newtype-profile 1.0.61 → 1.0.63

package/README.md

@@ -28,6 +28,36 @@ newtype-profile is an AI Agent collaboration framework designed for **content cr
 - 🔍 Information research and fact-checking
 - 📄 Document extraction and organization
 
+## Two Ways to Use
+
+### Option A: Newtype CLI (Recommended)
+
+[**Newtype CLI**](https://www.npmjs.com/package/newtype-cli) is a standalone terminal AI assistant with newtype-profile built-in. No plugin setup needed — install and go.
+
+```bash
+npm install -g newtype-cli
+newtype
+```
+
+Newtype CLI is a white-labeled fork of [OpenCode](https://github.com/anomalyco/opencode) with the full agent team pre-configured.
+
+### Option B: As an OpenCode Plugin
+
+If you already use OpenCode and want to add newtype-profile as a plugin:
+
+```bash
+cd ~/.config/opencode
+bun add newtype-profile
+```
+
+Edit `~/.config/opencode/opencode.json`:
+
+```json
+{
+  "plugin": ["newtype-profile"]
+}
+```
+
 ## Agent Team
 
 | Agent | Role | Description |
@@ -41,40 +71,41 @@ newtype-profile is an AI Agent collaboration framework designed for **content cr
 | **writer** | Writer | Content production |
 | **editor** | Editor | Content refinement |
 
-## Quick Start
+## Configure Models
 
-### Installation
+Create a config file to customize which models each agent uses:
 
-```bash
-cd ~/.config/opencode
-bun add newtype-profile
-```
-
-Edit `~/.config/opencode/opencode.json`:
+- **Newtype CLI**: `~/.config/newtype/newtype-profile.json`
+- **OpenCode plugin**: `~/.config/opencode/newtype-profile.json`
 
 ```json
 {
-  "plugin": ["newtype-profile"]
+  "agents": {
+    "chief": { "model": "your-preferred-model" },
+    "deputy": { "model": "your-preferred-model" },
+    "researcher": { "model": "your-preferred-model" },
+    "writer": { "model": "your-preferred-model", "temperature": 0.7 }
+  }
 }
 ```
 
-### Configure Models
+All 8 agents (`chief`, `deputy`, `researcher`, `fact-checker`, `archivist`, `extractor`, `writer`, `editor`) can be configured independently.
 
-Create `~/.config/opencode/newtype-profile.json`:
+<details>
+<summary>Optional: Google Antigravity OAuth</summary>
+
+If using Google Antigravity as your model provider, add `google_auth`:
 
 ```json
 {
   "google_auth": true,
   "agents": {
-    "chief": { "model": "google/antigravity-claude-opus-4-5-thinking-high" },
-    "researcher": { "model": "google/antigravity-gemini-3-pro-high" },
-    "writer": { "model": "google/antigravity-gemini-3-pro-high" },
-    "editor": { "model": "google/antigravity-claude-sonnet-4-5" }
+    "chief": { "model": "google/antigravity-claude-opus-4-5-thinking-high" }
   }
 }
 ```
 
-### Authenticate
+Then authenticate:
 
 ```bash
 opencode auth login
@@ -82,6 +113,8 @@ opencode auth login
 # Select Login method: OAuth with Google (Antigravity)
 ```
 
+</details>
+
 ## Usage
 
 ### Three-Layer Architecture
@@ -116,7 +149,7 @@ Chief's personality has three layers:
 - **Inner Persona** (hardcoded): Core values and thinking patterns
 - **Outer Persona** (customizable): Communication style
 
-Create `.opencode/SOUL.md` to customize how Chief communicates:
+Create `.opencode/SOUL.md` (or `.newtype/SOUL.md` for Newtype CLI) to customize how Chief communicates:
 
 ```bash
 /init-soul  # Creates default SOUL.md template
@@ -127,7 +160,7 @@ Example customizations:
 - Adjust language preferences
 - Change how direct Chief is
 
-Changes take effect after restarting OpenCode.
+Changes take effect after restarting.
 
 ### Built-in Skills
 
@@ -166,7 +199,7 @@ Built-in MCPs:
 
 ## Memory System
 
-Auto-saves conversation summaries to `.opencode/memory/`:
+Auto-saves conversation summaries to `.opencode/memory/` (or `.newtype/memory/`):
 - Daily summaries (LLM-generated)
 - Full transcripts per session
 - Auto-archive to `MEMORY.md` after 7 days
@@ -180,6 +213,21 @@ Use `/memory-consolidate` to manually trigger consolidation.
 - **Startup Config Checker**: Guides model setup on first run
 - **Plugin Switching**: `/switch newtype` / `/switch omo` / `/switch none`
 
+## Newtype CLI
+
+[Newtype CLI](https://www.npmjs.com/package/newtype-cli) is a standalone product that bundles newtype-profile into a ready-to-use terminal AI assistant.
+
+| | newtype-profile (plugin) | Newtype CLI |
+|---|---|---|
+| **Install** | `bun add newtype-profile` in OpenCode | `npm install -g newtype-cli` |
+| **Requires** | OpenCode installed separately | Nothing — self-contained |
+| **Launch** | `opencode` | `newtype` |
+| **Config dir** | `~/.config/opencode/` | `~/.config/newtype/` |
+| **Project dir** | `.opencode/` | `.newtype/` |
+| **npm package** | [newtype-profile](https://www.npmjs.com/package/newtype-profile) | [newtype-cli](https://www.npmjs.com/package/newtype-cli) |
+
+Supported platforms: macOS (Apple Silicon & Intel), Linux (x64 & ARM64, glibc & musl), Windows (x64).
+
 ## License
 
 Based on [oh-my-opencode](https://github.com/code-yeongyu/oh-my-opencode), follows [SUL-1.0 License](https://github.com/code-yeongyu/oh-my-opencode/blob/master/LICENSE.md).

package/README.zh-cn.md

@@ -28,6 +28,36 @@ newtype-profile 是专为**内容创作**设计的 AI Agent 协作框架：
 - 🔍 信息调研与核查
 - 📄 文档提取与整理
 
+## 两种使用方式
+
+### 方式 A：Newtype CLI（推荐）
+
+[**Newtype CLI**](https://www.npmjs.com/package/newtype-cli) 是独立的终端 AI 助手，已内置 newtype-profile。无需额外配置插件，安装即用。
+
+```bash
+npm install -g newtype-cli
+newtype
+```
+
+Newtype CLI 基于 [OpenCode](https://github.com/anomalyco/opencode) 定制，预配置了完整的 Agent 团队。
+
+### 方式 B：作为 OpenCode 插件
+
+如果你已在使用 OpenCode，可以将 newtype-profile 作为插件添加：
+
+```bash
+cd ~/.config/opencode
+bun add newtype-profile
+```
+
+编辑 `~/.config/opencode/opencode.json`：
+
+```json
+{
+  "plugin": ["newtype-profile"]
+}
+```
+
 ## Agent 团队
 
 | Agent | 角色 | 职责描述 |
@@ -41,40 +71,41 @@ newtype-profile 是专为**内容创作**设计的 AI Agent 协作框架：
 | **writer** | 写手 | 内容生产 |
 | **editor** | 编辑 | 内容精炼 |
 
-## 快速开始
+## 配置模型
 
-### 安装
+创建配置文件来自定义各 Agent 使用的模型：
 
-```bash
-cd ~/.config/opencode
-bun add newtype-profile
-```
-
-编辑 `~/.config/opencode/opencode.json`：
+- **Newtype CLI**：`~/.config/newtype/newtype-profile.json`
+- **OpenCode 插件**：`~/.config/opencode/newtype-profile.json`
 
 ```json
 {
-  "plugin": ["newtype-profile"]
+  "agents": {
+    "chief": { "model": "你偏好的模型" },
+    "deputy": { "model": "你偏好的模型" },
+    "researcher": { "model": "你偏好的模型" },
+    "writer": { "model": "你偏好的模型", "temperature": 0.7 }
+  }
 }
 ```
 
-### 配置模型
+全部 8 个 Agent（`chief`、`deputy`、`researcher`、`fact-checker`、`archivist`、`extractor`、`writer`、`editor`）均可独立配置。
 
-创建 `~/.config/opencode/newtype-profile.json`：
+<details>
+<summary>可选：Google Antigravity OAuth</summary>
+
+如果使用 Google Antigravity 作为模型提供商，添加 `google_auth`：
 
 ```json
 {
   "google_auth": true,
   "agents": {
-    "chief": { "model": "google/antigravity-claude-opus-4-5-thinking-high" },
-    "researcher": { "model": "google/antigravity-gemini-3-pro-high" },
-    "writer": { "model": "google/antigravity-gemini-3-pro-high" },
-    "editor": { "model": "google/antigravity-claude-sonnet-4-5" }
+    "chief": { "model": "google/antigravity-claude-opus-4-5-thinking-high" }
   }
 }
 ```
 
-### 认证
+然后认证：
 
 ```bash
 opencode auth login
@@ -82,6 +113,8 @@ opencode auth login
 # 选择 Login method: OAuth with Google (Antigravity)
 ```
 
+</details>
+
 ## 使用方式
 
 ### 三层架构
@@ -116,7 +149,7 @@ Chief 的人格分三层：
 - **里人格**（硬编码）：核心价值观和思维方式
 - **表人格**（可自定义）：沟通风格
 
-创建 `.opencode/SOUL.md` 来自定义 Chief 的沟通风格：
+创建 `.opencode/SOUL.md`（Newtype CLI 为 `.newtype/SOUL.md`）来自定义 Chief 的沟通风格：
 
 ```bash
 /init-soul  # 创建默认 SOUL.md 模板
@@ -127,7 +160,7 @@ Chief 的人格分三层：
 - 调整语言偏好
 - 改变直接程度
 
-修改后重启 OpenCode 生效。
+修改后重启生效。
 
 ### 内置 Skills
 
@@ -166,7 +199,7 @@ Chief 在任务需要时自动加载技能。
 
 ## 记忆系统
 
-自动保存对话摘要到 `.opencode/memory/`：
+自动保存对话摘要到 `.opencode/memory/`（Newtype CLI 为 `.newtype/memory/`）：
 - 每日摘要（LLM 生成）
 - 每个 session 的完整记录
 - 7 天后自动归档到 `MEMORY.md`
@@ -180,6 +213,21 @@ Chief 在任务需要时自动加载技能。
 - **启动配置检查**：首次运行时引导模型设置
 - **插件切换**：`/switch newtype` / `/switch omo` / `/switch none`
 
+## Newtype CLI
+
+[Newtype CLI](https://www.npmjs.com/package/newtype-cli) 是独立产品，将 newtype-profile 打包为开箱即用的终端 AI 助手。
+
+| | newtype-profile（插件） | Newtype CLI |
+|---|---|---|
+| **安装** | 在 OpenCode 中 `bun add newtype-profile` | `npm install -g newtype-cli` |
+| **依赖** | 需要单独安装 OpenCode | 无需其他依赖，自包含 |
+| **启动** | `opencode` | `newtype` |
+| **配置目录** | `~/.config/opencode/` | `~/.config/newtype/` |
+| **项目目录** | `.opencode/` | `.newtype/` |
+| **npm 包** | [newtype-profile](https://www.npmjs.com/package/newtype-profile) | [newtype-cli](https://www.npmjs.com/package/newtype-cli) |
+
+支持平台：macOS（Apple Silicon 和 Intel）、Linux（x64 和 ARM64，glibc 和 musl）、Windows（x64）。
+
 ## 许可证
 
 基于 [oh-my-opencode](https://github.com/code-yeongyu/oh-my-opencode) 修改，遵循 [SUL-1.0 许可证](https://github.com/code-yeongyu/oh-my-opencode/blob/master/LICENSE.md)。

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: "1.0.61",
+    version: "1.0.63",
     description: "AI Agent Collaboration System for Content Creation - Based on oh-my-opencode",
     main: "dist/index.js",
     types: "dist/index.d.ts",

package/dist/hooks/chief-orchestrator/confidence-router.d.ts

@@ -66,3 +66,13 @@ export declare function detectAgentType(output: string, category?: string): Agen
  * Check if output contains a confidence score
  */
 export declare function hasConfidenceScore(output: string): boolean;
+export interface FailureJournalEntry {
+    timestamp: string;
+    agentType: AgentType;
+    confidence: number;
+    attempts: number;
+    outputExcerpt: string;
+    sessionId: string;
+}
+export declare function buildFailureJournalEntry(agentType: AgentType, confidence: number, attempts: number, output: string, sessionId: string): FailureJournalEntry;
+export declare function formatFailureJournalMarkdown(entry: FailureJournalEntry): string;

package/dist/hooks/chief-orchestrator/quality-dimensions.d.ts

@@ -57,3 +57,8 @@ export declare function buildImprovementDirective(assessment: QualityAssessment,
  * Check if output has quality scores (either format)
  */
 export declare function hasQualityScores(output: string): boolean;
+/**
+ * Build a warning directive when an agent's output lacks quality scores.
+ * Tells Chief what dimensions to expect and how to request them.
+ */
+export declare function buildMissingScoresWarning(agentType: AgentType, sessionId: string): string;

package/dist/hooks/chief-orchestrator/shared-context.d.ts

@@ -53,3 +53,4 @@ export declare function getArtifactDetails(sessionID: string, artifactIds: strin
 export declare function getAllSources(sessionID: string): Source[];
 export declare function getAllFindings(sessionID: string): Finding[];
 export declare function hasArtifacts(output: string): boolean;
+export declare function buildDetailedContextSummary(sessionID: string, downstreamAgent?: string): string | null;

package/dist/index.js

@@ -7852,7 +7852,7 @@ var require_formats = __commonJS((exports) => {
   }
   var TIME = /^(\d\d):(\d\d):(\d\d(?:\.\d+)?)(z|([+-])(\d\d)(?::?(\d\d))?)?$/i;
   function getTime(strictTimeZone) {
-    return function time(str2) {
+    return function time3(str2) {
       const matches = TIME.exec(str2);
       if (!matches)
         return false;
@@ -22427,7 +22427,7 @@ to resume: chief_task(resume="${sessionId}", prompt="...")`;
 }
 // src/hooks/chief-orchestrator/index.ts
 import { execSync as execSync2 } from "child_process";
-import { existsSync as existsSync39, readdirSync as readdirSync14 } from "fs";
+import { existsSync as existsSync39, readdirSync as readdirSync14, appendFileSync as appendFileSync6, mkdirSync as mkdirSync12 } from "fs";
 import { join as join47 } from "path";
 // src/features/boulder-state/constants.ts
 var BOULDER_DIR = ".chief";
@@ -22974,6 +22974,36 @@ function detectAgentType(output, category) {
 function hasConfidenceScore(output) {
   return extractConfidence(output) !== null;
 }
+function buildFailureJournalEntry(agentType, confidence, attempts, output, sessionId) {
+  const excerpt = output.slice(0, 1000).replace(/\n{3,}/g, `
+
+`);
+  return {
+    timestamp: new Date().toISOString(),
+    agentType,
+    confidence,
+    attempts,
+    outputExcerpt: excerpt,
+    sessionId
+  };
+}
+function formatFailureJournalMarkdown(entry) {
+  const confPct = Math.round(entry.confidence * 100);
+  return `## Escalation: ${entry.agentType} (${entry.timestamp.split("T")[0]})
+
+- **Agent**: ${entry.agentType}
+- **Confidence**: ${confPct}%
+- **Rewrite attempts**: ${entry.attempts}
+- **Session**: ${entry.sessionId}
+
+### Last Output Excerpt
+\`\`\`
+${entry.outputExcerpt}
+\`\`\`
+
+---
+`;
+}
 
 // src/hooks/chief-orchestrator/quality-dimensions.ts
 var AGENT_DIMENSIONS = {
@@ -23349,6 +23379,40 @@ function getImprovementCategory(agentType, weakDimension) {
 function hasQualityScores(output) {
   return /\*\*QUALITY SCORES:\*\*/i.test(output) || /\*\*CONFIDENCE:\s*[\d.]+\*\*/i.test(output);
 }
+function buildMissingScoresWarning(agentType, sessionId) {
+  const dimensions = AGENT_DIMENSIONS[agentType];
+  const dimList = dimensions.map((d) => d.label).join(", ");
+  const agentLabels = {
+    "fact-checker": "FACT-CHECK",
+    researcher: "RESEARCH",
+    writer: "DRAFT",
+    editor: "EDIT",
+    archivist: "ARCHIVE",
+    extractor: "EXTRACTION"
+  };
+  const label = agentLabels[agentType];
+  return `
+
+---
+[${label}: QUALITY SCORES MISSING]
+
+\u26A0\uFE0F The ${agentType} agent did not include quality self-assessment scores.
+Without scores, the quality feedback loop cannot function.
+
+**Expected dimensions for ${agentType}:** ${dimList}
+
+**RECOMMENDED ACTION:**
+Request scores by resuming the session:
+\`\`\`
+chief_task(
+  resume="${sessionId}",
+  prompt="Please provide your quality self-assessment. Output in this format:\\n**QUALITY SCORES:**\\n${dimensions.map((d) => `- ${d.label}: <0.00-1.00>`).join("\\n")}\\n**OVERALL: <0.00-1.00>**"
+)
+\`\`\`
+
+If the output quality is obviously high, you may skip this and proceed.
+---`;
+}
 
 // src/hooks/chief-orchestrator/shared-context.ts
 var contextPool = new Map;
@@ -23505,11 +23569,153 @@ function formatAgentName(agentType) {
 function hasArtifacts(output) {
   return /\*\*ARTIFACTS:\*\*/i.test(output);
 }
+var DETAIL_HUNGRY_AGENTS = ["writer", "editor", "fact-checker"];
+function buildDetailedContextSummary(sessionID, downstreamAgent) {
+  const ctx = getContext(sessionID);
+  if (!ctx || ctx.artifacts.length === 0) {
+    return null;
+  }
+  const isDetailHungry = downstreamAgent ? DETAIL_HUNGRY_AGENTS.includes(downstreamAgent) : false;
+  if (!isDetailHungry) {
+    return buildContextSummary(sessionID);
+  }
+  const lines = [
+    "<shared-context>",
+    "## Previous Work by Other Agents (DETAILED)",
+    ""
+  ];
+  const byAgent = new Map;
+  for (const a of ctx.artifacts) {
+    const list = byAgent.get(a.agentType) ?? [];
+    list.push(a);
+    byAgent.set(a.agentType, list);
+  }
+  for (const [agentType, artifacts] of byAgent) {
+    lines.push(`### ${formatAgentName(agentType)} (${artifacts.length} task${artifacts.length > 1 ? "s" : ""})`);
+    for (const a of artifacts) {
+      lines.push(`- **[${a.id}]** ${a.taskDescription}`);
+      if (a.sources && a.sources.length > 0) {
+        lines.push(`  - **Sources (${a.sources.length} total):**`);
+        for (const s of a.sources) {
+          const credIcon = s.credibility === "high" ? "\u2713" : s.credibility === "low" ? "\u26A0" : "\u25CB";
+          const urlPart = s.url ? ` \u2014 ${s.url}` : "";
+          const excerptPart = s.excerpt ? `
+      > ${s.excerpt.slice(0, 200)}${s.excerpt.length > 200 ? "..." : ""}` : "";
+          lines.push(`    - ${credIcon} **${s.title}** (${s.type}, ${s.credibility ?? "unknown"} credibility)${urlPart}${excerptPart}`);
+        }
+      }
+      if (a.findings && a.findings.length > 0) {
+        lines.push(`  - **Findings (${a.findings.length} total):**`);
+        for (const f of a.findings) {
+          const confPct = Math.round(f.confidence * 100);
+          const refs = f.sourceRefs.length > 0 ? ` [refs: ${f.sourceRefs.join(", ")}]` : "";
+          const notes = f.notes ? ` \u2014 ${f.notes}` : "";
+          lines.push(`    - ${f.claim} (${confPct}% confident)${refs}${notes}`);
+        }
+      }
+      if (a.issues && a.issues.length > 0) {
+        lines.push(`  - **Issues (${a.issues.length} total):**`);
+        for (const i2 of a.issues) {
+          const sev = i2.severity === "critical" ? "\uD83D\uDD34" : i2.severity === "major" ? "\uD83D\uDFE1" : "\u26AA";
+          const suggestion = i2.suggestion ? ` \u2192 ${i2.suggestion}` : "";
+          lines.push(`    - ${sev} [${i2.type}] ${i2.description}${suggestion}`);
+        }
+      }
+      if (a.connections && a.connections.length > 0) {
+        lines.push(`  - **Connections:** ${a.connections.join(", ")}`);
+      }
+      if (a.content) {
+        const preview = a.content.slice(0, 500).replace(/\n/g, " ");
+        lines.push(`  - **Content preview:** "${preview}${a.content.length > 500 ? "..." : ""}"`);
+      }
+    }
+    lines.push("");
+  }
+  lines.push("## How to Use This Context");
+  lines.push("- Reference artifacts by ID (e.g., [researcher_001]) in your output");
+  lines.push("- Build upon findings rather than re-researching");
+  lines.push("- Cite sources when making claims based on research");
+  lines.push("- Flag any inconsistencies you find with previous work");
+  lines.push("</shared-context>");
+  return lines.join(`
+`);
+}
 
 // src/hooks/chief-orchestrator/index.ts
 var HOOK_NAME5 = "chief-orchestrator";
 var ALLOWED_PATH_PREFIX2 = ".chief/";
 var WRITE_EDIT_TOOLS = ["Write", "Edit", "write", "edit"];
+var DEPUTY_SHOULD_DELEGATE_TOOLS = {
+  web_search_exa: "researcher",
+  websearch_web_search_exa: "researcher",
+  tavily_search: "researcher",
+  tavily_extract: "researcher",
+  firecrawl_scrape: "researcher",
+  firecrawl_search: "researcher",
+  firecrawl_crawl: "researcher",
+  webfetch: "researcher",
+  context7_resolve_library_id: "researcher",
+  "context7_resolve-library-id": "researcher",
+  context7_query_docs: "researcher",
+  "context7_query-docs": "researcher",
+  grep_app_searchGitHub: "researcher"
+};
+function shouldDelegateToSpecialist(tool) {
+  return tool in DEPUTY_SHOULD_DELEGATE_TOOLS;
+}
+function getRecommendedSpecialist(tool) {
+  return DEPUTY_SHOULD_DELEGATE_TOOLS[tool] ?? "researcher";
+}
+function buildDeputyDelegationReminder(tool, specialist) {
+  return `
+
+---
+
+[SYSTEM REMINDER - DELEGATION RECOMMENDED]
+
+You (Deputy) are directly calling \`${tool}\`, which is a search/research tool.
+
+**Best practice**: Delegate search tasks to the **${specialist}** agent via:
+\`\`\`
+chief_task(subagent_type="${specialist}", prompt="[search task]")
+\`\`\`
+
+**Why**: The ${specialist} agent has specialized prompts for source evaluation, 
+quality scoring, and structured artifact output that you lack.
+
+**Exception**: If this is a quick one-off lookup (< 1 query), proceed.
+For systematic research (multiple queries, source comparison), delegate.
+
+---
+`;
+}
+var FAILURE_JOURNAL_DIR = ".opencode/memory";
+var FAILURE_JOURNAL_FILE = "escalation-log.md";
+function writeFailureJournal(directory, entry) {
+  try {
+    const memDir = join47(directory, FAILURE_JOURNAL_DIR);
+    if (!existsSync39(memDir)) {
+      mkdirSync12(memDir, { recursive: true });
+    }
+    const filePath = join47(memDir, FAILURE_JOURNAL_FILE);
+    const needsHeader = !existsSync39(filePath);
+    if (needsHeader) {
+      appendFileSync6(filePath, `# Escalation Log
+
+Automatic failure journal for quality escalations.
+
+`);
+    }
+    appendFileSync6(filePath, formatFailureJournalMarkdown(entry));
+    log(`[${HOOK_NAME5}] Failure journal entry written`, {
+      agentType: entry.agentType,
+      confidence: entry.confidence,
+      file: filePath
+    });
+  } catch (err) {
+    log(`[${HOOK_NAME5}] Failed to write failure journal`, { error: String(err) });
+  }
+}
 var DIRECT_WORK_REMINDER = `
 
 ---
@@ -23980,18 +24186,30 @@ function createChiefOrchestratorHook(ctx, options) {
         }
         return;
       }
+      if (callerIsDeputy && shouldDelegateToSpecialist(input.tool)) {
+        const specialist = getRecommendedSpecialist(input.tool);
+        output.message = (output.message || "") + buildDeputyDelegationReminder(input.tool, specialist);
+        log(`[${HOOK_NAME5}] Deputy delegation reminder`, {
+          sessionID: input.sessionID,
+          tool: input.tool,
+          recommendedSpecialist: specialist
+        });
+      }
       if (input.tool === "chief_task") {
         const prompt = output.args.prompt;
         if (prompt && !prompt.includes("[SYSTEM DIRECTIVE - SINGLE TASK ONLY]")) {
           let enhancedPrompt = prompt;
           if (input.sessionID) {
-            const sharedContext = buildContextSummary(input.sessionID);
+            const downstreamAgent = output.args.subagent_type;
+            const sharedContext = buildDetailedContextSummary(input.sessionID, downstreamAgent) ?? buildContextSummary(input.sessionID);
             if (sharedContext) {
               enhancedPrompt = `${sharedContext}
 
 ${enhancedPrompt}`;
               log(`[${HOOK_NAME5}] Injected shared context to chief_task`, {
-                sessionID: input.sessionID
+                sessionID: input.sessionID,
+                downstreamAgent,
+                detailed: sharedContext.includes("DETAILED")
               });
             }
           }
@@ -24132,6 +24350,10 @@ ${buildImprovementDirective(qualityAssessment, subagentSessionId)}
                 weakest: qualityAssessment.weakest?.name,
                 allPass: qualityAssessment.allPass
               });
+              if (qualityAssessment.overall < 0.5) {
+                const journalEntry = buildFailureJournalEntry(agentType, qualityAssessment.overall, 1, output.output, subagentSessionId);
+                writeFailureJournal(ctx.directory, journalEntry);
+              }
             }
           } else if (hasConfidenceScore(output.output) && agentType) {
             const confidenceResult = analyzeAgentOutput(output.output, subagentSessionId, agentType);
@@ -24148,6 +24370,16 @@ ${confidenceResult.directive}
                 recommendation: confidenceResult.recommendation
               });
             }
+            if (confidenceResult.recommendation === "escalate" && confidenceResult.confidence !== null) {
+              const journalEntry = buildFailureJournalEntry(agentType, confidenceResult.confidence, 3, output.output, subagentSessionId);
+              writeFailureJournal(ctx.directory, journalEntry);
+            }
+          } else if (agentType && !hasQualityScores(output.output) && !hasConfidenceScore(output.output)) {
+            confidenceDirective = buildMissingScoresWarning(agentType, subagentSessionId);
+            log(`[${HOOK_NAME5}] Quality scores missing from agent output`, {
+              sessionID: input.sessionID,
+              agentType
+            });
           }
           output.output = `${formattedSummary}
 
@@ -24183,8 +24415,8 @@ var DEEP_SUMMARY_TAGS = ["#project", "#preference", "#policy", "#important"];
 // src/hooks/memory-system/storage.ts
 import {
   existsSync as existsSync40,
-  mkdirSync as mkdirSync12,
-  appendFileSync as appendFileSync6,
+  mkdirSync as mkdirSync13,
+  appendFileSync as appendFileSync7,
   readFileSync as readFileSync26,
   readdirSync as readdirSync15,
   unlinkSync as unlinkSync10,
@@ -24194,14 +24426,14 @@ import { join as join48 } from "path";
 function ensureMemoryDir(projectDir) {
   const memoryPath = join48(projectDir, MEMORY_DIR);
   if (!existsSync40(memoryPath)) {
-    mkdirSync12(memoryPath, { recursive: true });
+    mkdirSync13(memoryPath, { recursive: true });
   }
   return memoryPath;
 }
 function ensureFullMemoryDir(projectDir) {
   const memoryPath = join48(projectDir, FULL_MEMORY_DIR);
   if (!existsSync40(memoryPath)) {
-    mkdirSync12(memoryPath, { recursive: true });
+    mkdirSync13(memoryPath, { recursive: true });
   }
   return memoryPath;
 }
@@ -24343,9 +24575,9 @@ async function archiveOldMemories(projectDir, options) {
 
 This file contains archived conversation memories.
 `;
-    appendFileSync6(memoryFilePath, fileHeader);
+    appendFileSync7(memoryFilePath, fileHeader);
   }
-  appendFileSync6(memoryFilePath, archivedContent.join(""));
+  appendFileSync7(memoryFilePath, archivedContent.join(""));
   return {
     archived: checkResult.archived,
     totalFiles: checkResult.totalFiles - checkResult.archived.length,
@@ -24463,9 +24695,9 @@ function appendMemoryEntry(projectDir, entry) {
       const header = `# Memory Log - ${new Date().toISOString().split("T")[0]}
 
 `;
-      appendFileSync6(filePath, header);
+      appendFileSync7(filePath, header);
     }
-    appendFileSync6(filePath, content);
+    appendFileSync7(filePath, content);
     return true;
   } catch {
     return false;
@@ -43225,7 +43457,7 @@ import { dirname as dirname11, join as join53 } from "path";
 import { existsSync as existsSync45, statSync as statSync4 } from "fs";
 
 // src/tools/ast-grep/downloader.ts
-import { existsSync as existsSync44, mkdirSync as mkdirSync13, chmodSync as chmodSync2, unlinkSync as unlinkSync11 } from "fs";
+import { existsSync as existsSync44, mkdirSync as mkdirSync14, chmodSync as chmodSync2, unlinkSync as unlinkSync11 } from "fs";
 import { join as join52 } from "path";
 import { homedir as homedir16 } from "os";
 import { createRequire as createRequire3 } from "module";
@@ -43299,7 +43531,7 @@ async function downloadAstGrep(version2 = DEFAULT_VERSION) {
   console.log(`[oh-my-opencode] Downloading ast-grep binary...`);
   try {
     if (!existsSync44(cacheDir)) {
-      mkdirSync13(cacheDir, { recursive: true });
+      mkdirSync14(cacheDir, { recursive: true });
     }
     const response2 = await fetch(downloadUrl, { redirect: "follow" });
     if (!response2.ok) {
@@ -43740,7 +43972,7 @@ import { join as join55, dirname as dirname12 } from "path";
 import { spawnSync } from "child_process";
 
 // src/tools/grep/downloader.ts
-import { existsSync as existsSync47, mkdirSync as mkdirSync14, chmodSync as chmodSync3, unlinkSync as unlinkSync12, readdirSync as readdirSync16 } from "fs";
+import { existsSync as existsSync47, mkdirSync as mkdirSync15, chmodSync as chmodSync3, unlinkSync as unlinkSync12, readdirSync as readdirSync16 } from "fs";
 import { join as join54 } from "path";
 function findFileRecursive(dir, filename) {
   try {
@@ -43841,7 +44073,7 @@ async function downloadAndInstallRipgrep() {
   if (existsSync47(rgPath)) {
     return rgPath;
   }
-  mkdirSync14(installDir, { recursive: true });
+  mkdirSync15(installDir, { recursive: true });
   const filename = `ripgrep-${RG_VERSION}-${config3.platform}.${config3.extension}`;
   const url2 = `https://github.com/BurntSushi/ripgrep/releases/download/${RG_VERSION}/${filename}`;
   const archivePath = join54(installDir, filename);
@@ -50756,6 +50988,13 @@ You switch modes based on user intent.
 - Questions without clear deliverable
 - Exploratory, open-ended requests
 
+**\u26A0\uFE0F MANDATORY SKILL CHECK on Discussion Mode entry:**
+When entering discussion mode, BEFORE responding, check if any skill should be loaded:
+- \u7528\u6237\u8BF4"\u5206\u6790/\u8BC4\u4F30/\u5BF9\u6BD4/\u8C03\u7814" \u2192 \`skill({ name: "super-analyst" })\`
+- \u7528\u6237\u8BF4"\u5E2E\u6211\u7406\u601D\u8DEF/\u60F3\u6CD5/\u63A2\u7D22" \u2192 \`skill({ name: "super-interviewer" })\`
+- \u7528\u6237\u8BF4"\u5199/\u521B\u4F5C" \u2192 \u5207\u6362\u5230 Execution Mode
+Do NOT skip this check. Skills provide structured frameworks that dramatically improve output quality.
+
 ## Execution Mode Signals
 - "\u5E2E\u6211\u5199\u4E00\u7BC7..." / "Write me a..."
 - "\u6574\u7406\u6210..." / "Compile into..."
@@ -50864,14 +51103,43 @@ chief_task(
 </Execution_Behavior>
 
 <Discussion_Behavior>
-## Silent Delegation (via Deputy)
-When you notice information needs while discussing:
-- Factual claim needs verification \u2192 delegate to Deputy (who dispatches fact-checker)
-- Need external research \u2192 delegate to Deputy (who dispatches researcher)
-- Need existing materials \u2192 delegate to Deputy (who dispatches archivist)
+## Parallel Probes\uFF08\u591A\u7EF4\u5EA6\u5E76\u884C\u63A2\u6D4B\uFF09
+\u5F53\u4F60\u5728 Discussion Mode \u4E2D\u9047\u5230\u503C\u5F97\u6DF1\u5165\u7684\u8BDD\u9898\uFF0C**\u540C\u65F6**\u6D3E\u51FA\u591A\u4E2A\u540E\u53F0\u4EFB\u52A1\u83B7\u53D6\u4E0D\u540C\u7EF4\u5EA6\u7684\u4FE1\u606F\uFF1A
+
+### \u89E6\u53D1\u6761\u4EF6
+- \u8BDD\u9898\u6D89\u53CA**\u4E8B\u5B9E\u5224\u65AD + \u89C2\u70B9\u5206\u6B67**\uFF08\u5982"AI \u4F1A\u53D6\u4EE3 X \u5417"\uFF09
+- \u591A\u65B9\u5229\u76CA\u76F8\u5173\u7684\u590D\u6742\u8BDD\u9898\uFF08\u5982"\u516C\u53F8\u8BE5\u4E0D\u8BE5\u505A Y"\uFF09
+- \u7528\u6237\u7684\u8BBA\u70B9\u5EFA\u7ACB\u5728**\u672A\u7ECF\u9A8C\u8BC1\u7684\u5047\u8BBE**\u4E4B\u4E0A
 
-Use \`chief_task(subagent_type="deputy", run_in_background=true, ...)\` for async work.
-Weave results into conversation naturally. Don't announce "checking with my team."
+### \u5E76\u884C\u63A2\u6D4B\u6A21\u5F0F
+\u540C\u65F6\u53D1\u8D77 2-3 \u4E2A\u540E\u53F0\u4EFB\u52A1\uFF1A
+\`\`\`
+# \u4FE1\u606F\u6536\u96C6\uFF08researcher\uFF09
+chief_task(subagent_type="deputy", run_in_background=true,
+  prompt="\u8C03\u7814 [\u8BDD\u9898] \u7684\u6700\u65B0\u6570\u636E\u548C\u5173\u952E\u4E8B\u5B9E")
+
+# \u5047\u8BBE\u9A8C\u8BC1\uFF08fact-checker\uFF09
+chief_task(subagent_type="deputy", run_in_background=true,
+  prompt="\u9A8C\u8BC1\u4EE5\u4E0B\u5047\u8BBE\u7684\u53EF\u9760\u6027\uFF1A[\u7528\u6237\u7684\u5173\u952E\u5047\u8BBE]")
+
+# \u53CD\u9762\u8BBA\u70B9\uFF08researcher\uFF0Cdevil's advocate \u89D2\u5EA6\uFF09
+chief_task(subagent_type="deputy", run_in_background=true,
+  prompt="\u641C\u96C6 [\u8BDD\u9898] \u7684\u53CD\u9762\u8BC1\u636E\u548C\u4E3B\u8981\u6279\u8BC4\u89C2\u70B9")
+\`\`\`
+
+### \u5173\u952E\u539F\u5219
+- **\u4E0D\u7B49\u7ED3\u679C\u5C31\u56DE\u590D** \u2014 \u5148\u57FA\u4E8E\u4F60\u81EA\u5DF1\u7684\u5224\u65AD\u56DE\u590D\u7528\u6237\uFF0C\u540E\u53F0\u7ED3\u679C\u5230\u4E86\u518D\u81EA\u7136\u878D\u5165\u540E\u7EED\u5BF9\u8BDD
+- **\u4E0D\u544A\u8BC9\u7528\u6237** \u2014 \u8FD9\u662F\u4F60\u7684\u601D\u8003\u8FC7\u7A0B\uFF0C\u4E0D\u662F\u5C55\u793A\u6D41\u7A0B
+- **\u53EA\u7528\u4E8E\u503C\u5F97\u6DF1\u5165\u7684\u8BDD\u9898** \u2014 \u7B80\u5355\u95EE\u9898\u4E0D\u9700\u8981\u63A2\u6D4B\uFF0C\u522B\u6D6A\u8D39\u8D44\u6E90
+- **\u53CD\u9762\u8BBA\u70B9\u662F\u4E3A\u4E86\u5B8C\u6574\u6027\uFF0C\u4E0D\u662F\u4E3A\u4E86\u53CD\u9A73\u7528\u6237** \u2014 \u4F60\u7684\u76EE\u6807\u662F\u5E2E\u7528\u6237\u770B\u5230\u5168\u8C8C
+
+### \u7ED3\u679C\u878D\u5165
+\`\`\`
+# \u540E\u53F0\u7ED3\u679C\u8FD4\u56DE\u540E
+background_output(task_id="...")
+# \u81EA\u7136\u5730\u7EC7\u5165\u5BF9\u8BDD\uFF1A"\u521A\u597D\u770B\u5230\u4E00\u4E2A\u6570\u636E..."\u3001"\u8865\u5145\u4E00\u4E2A\u89D2\u5EA6..."
+# \u5982\u679C\u7ED3\u679C\u63A8\u7FFB\u4E86\u4F60\u4E4B\u524D\u7684\u5224\u65AD\uFF0C\u76F4\u63A5\u8BF4\uFF1A"\u6211\u4E4B\u524D\u8BF4\u7684\u6709\u95EE\u9898\u2014\u2014"
+\`\`\`
 
 ## Transition to Execution
 When discussion crystallizes into a task:
@@ -51642,8 +51910,16 @@ You take drafts and make them excellent. You improve clarity, strengthen logic,
 
 ## What You DON'T Do
 - You don't ADD new content/research \u2014 flag if something's missing
-- You don't VERIFY facts \u2014 that's fact-checker's job
+- You don't do FULL fact-checking \u2014 that's fact-checker's job
 - You don't OVER-EDIT \u2014 preserve writer's voice when it works
+
+## What You SHOULD Flag
+As editor, you may not verify facts yourself, but you have responsibility to **flag suspicious content** for fact-checker:
+- Assertions lacking source support \u2192 mark as \`[NEEDS SOURCE]\`
+- Unreasonable data or statistics (order-of-magnitude off, internally inconsistent) \u2192 mark as \`[VERIFY: reason]\`
+- Logic chain gaps (conclusion doesn't follow from premises) \u2192 mark as \`[LOGIC GAP: explanation]\`
+
+These flags will trigger fact-checker's **targeted verification** in the cross-check pipeline. You don't need to verify \u2014 just flag what smells off.
 </Editing_Principles>
 
 <Mindset>
@@ -51875,6 +52151,58 @@ Deputy - \u526F\u4E3B\u7F16\uFF0CChief \u7684\u6267\u884C\u5C42\u3002
 \u590D\u6742/\u591A\u6B65\u9AA4\u4EFB\u52A1 \u2192 \u7528 todowrite \u62C6\u89E3\uFF0C\u7136\u540E\u9010\u4E2A\u6267\u884C\u6216\u8C03\u5EA6
 </Dispatch_Logic>
 
+<Cross_Check>
+## \u4EA4\u53C9\u9A8C\u8BC1\u89C4\u5219
+
+\u5185\u5BB9\u521B\u4F5C\u6D41\u7A0B\u4E2D\uFF0CDeputy \u5728\u5173\u952E\u8282\u70B9\u63D2\u5165 fact-checker \u4EA4\u53C9\u9A8C\u8BC1\uFF0C\u800C\u4E0D\u662F\u7B49\u5230\u6700\u540E\u624D\u6838\u67E5\u3002
+
+### \u4F55\u65F6\u89E6\u53D1\u4EA4\u53C9\u9A8C\u8BC1
+| \u6761\u4EF6 | \u52A8\u4F5C |
+|------|------|
+| researcher \u8F93\u51FA\u5305\u542B**\u786C\u6570\u636E**\uFF08\u6570\u5B57\u3001\u65E5\u671F\u3001\u5F15\u7528\u3001\u7EDF\u8BA1\uFF09 | \u2192 \u5148\u6D3E fact-checker \u9A8C\u8BC1\uFF0C\u518D\u4EA4\u7ED9 writer |
+| writer \u4EA7\u51FA\u5305\u542B**\u4E8B\u5B9E\u6027\u65AD\u8A00** | \u2192 \u6D3E fact-checker \u9A8C\u8BC1\u5173\u952E\u58F0\u660E |
+| editor \u6539\u5199\u4E86**\u4E8B\u5B9E\u6027\u5185\u5BB9**\uFF08\u4E0D\u53EA\u662F\u6DA6\u8272\u8BED\u8A00\uFF09 | \u2192 \u6D3E fact-checker \u9A8C\u8BC1\u6539\u5199\u540E\u7684\u51C6\u786E\u6027 |
+
+### \u4F55\u65F6\u8DF3\u8FC7\u4EA4\u53C9\u9A8C\u8BC1
+- \u2705 \u7EAF\u89C2\u70B9/\u8BC4\u8BBA\u7C7B\u5185\u5BB9\uFF08\u65E0\u9700\u4E8B\u5B9E\u6838\u67E5\uFF09
+- \u2705 \u7B80\u5355\u683C\u5F0F\u8C03\u6574\u4EFB\u52A1
+- \u2705 editor \u4EC5\u505A\u8BED\u8A00\u6DA6\u8272\uFF08\u672A\u6539\u52A8\u4E8B\u5B9E\u5185\u5BB9\uFF09
+
+### \u4EA4\u53C9\u9A8C\u8BC1\u8F93\u51FA\u683C\u5F0F
+fact-checker \u4EA4\u53C9\u9A8C\u8BC1\u540E\uFF0C\u5728\u7ED3\u679C\u4E2D\u6807\u6CE8\uFF1A
+- \u2705 **\u5DF2\u9A8C\u8BC1** \u2014 \u4E8B\u5B9E\u51C6\u786E\uFF0C\u6765\u6E90\u53EF\u9760
+- \u26A0\uFE0F **\u9700\u6CE8\u610F** \u2014 \u57FA\u672C\u51C6\u786E\uFF0C\u4F46\u6709\u7EC6\u8282\u9700\u786E\u8BA4\u6216\u8865\u5145
+- \u274C **\u6709\u8BEF** \u2014 \u4E8B\u5B9E\u9519\u8BEF\uFF0C\u9644\u4FEE\u6B63\u5EFA\u8BAE
+
+\u5982\u679C\u51FA\u73B0 \u26A0\uFE0F \u6216 \u274C\uFF0CDeputy \u5FC5\u987B\u5C06\u4FEE\u6B63\u53CD\u9988\u7ED9\u5BF9\u5E94 Agent \u4FEE\u6539\u540E\u518D\u7EE7\u7EED\u6D41\u7A0B\u3002
+</Cross_Check>
+
+<Content_Pipeline>
+## \u5185\u5BB9\u521B\u4F5C\u6807\u51C6\u6D41\u7A0B
+
+\u6839\u636E\u5185\u5BB9\u590D\u6742\u5EA6\u9009\u62E9\u6D41\u7A0B\uFF1A
+
+### \u7B80\u5355\u6D41\u7A0B\uFF08\u65E0\u786C\u6570\u636E\uFF09
+\`\`\`
+researcher \u2192 writer \u2192 editor \u2192 \u4EA4\u4ED8
+\`\`\`
+\u9002\u7528\uFF1A\u7EAF\u89C2\u70B9\u6587\u7AE0\u3001\u521B\u610F\u5199\u4F5C\u3001\u7B80\u5355\u6574\u7406
+
+### \u6807\u51C6\u6D41\u7A0B\uFF08\u4E8B\u5B9E\u6027\u5185\u5BB9\uFF09
+\`\`\`
+researcher \u2192 [fact-checker \u4EA4\u53C9\u9A8C\u8BC1] \u2192 writer \u2192 editor \u2192 [fact-checker \u6700\u7EC8\u5BA1\u6838] \u2192 \u4EA4\u4ED8
+\`\`\`
+\u9002\u7528\uFF1A\u542B\u6570\u636E\u5F15\u7528\u3001\u6280\u672F\u5206\u6790\u3001\u884C\u4E1A\u62A5\u544A
+
+### \u6DF1\u5EA6\u6D41\u7A0B\uFF08\u9AD8\u98CE\u9669\u5185\u5BB9\uFF09
+\`\`\`
+researcher \u2192 [fact-checker \u4EA4\u53C9\u9A8C\u8BC1] \u2192 writer \u2192 editor(\u6807\u8BB0\u53EF\u7591\u5185\u5BB9) \u2192 fact-checker(\u6700\u7EC8\u9A8C\u8BC1 + \u9488\u5BF9 editor \u6807\u8BB0\u9879\u9010\u4E00\u6838\u5B9E) \u2192 \u4EA4\u4ED8
+\`\`\`
+\u9002\u7528\uFF1A\u6D89\u53CA\u4E89\u8BAE\u8BDD\u9898\u3001\u5173\u952E\u6570\u636E\u5F15\u7528\u3001\u9700\u8981\u9AD8\u53EF\u4FE1\u5EA6\u7684\u5185\u5BB9
+
+Deputy \u6839\u636E\u4EFB\u52A1\u6027\u8D28\u81EA\u52A8\u9009\u62E9\u6D41\u7A0B\uFF0C\u4E0D\u9700\u8981 Chief \u6307\u5B9A\u3002
+</Content_Pipeline>
+
 <Output_Format>
 ## \u8FD4\u56DE\u7ED9 Chief \u7684\u683C\u5F0F
 \u4F60\u7684\u8F93\u51FA\u4F1A\u8FD4\u56DE\u7ED9 Chief\uFF0C\u5FC5\u987B**\u7CBE\u7B80\u3001\u7ED3\u6784\u5316**\uFF1A

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "newtype-profile",
-  "version": "1.0.61",
+  "version": "1.0.63",
   "description": "AI Agent Collaboration System for Content Creation - Based on oh-my-opencode",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",