDir, FULL_MEMORY_DIR, `${safeSessionID}.md`);
+}
+function parseFullTranscript(content) {
+  const cleaned = content.replace(/^# Full Transcript.*\n.*\n\n?/, "");
+  const blocks = cleaned.split(/\n---\n/).map((block) => block.trim()).filter(Boolean);
+  return blocks.map((block) => {
+    const lines = block.split(`
+`);
+    const title = lines.shift() || "";
+    const roleMatch = title.match(/^##\s+([A-Z]+)\b/);
+    return {
+      role: roleMatch?.[1]?.toLowerCase() || "unknown",
+      content: lines.join(`
+`).trim()
+    };
+  });
+}
+function summarizeFullTranscript(content) {
+  const blocks = parseFullTranscript(content);
+  const userPreferences = [];
+  const decisions = [];
+  const lessons = [];
+  for (const block of blocks) {
+    const text = block.content;
+    const preferencePatterns = [
+      /\u504F\u597D[\uFF1A:](.+)/g,
+      /\u559C\u6B22[\uFF1A:](.+)/g,
+      /prefer(?:s|red)?[\uFF1A:](.+)/gi,
+      /preference(?:s)?[\uFF1A:](.+)/gi
+    ];
+    for (const pattern of preferencePatterns) {
+      for (const match of text.matchAll(pattern)) {
+        if (match[1])
+          userPreferences.push(match[1].trim());
+      }
+    }
+    const decisionPatterns = [
+      /\u51B3\u5B9A[\uFF1A:](.+)/g,
+      /\u51B3\u7B56[\uFF1A:](.+)/g,
+      /decided?[\uFF1A:](.+)/gi,
+      /chosen?[\uFF1A:](.+)/gi,
+      /going with\s+(.+)/gi,
+      /will use\s+(.+)/gi
+    ];
+    for (const pattern of decisionPatterns) {
+      for (const match of text.matchAll(pattern)) {
+        if (match[1])
+          decisions.push(match[1].trim());
+      }
+    }
+    const lessonPatterns = [
+      /\u7ED3\u8BBA[\uFF1A:](.+)/g,
+      /\u7ECF\u9A8C[\uFF1A:](.+)/g,
+      /\u6559\u8BAD[\uFF1A:](.+)/g,
+      /lesson(?:s)?[\uFF1A:](.+)/gi,
+      /insight(?:s)?[\uFF1A:](.+)/gi
+    ];
+    for (const pattern of lessonPatterns) {
+      for (const match of text.matchAll(pattern)) {
+        if (match[1])
+          lessons.push(match[1].trim());
+      }
+    }
+  }
+  const unique = (items) => Array.from(new Set(items)).slice(0, 10);
+  return {
+    userPreferences: unique(userPreferences),
+    decisions: unique(decisions),
+    lessons: unique(lessons)
+  };
+}
 function appendMemoryEntry(projectDir, entry) {
   try {
     const memoryDir = ensureMemoryDir(projectDir);
     const filePath = join48(memoryDir, getDateFileName());
     const sections = [
       `## Session: ${entry.sessionID.slice(0, 12)} (${formatTime(new Date(entry.timestamp))})`,
+      `SessionID: ${entry.sessionID}`,
       ""
     ];
     if (entry.summary) {
@@ -24457,6 +24629,11 @@ function appendMemoryEntry(projectDir, entry) {
       entry.todos.forEach((todo) => sections.push(`- [ ] ${todo}`));
       sections.push("");
     }
+    if (entry.tags && entry.tags.length > 0) {
+      sections.push("**Tags:**");
+      entry.tags.forEach((tag) => sections.push(`- ${tag}`));
+      sections.push("");
+    }
     sections.push("---", "");
     const content = sections.join(`
 `);
@@ -24480,7 +24657,28 @@ function hasMemoryForSession(projectDir, sessionID) {
     if (!existsSync40(filePath))
       return false;
     const content = readFileSync26(filePath, "utf-8");
-    return content.includes(`Session: ${sessionID.slice(0, 12)}`);
+    return content.includes(`SessionID: ${sessionID}`) || content.includes(`Session: ${sessionID.slice(0, 12)}`);
+  } catch {
+    return false;
+  }
+}
+function saveFullTranscript(projectDir, sessionID, messages) {
+  try {
+    const memoryDir = ensureFullMemoryDir(projectDir);
+    const safeSessionID = sessionID.replace(/[^a-zA-Z0-9_-]/g, "_");
+    const filePath = join48(memoryDir, `${safeSessionID}.md`);
+    const sections = [
+      `# Full Transcript - ${sessionID}`,
+      `Generated: ${new Date().toISOString()}`,
+      ""
+    ];
+    for (const message of messages) {
+      const timestamp2 = message.timestamp ? ` (${message.timestamp})` : "";
+      sections.push(`## ${message.role.toUpperCase()}${timestamp2}`, "", message.text || "", "", "---", "");
+    }
+    writeFileSync16(filePath, sections.join(`
+`));
+    return true;
   } catch {
     return false;
   }
@@ -24499,10 +24697,17 @@ function truncateText(text, maxLength) {
 function extractSessionSummary(sessionID, messages) {
   const userMessages = [];
   const assistantMessages = [];
+  const tags = [];
   for (const msg of messages) {
     const text = extractTextFromParts(msg.parts);
     if (!text.trim())
       continue;
+    const tagMatches = text.matchAll(/#([a-zA-Z][\w-]{1,30})/g);
+    for (const match of tagMatches) {
+      const tag = match[1]?.toLowerCase();
+      if (tag)
+        tags.push(`#${tag}`);
+    }
     if (msg.info.role === "user") {
       userMessages.push(truncateText(text, 500));
     } else if (msg.info.role === "assistant") {
@@ -24559,15 +24764,60 @@ function extractSessionSummary(sessionID, messages) {
     summary,
     keyPoints: [...new Set(keyPoints)].slice(0, 5),
     decisions: [...new Set(decisions)].slice(0, 3),
-    todos: [...new Set(todos)].slice(0, 3)
+    todos: [...new Set(todos)].slice(0, 3),
+    tags: [...new Set(tags)].slice(0, 5)
   };
 }
 
 // src/hooks/memory-system/index.ts
+function extractMessageText(parts) {
+  return parts.filter((p) => p.type === "text" && p.text).map((p) => p.text).join(`
+`);
+}
+function extractFullTranscript(messages) {
+  return messages.map((message) => {
+    const text = extractMessageText(message.parts);
+    return {
+      role: message.info.role,
+      text
+    };
+  }).filter((message) => message.text.trim().length > 0);
+}
 var ARCHIVE_CHECK_COOLDOWN_MS = 60 * 60 * 1000;
 function createMemorySystemHook(ctx) {
   const sessionStates = new Map;
   const archiveState = { inProgress: false, lastCheck: 0 };
+  async function runArchivistSummary(prompt) {
+    try {
+      const createResult = await ctx.client.session.create({
+        body: {
+          title: "Memory: Deep Summary"
+        }
+      });
+      if (createResult.error)
+        return null;
+      const sessionID = createResult.data.id;
+      subagentSessions.add(sessionID);
+      await ctx.client.session.prompt({
+        path: { id: sessionID },
+        body: {
+          agent: "archivist",
+          parts: [{ type: "text", text: prompt }]
+        }
+      });
+      const messagesResult = await ctx.client.session.messages({
+        path: { id: sessionID }
+      });
+      const messages = messagesResult.data ?? messagesResult;
+      const assistantMessages = messages.filter((m) => m.info.role === "assistant");
+      const last = assistantMessages[assistantMessages.length - 1];
+      const text = last?.parts.filter((p) => p.type === "text" && p.text).map((p) => p.text).join(`
+`).trim();
+      return text || null;
+    } catch {
+      return null;
+    }
+  }
   function getOrCreateState(sessionID) {
     let state2 = sessionStates.get(sessionID);
     if (!state2) {
@@ -24611,7 +24861,34 @@ function createMemorySystemHook(ctx) {
       count: checkResult.archived.length
     });
     try {
-      const result = archiveOldMemories(ctx.directory);
+      const result = await archiveOldMemories(ctx.directory, {
+        deepSummarizer: async (session, fullContent) => {
+          const prompt = `Summarize the full transcript into long-term memory entries.
+
+Rules:
+- Output Markdown only.
+- Use sections only when relevant.
+- Keep each bullet concise (<120 chars).
+- Do NOT include sensitive data or raw conversation.
+- If nothing important, return "".
+
+Required sections (only if non-empty):
+**User Preferences:**
+- ...
+**Decisions Made:**
+- ...
+**Lessons Learned:**
+- ...
+
+Transcript:
+${fullContent}`;
+          try {
+            return await runArchivistSummary(prompt);
+          } catch {
+            return null;
+          }
+        }
+      });
       await ctx.client.tui.showToast({
         body: {
           title: "Memory Archived",
@@ -24647,17 +24924,11 @@ function createMemorySystemHook(ctx) {
         query: { directory: ctx.directory }
       });
       const messages = resp.data ?? resp;
-      if (messages.length < MIN_MESSAGES_TO_SAVE) {
-        log(`[${HOOK_NAME6}] Too few messages`, {
-          sessionID,
-          count: messages.length,
-          threshold: MIN_MESSAGES_TO_SAVE
-        });
-        return;
-      }
       const entry = extractSessionSummary(sessionID, messages);
+      const fullTranscript = extractFullTranscript(messages);
       const success = appendMemoryEntry(ctx.directory, entry);
-      if (success) {
+      const fullSuccess = saveFullTranscript(ctx.directory, sessionID, fullTranscript);
+      if (success && fullSuccess) {
         state2.saved = true;
         log(`[${HOOK_NAME6}] Memory saved`, {
           sessionID,
@@ -24666,7 +24937,11 @@ function createMemorySystemHook(ctx) {
         });
         await checkAndArchive();
       } else {
-        log(`[${HOOK_NAME6}] Failed to save memory`, { sessionID });
+        log(`[${HOOK_NAME6}] Failed to save memory`, {
+          sessionID,
+          summarySaved: success,
+          fullSaved: fullSuccess
+        });
       }
     } catch (err) {
       log(`[${HOOK_NAME6}] Error saving memory`, { sessionID, error: String(err) });
@@ -30031,7 +30306,7 @@ ${msg}`);
 // src/tools/lsp/utils.ts
 import { extname as extname2, resolve as resolve7 } from "path";
 import { fileURLToPath as fileURLToPath3 } from "url";
-import { existsSync as existsSync43, readFileSync as readFileSync30, writeFileSync as writeFileSync16 } from "fs";
+import { existsSync as existsSync43, readFileSync as readFileSync30, writeFileSync as writeFileSync17 } from "fs";
 function findWorkspaceRoot(filePath) {
   let dir = resolve7(filePath);
   if (!existsSync43(dir) || !__require("fs").statSync(dir).isDirectory()) {
@@ -30261,7 +30536,7 @@ function applyTextEditsToFile(filePath, edits) {
 `));
       }
     }
-    writeFileSync16(filePath, lines.join(`
+    writeFileSync17(filePath, lines.join(`
 `), "utf-8");
     return { success: true, editCount: edits.length };
   } catch (err) {
@@ -30292,7 +30567,7 @@ function applyWorkspaceEdit(edit) {
         if (change.kind === "create") {
           try {
             const filePath = uriToPath(change.uri);
-            writeFileSync16(filePath, "", "utf-8");
+            writeFileSync17(filePath, "", "utf-8");
             result.filesModified.push(filePath);
           } catch (err) {
             result.success = false;
@@ -30303,7 +30578,7 @@ function applyWorkspaceEdit(edit) {
             const oldPath = uriToPath(change.oldUri);
             const newPath = uriToPath(change.newUri);
             const content = readFileSync30(oldPath, "utf-8");
-            writeFileSync16(newPath, content, "utf-8");
+            writeFileSync17(newPath, content, "utf-8");
             __require("fs").unlinkSync(oldPath);
             result.filesModified.push(newPath);
           } catch (err) {
@@ -50700,8 +50975,10 @@ When analyzing problems:
 \u4F60\u6709\u4E00\u4E2A\u6587\u4EF6\u7CFB\u7EDF\u8BB0\u5FC6\uFF0C\u7528\u4E8E\u8DE8\u4F1A\u8BDD\u4FDD\u7559\u91CD\u8981\u4FE1\u606F\uFF1A
 
 **\u5B58\u50A8\u4F4D\u7F6E\uFF1A**
+- \`KNOWLEDGE.md\` \u2014 \u9879\u76EE\u77E5\u8BC6\u5E93\uFF08\u4ED3\u5E93\u7EA7\u3001\u7ED3\u6784\u5316\u4E8B\u5B9E\uFF09
 - \`.opencode/MEMORY.md\` \u2014 \u957F\u671F\u77E5\u8BC6\u6C89\u6DC0\uFF08\u6574\u7406\u540E\u7684\u7CBE\u534E\uFF09
 - \`.opencode/memory/\` \u2014 \u6309\u65E5\u671F\u5B58\u50A8\u7684\u5BF9\u8BDD\u8BB0\u5FC6
+- \`.opencode/memory/full/\` \u2014 \u6BCF\u4E2A session \u7684\u5B8C\u6574\u5BF9\u8BDD
 
 **\u4F55\u65F6\u67E5\u9605\u8BB0\u5FC6\uFF1A**
 - \u7528\u6237\u63D0\u5230"\u4E4B\u524D\u8BA8\u8BBA\u8FC7"\u3001"\u4E0A\u6B21"\u3001"\u6211\u4EEC\u51B3\u5B9A\u7684"
@@ -50709,9 +50986,10 @@ When analyzing problems:
 - \u7528\u6237\u95EE"\u4F60\u8FD8\u8BB0\u5F97...\u5417"
 
 **\u5982\u4F55\u4F7F\u7528\uFF1A**
-1. \`read(".opencode/MEMORY.md")\` \u2014 \u67E5\u770B\u957F\u671F\u8BB0\u5FC6
-2. \`glob(".opencode/memory/*.md")\` \u2014 \u5217\u51FA\u6240\u6709\u8BB0\u5FC6\u6587\u4EF6
-3. \`grep("\u5173\u952E\u8BCD", ".opencode/memory/")\` \u2014 \u641C\u7D22\u7279\u5B9A\u8BDD\u9898
+1. \`read("KNOWLEDGE.md")\` \u2014 \u67E5\u770B\u9879\u76EE\u77E5\u8BC6\u5E93
+2. \`read(".opencode/MEMORY.md")\` \u2014 \u67E5\u770B\u957F\u671F\u8BB0\u5FC6
+3. \`glob(".opencode/memory/*.md")\` \u2014 \u5217\u51FA\u6240\u6709\u8BB0\u5FC6\u6587\u4EF6
+4. \`grep("\u5173\u952E\u8BCD", ".opencode/memory/")\` \u2014 \u641C\u7D22\u7279\u5B9A\u8BDD\u9898
 
 **\u8BB0\u5FC6\u662F\u4F60\u7684\u8D44\u4EA7**\uFF1A\u5584\u7528\u5B83\u6765\u4FDD\u6301\u8FDE\u8D2F\u6027\uFF0C\u907F\u514D\u91CD\u590D\u8BA8\u8BBA\u5DF2\u51B3\u5B9A\u7684\u4E8B\u9879\u3002
 </Memory_System>
@@ -52221,6 +52499,8 @@ var MEMORY_CONSOLIDATE_TEMPLATE = `Consolidate daily memory logs into long-term
 ## WHAT TO DO
 
 Read all files in \`.opencode/memory/\` directory and consolidate important information into \`.opencode/MEMORY.md\`.
+If a daily log entry contains Decisions, TODOs, or critical tags (#project, #preference, #policy, #important),
+also pull the full transcript from \`.opencode/memory/full/<sessionID>.md\` for deeper summarization.
 
 ### Step 1: Read Daily Logs
 
@@ -52240,6 +52520,9 @@ From the daily logs, identify:
 4. **Recurring Problems** - Issues that came up multiple times and their solutions
 5. **Key Insights** - Valuable learnings worth preserving
 
+If deep-summary trigger conditions are met, read the full transcript for that session and extract additional
+preferences, decisions, and lessons.
+
 ### Step 3: Update MEMORY.md
 
 Read existing \`.opencode/MEMORY.md\` (create if doesn't exist).
@@ -52259,6 +52542,9 @@ Append new consolidated information using this structure:
 ### Lessons Learned
 - [insight or lesson]
 
+### Deep Summaries (when triggered)
+- [sessionID] preference/decision/lesson from full transcript
+
 ---
 \`\`\`
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "newtype-profile",
-  "version": "1.0.47",
+  "version": "1.0.49",
   "description": "AI Agent Collaboration System for Content Creation - Based on oh-my-opencode",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",