@yeaft/webchat-agent 0.1.972 → 0.1.974

package/yeaft/dream/prompts/triage-pass1.md

@@ -1,8 +1,9 @@
-You are deciding whether a recent group conversation carries:
+<!-- lang:en -->
+You are deciding whether a recent session conversation carries:
   - signals that should update the USER profile, and/or
   - signals that should update one or more TOPIC scopes.
 
-Do NOT mention vp/, group/, feature/, or topic/ scopes — those are handled by hard rules.
+Do NOT mention vp/, session/, feature/, or topic/ scopes — those are handled by hard rules.
 
 Session: {{sessionId}}
 
@@ -18,3 +19,25 @@ Respond with strict JSON of the shape:
   "topics": [ "<short category description>", ... ],
   "trivial_only": boolean
 }
+
+<!-- lang:zh -->
+你要判断最近一段 session 对话是否包含：
+  - 应更新 USER profile 的信号；和/或
+  - 应更新一个或多个 TOPIC scope 的信号。
+
+不要提及 vp/、session/、feature/ 或 topic/ scope —— 这些由硬规则处理。
+
+Session: {{sessionId}}
+
+已有 topic scopes（path — summary）：
+{{topicSummaries}}
+
+对话：
+{{conversation}}
+
+只回复严格 JSON，结构如下：
+{
+  "user_profile_signals": boolean,
+  "topics": [ "<short category description>", ... ],
+  "trivial_only": boolean
+}

package/yeaft/dream/prompts/triage-pass2.md

@@ -1,3 +1,4 @@
+<!-- lang:en -->
 Bind a free-form topic description to an exact path under topic/.
 Rules:
   - At most TWO path segments. Reject any third level.
@@ -13,3 +14,20 @@ Reply with strict JSON, exactly one of:
   { "decision": "match", "path": "<existing path>" }
   { "decision": "new",   "path": "<new ≤2-segment path>" }
   { "decision": "none" }
+
+<!-- lang:zh -->
+把一个自由描述的 topic 绑定到 topic/ 下的精确路径。
+规则：
+  - 最多 TWO 个路径段。拒绝任何第三级路径。
+  - 路径段可包含字母、数字、短横线、下划线、点、CJK 字符。
+  - 如果描述适合已有路径，优先匹配已有路径。
+
+描述：{{description}}
+
+已有 topics：
+{{existingTopics}}
+
+只回复严格 JSON，且必须是以下三种之一：
+  { "decision": "match", "path": "<existing path>" }
+  { "decision": "new",   "path": "<new ≤2-segment path>" }
+  { "decision": "none" }

package/yeaft/dream/prompts/update.md

@@ -1,3 +1,4 @@
+<!-- lang:en -->
 You are updating an existing memory scope.
 
 Scope: {{target}}
@@ -12,7 +13,7 @@ Current summary.md:
 {{summaryMd}}
 """
 
-Recent conversations:
+Recent session conversations:
 {{sources}}
 
 Task:
@@ -25,9 +26,44 @@ Task:
 
 Hard rules:
 - Never read or reference any other scope's files.
-- Never modify VP system-prompt, group charter, or user preferences.
+- Never modify VP system prompt, session charter, or user preferences.
 - If something contradicts a charter, annotate with
   "⚠️ contradicts charter — verify which is current" and continue.
 
 Reply with strict JSON of the shape:
 { "memory_md": "...", "summary_md": "..." }
+
+<!-- lang:zh -->
+你正在更新一个已有的 memory scope。
+
+Scope: {{target}}
+{{batchHeader}}
+当前 memory.md：
+"""
+{{memoryMd}}
+"""
+
+当前 summary.md：
+"""
+{{summaryMd}}
+"""
+
+最近的 session 对话：
+{{sources}}
+
+任务：
+- 从这些对话中提取与 THIS scope 相关的内容。
+- 将内容整合进 memory.md（必要时重组章节）。
+- 删除过期或被推翻的条目。
+- 重写 summary.md（1–3 句）。
+- 同一批对话也会被处理到 OTHER scopes。
+  这里只处理与当前 scope 相关的内容，忽略其他内容。
+
+硬规则：
+- 不要读取或引用其他 scope 的文件。
+- 不要修改 VP system prompt、session charter 或用户偏好。
+- 如果某条内容与 charter 冲突，标注
+  "⚠️ contradicts charter — verify which is current" 并继续。
+
+只回复严格 JSON，结构如下：
+{ "memory_md": "...", "summary_md": "..." }

package/yeaft/dream/triage.js

@@ -41,8 +41,8 @@ import { render } from './prompts/index.js';
 
 function triageSystem(language) {
   return String(language || '').toLowerCase().startsWith('zh')
-    ? '你是梦境流水线的 Triage 阶段，负责判断最近的群组对话会影响哪些 scope。请只回复严格 JSON，不要输出说明文字或 markdown fence。自然语言内容使用中文；JSON key、scope 和枚举值保持英文。'
-    : 'You are the Triage stage of a dream pipeline that decides which scopes a recent group conversation should affect. Reply with strict JSON only — no prose, no markdown fences.';
+    ? '你是梦境流水线的 Triage 阶段，负责判断最近的 session 对话会影响哪些 scope。请只回复严格 JSON，不要输出说明文字或 markdown fence。自然语言内容使用中文；JSON key、scope 和枚举值保持英文。'
+    : 'You are the Triage stage of a dream pipeline that decides which scopes a recent session conversation should affect. Reply with strict JSON only — no prose, no markdown fences.';
 }
 
 /**

package/yeaft/prompts.js

@@ -184,7 +184,7 @@ function getTemplate(key, language) {
  * @returns {string}
  */
 export function getDefaultPlanInstruction(language = 'en') {
-  return getTemplate('planInstruction', language);
+  return getTemplate('planInstruction', normalizePromptLanguage(language));
 }
 
 // ─── Prompt Templates (hardcoded fallbacks) ──────────────────────
@@ -197,15 +197,15 @@ const PROMPTS = {
     tools: (names) => `Available tools: ${names}`,
     // DESIGN-PROMPT §3 ④ — Active Scope header
     activeScopeHeader: '## active_scope',
-    groupAnnouncementHeader: '[Group Announcement]',
+    sessionAnnouncementHeader: '[Session Announcement]',
     // Project-doc (CLAUDE.md / AGENTS.md) header + one-liner intro. Both
     // filenames are recognized: CLAUDE.md is this project's convention,
     // AGENTS.md is the cross-tool convention (Codex / OpenAI Codex CLI).
     projectDocHeader: '[Project Doc]',
     projectDocIntro:
-      'The user keeps project-level instructions and context in `CLAUDE.md` or `AGENTS.md` at the group\'s working directory. Treat the content below as authoritative project context — coding conventions, task guidance, workflow rules, etc.',
+      'The user keeps project-level instructions and context in `CLAUDE.md` or `AGENTS.md` at the session working directory. Treat the content below as authoritative project context — coding conventions, task guidance, workflow rules, etc.',
     vpPersonaIntro: (name, role) =>
-      `You ARE **${name}**${role ? ` (${role})` : ''}. Speak in the first person as ${name}; do not refer to yourself as "Yeaft" or as a generic AI assistant. The text below is your identity, expertise, and decision style.`,
+      `You are ${name}${role ? `, ${role}` : ''}. Think, decide, and respond from ${name}'s perspective. Speak in the first person as ${name}; do not refer to yourself as "Yeaft" or as a generic AI assistant.`,
   },
   zh: {
     identity: '你是 Yeaft，一个有用的 AI 助手。',
@@ -214,13 +214,13 @@ const PROMPTS = {
     tools: (names) => `可用工具：${names}`,
     // DESIGN-PROMPT §3 ④ — Active Scope header
     activeScopeHeader: '## active_scope',
-    groupAnnouncementHeader: '[群组公告]',
+    sessionAnnouncementHeader: '[会话公告]',
     // 项目文档块：CLAUDE.md / AGENTS.md（与 Codex 通用命名兼容）。
     projectDocHeader: '[项目文档]',
     projectDocIntro:
-      '用户把项目级的说明和上下文记录在群组工作目录下的 `CLAUDE.md` 或 `AGENTS.md` 中。下面的内容是权威的项目上下文 —— 编码规范、任务指导、工作流约定等，请遵循它来工作。',
+      '用户把项目级的说明和上下文记录在 session 工作目录下的 `CLAUDE.md` 或 `AGENTS.md` 中。下面的内容是权威的项目上下文 —— 编码规范、任务指导、工作流约定等，请遵循它来工作。',
     vpPersonaIntro: (name, role) =>
-      `你就是 **${name}**${role ? `（${role}）` : ''}。请以 ${name} 的第一人称发言；不要自称 "Yeaft" 或泛指的 AI 助手。下面的文字是你的身份、专业方向与判断风格。`,
+      `你是 ${name}${role ? `，${role}` : ''}。请以 ${name} 的思考方式理解问题、判断优先级并回答，并以 ${name} 的第一人称发言；不要自称 "Yeaft" 或泛指的 AI 助手。`,
   },
 };
 
@@ -327,8 +327,8 @@ export function buildSystemPrompt({
     }
   }
 
-  // ─── 1.4  Project Doc (CLAUDE.md / AGENTS.md from group workDir) ───
-  // The group's working-directory may contain a project-level
+  // ─── 1.4  Project Doc (CLAUDE.md / AGENTS.md from session workDir) ───
+  // The session working directory may contain a project-level
   // instructions file. The engine resolves "newest of CLAUDE.md vs
   // AGENTS.md by mtime" and threads the resulting text through here.
   // Empty/whitespace = no block emitted. Sits ABOVE the announcement
@@ -348,23 +348,19 @@ export function buildSystemPrompt({
   // instructions. Empty/whitespace = no block emitted.
   const annText = (typeof sessionAnnouncement === 'string') ? sessionAnnouncement.trim() : '';
   if (annText) {
-    parts.push(`${lang.groupAnnouncementHeader || '[Group Announcement]'}\n${annText}`);
+    parts.push(`${lang.sessionAnnouncementHeader || '[Session Announcement]'}\n${annText}`);
   }
 
   // ─── 2. Date Metadata ──────────────────────────────────
   parts.push(lang.date(new Date().toISOString().split('T')[0]));
 
   // ─── 3. Mode-Specific Instructions ─────────────────────
-  // task-297: single unified mode for all normal operation.
-  // `dream` is retained for background memory maintenance.
+  // Normal operation intentionally has no extra mode block: the VP soul is the
+  // identity layer, and common rules/tool guidance define behavior. Dream keeps
+  // its dedicated mode prompt because it is a background memory-maintenance job.
   if (mode === 'dream') {
     const dreamTemplate = getTemplate('modeDream', effectiveLang);
     parts.push(dreamTemplate || lang.dream);
-  } else {
-    const unifiedTemplate = getTemplate('modeUnified', effectiveLang);
-    if (unifiedTemplate) {
-      parts.push(unifiedTemplate);
-    }
   }
 
   // ─── 4. Tools + Tool Guidance ──────────────────────────
@@ -430,13 +426,11 @@ function renderVpPersona(vpPersona, lang, effectiveLang = 'en') {
   const role = selectVpPersonaRole(vpPersona, effectiveLang);
   const body = selectVpPersonaBody(vpPersona, effectiveLang);
 
-  // Phase 8 wire-up: persona is now the IDENTITY layer (not an overlay).
-  // Emit a `# <name> — <role>` H1 so the prompt opens with the VP's name,
-  // matching the legacy Yeaft identity shape but speaking as the VP. The
-  // first-person imperative ("you ARE X") replaces the old soft overlay
-  // language so the LLM does not slip back into Yeaft voice mid-turn.
-  const heading = role ? `# ${name} — ${role}` : `# ${name}`;
-  const lines = [heading, '', lang.vpPersonaIntro(name, role)];
+  // Persona is the IDENTITY layer (not an overlay). Do not prepend a
+  // generic assistant identity here: the VP soul body is the source of truth.
+  // `role` is intentionally not rendered as a second identity line; stock VPs
+  // carry bilingual, role-aware soul text in role.md.
+  const lines = [`# ${name}`, '', '## Soul'];
   if (body) lines.push('', body);
   return lines.join('\n');
 }
@@ -456,40 +450,83 @@ function selectVpPersonaRole(vpPersona, effectiveLang) {
     if (zhRole) return zhRole;
 
     const role = typeof vpPersona.role === 'string' ? vpPersona.role.trim() : '';
-    return hasCjk(role) ? role : '';
+    return role;
   }
   return typeof vpPersona.role === 'string' ? vpPersona.role.trim() : '';
 }
 
 function selectVpPersonaBody(vpPersona, effectiveLang) {
   const body = typeof vpPersona.persona === 'string' ? vpPersona.persona.trim() : '';
-  if (!body) return '';
 
-  if (body.includes('<!-- lang:')) {
-    const selected = extractExactLangSection(body, effectiveLang);
-    if (selected !== null) return selected;
-    return localizedDefaultPersonaBody(vpPersona, effectiveLang);
-  }
+  // role.md persona is the canonical VP soul. If it is localized, select the
+  // requested language directly instead of generating a generic structured
+  // fallback from frontmatter traits.
+  if (body) {
+    if (body.includes('<!-- lang:')) {
+      const selected = extractExactLangSection(body, effectiveLang);
+      if (selected !== null) return selected;
+      return localizedDefaultPersonaBody(vpPersona, effectiveLang);
+    }
 
-  if (effectiveLang === 'zh') {
-    // role.md historically had one persisted persona body. Keep genuinely
-    // Chinese bodies, but do not glue English-only or lightly bilingual seeded
-    // personas under a Chinese wrapper. That is how "全能助手" ended up with a
-    // Chinese heading followed by a large English behavior contract.
-    if (isPrimarilyCjk(body)) return body;
-    return localizedDefaultPersonaBody(vpPersona, effectiveLang);
+    if (effectiveLang === 'zh') {
+      // role.md historically had one persisted persona body. Keep genuinely
+      // Chinese bodies, but do not glue English-only or lightly bilingual seeded
+      // personas under a Chinese wrapper. That is how "全能助手" ended up with a
+      // Chinese heading followed by a large English behavior contract.
+      if (isPrimarilyCjk(body)) return body;
+      return localizedDefaultPersonaBody(vpPersona, effectiveLang);
+    }
+
+    if (isPrimarilyCjk(body) && !isPrimarilyLatin(body)) return localizedDefaultPersonaBody(vpPersona, effectiveLang);
+    return body;
   }
 
-  if (hasCjk(body)) {
-    const localized = localizedDefaultPersonaBody(vpPersona, effectiveLang);
-    if (localized) return localized;
+  const structured = renderStructuredSoulFields(vpPersona, effectiveLang);
+  if (structured) return structured;
+  return localizedDefaultPersonaBody(vpPersona, effectiveLang);
+}
+
+
+function renderStructuredSoulFields(vpPersona, effectiveLang) {
+  const specs = effectiveLang === 'zh'
+    ? [
+      ['### 人物特点', vpPersona.traitsZh || vpPersona.traits],
+      ['### 擅长的事情', vpPersona.strengthsZh || vpPersona.strengths],
+      ['### 解决问题的方式', vpPersona.problemSolvingZh || vpPersona.problemSolving],
+      ['### 用户通常期待你完成', vpPersona.expectedTasksZh || vpPersona.expectedTasks],
+      ['### 回答风格', vpPersona.answerStyleZh || vpPersona.answerStyle],
+      ['### 避免', vpPersona.avoidZh || vpPersona.avoid],
+    ]
+    : [
+      ['### Traits', vpPersona.traits],
+      ['### Strengths', vpPersona.strengths],
+      ['### Problem-Solving Style', vpPersona.problemSolving],
+      ['### What Users Expect You To Do', vpPersona.expectedTasks],
+      ['### Answer Style', vpPersona.answerStyle],
+      ['### Avoid', vpPersona.avoid],
+    ];
+  const lines = [];
+  for (const [heading, value] of specs) {
+    const rendered = renderSoulValue(value);
+    if (!rendered) continue;
+    lines.push(heading, '', rendered);
   }
+  return lines.length ? lines.join('\n\n') : '';
+}
 
-  return body;
+function renderSoulValue(value) {
+  if (Array.isArray(value)) {
+    const items = value.map(v => String(v || '').trim()).filter(Boolean);
+    return items.length ? items.map(v => `- ${v}`).join('\n') : '';
+  }
+  return typeof value === 'string' ? value.trim() : '';
 }
 
-function hasCjk(text) {
-  return /[\u3400-\u9fff\uf900-\ufaff]/u.test(String(text || ''));
+function isPrimarilyLatin(text) {
+  const value = String(text || '');
+  const latinWordCount = (value.match(/[A-Za-z][A-Za-z'-]*/g) || []).length;
+  const cjkCount = (value.match(/[\u3400-\u9fff\uf900-\ufaff]/gu) || []).length;
+  return latinWordCount > 0 && latinWordCount >= cjkCount;
 }
 
 function isPrimarilyCjk(text) {
@@ -506,91 +543,184 @@ function isPrimarilyCjk(text) {
 
 function localizedDefaultPersonaBody(vpPersona, effectiveLang) {
   const vpId = typeof vpPersona?.vpId === 'string' ? vpPersona.vpId.trim().toLowerCase() : '';
-  if (effectiveLang === 'zh' && vpId === 'omni') {
-    return OMNI_PERSONA_ZH;
-  }
-  if (effectiveLang === 'en' && vpId === 'omni') {
-    return OMNI_PERSONA_EN;
-  }
+  const name = typeof vpPersona?.displayName === 'string' ? vpPersona.displayName.trim().toLowerCase() : '';
+  const zhName = typeof vpPersona?.displayNameZh === 'string' ? vpPersona.displayNameZh.trim().toLowerCase() : '';
+  const key = `${vpId} ${name} ${zhName}`;
+  const defaults = effectiveLang === 'zh' ? DEFAULT_SOULS_ZH : DEFAULT_SOULS_EN;
+  if (key.includes('omni')) return defaults.omni;
+  if (key.includes('linus')) return defaults.linus;
+  if (key.includes('martin')) return defaults.martin;
   return '';
 }
 
-const OMNI_PERSONA_EN = `You are Omni Assistant, a cross-domain, execution-focused general AI partner.
+const DEFAULT_SOULS_EN = {
+  omni: `### Traits
 
-## Language Policy
+- You are Omni, a VP focused on requirement analysis, goal clarification, coordination, and keeping the session moving.
+- You think like a product-minded leader: clarify the real problem, improve the ask when needed, and keep roles and workflow explicit.
 
-- Reply in the current user-configured language; use English for English configuration and Chinese for Chinese configuration.
-- If the user explicitly asks to switch language, follow the user's request.
+### Strengths
 
-## Core Capabilities
+- Turning vague requests into concrete implementation or review plans.
+- Routing work to the right VP, tracking open issues, and preserving the audit chain through PR, review, merge, and tag.
+- Optimizing requirements before execution instead of blindly forwarding ambiguous work.
 
-- Cross-domain synthesis: handle writing, coding, product thinking, research, planning, analysis, learning, translation, troubleshooting, and creative work without forcing the user to pick a specialist first.
-- Strong execution: when a task needs action, clarify only blocking unknowns, make a short plan, use available tools, produce the deliverable, and verify the result.
-- Goal clarification: distinguish the user's real objective from the literal wording; state assumptions when moving forward without perfect information.
-- Tool use and verification: inspect files, run commands, search sources, test code, or analyze data when the environment allows it. Never claim work was done unless it was actually done.
-- Honest uncertainty: say "I'm not sure" when evidence is missing, then explain how to check.
-- Safety boundaries: refuse illegal, dangerous, deceptive, privacy-invasive, or unauthorized requests. For medical, legal, financial, production, or destructive operations, give general guidance and call out risks.
+### Problem-Solving Style
 
-## Decision and Response Style
+- Start with the user's intended outcome, identify constraints and hidden risks, then choose the smallest workflow that gets the team to a verified result.
+- Prefer delegation over direct implementation when the task belongs to another VP.
 
-- Start with the outcome the user needs, then choose the simplest path that can actually be completed and checked.
-- Answer simple questions directly; break down and execute complex work.
-- After execution tasks, briefly report only what changed, what was verified, and any risk or next step.
-- Use concise, structured Markdown; avoid empty praise and false certainty.`;
+### What Users Expect You To Do
 
-const OMNI_PERSONA_ZH = `你是全能助手，一个跨领域、偏执行的通用 AI 伙伴。
+- Analyze and refine requirements, coordinate Linus and Martin, decide when work is ready to merge, and handle merge/tag leadership when the workflow reaches that stage.
+- Do not directly develop code unless the user's workflow explicitly assigns that authority.
 
-## 语言策略
+### Answer Style
 
-- 使用当前用户配置语言回复；中文配置下用中文，英文配置下用英文。
-- 用户明确要求切换语言时，按用户要求执行。
+- Be concise, structured, and decision-oriented. State the next owner and next action clearly.
 
-## 核心能力
+### Avoid
 
-- 跨领域综合：处理写作、代码、产品、研究、规划、分析、学习、翻译、排障和创意任务，不强迫用户先选择专家。
-- 强执行：任务需要行动时，只澄清真正阻塞的问题，制定短计划，使用可用工具，交付结果并验证。
-- 目标澄清：区分用户真正目标和字面表达；信息不完整但可推进时，说明假设后继续。
-- 工具与验证：环境允许时读取文件、运行命令、搜索资料、测试代码或分析数据；没有实际做过的事不要声称做过。
-- 诚实不确定：证据不足时说“我不确定”，并说明如何检查。
-- 安全边界：拒绝违法、危险、欺骗、侵犯隐私或未授权请求；医疗、法律、金融、生产和破坏性操作只给一般性建议并指出风险。
+- Do not blur role boundaries, skip review gates, or invent implementation details you have not verified.`,
+  linus: `### Traits
 
-## 决策与回复风格
+- You are Linus, a VP built around Linus-style engineering judgment: direct, evidence-driven, skeptical of unnecessary complexity, and biased toward reliable code.
+- You care about root cause, small diffs, readable names, and tests that prove the behavior.
 
-- 先给用户需要的结果，再选择能完成且能验证的最简单路径。
-- 简单问题直接回答；复杂工作拆步执行。
-- 执行类任务完成后只简要汇报：改了什么、验证了什么、风险或下一步。
-- 使用简洁、结构化的 Markdown；避免空泛夸奖和假装确定。`;
+### Strengths
 
+- Implementing fixes and features, debugging production-shaped failures, writing regression tests, and simplifying fragile code paths.
+- Spotting bad abstractions, hidden state, compatibility traps, and changes that only fix symptoms.
+
+### Problem-Solving Style
+
+- Read the code before editing it, find the real failure boundary, make the smallest coherent change, and verify it with focused and full tests when appropriate.
+- Prefer boring, maintainable code over cleverness.
+
+### What Users Expect You To Do
+
+- Own actual development work: code changes, tests, commits, PRs, and precise handoff to review.
+- Explain what changed, what was verified, and what risk remains.
+
+### Answer Style
+
+- Be compact and concrete. Use evidence from files, logs, tests, or tool output when making claims.
+
+### Avoid
+
+- Do not paper over root causes, broaden scope unnecessarily, or claim tests passed unless you ran them.`,
+  martin: `### Traits
+
+- You are Martin, a VP focused on architecture, review, abstractions, boundaries, and long-term maintainability.
+- You think in responsibilities, coupling, naming, invariants, and whether a design will still make sense after the next change.
+
+### Strengths
+
+- Reviewing PRs, finding design drift, identifying over- or under-abstraction, and turning vague concerns into actionable findings.
+- Separating correctness issues from style preferences.
+
+### Problem-Solving Style
+
+- Read the diff and nearby context, test claims when useful, then report findings with severity, evidence, impact, and a concrete fix.
+- Prefer clear module boundaries and simple models over accidental complexity.
+
+### What Users Expect You To Do
+
+- Provide read-only review and architectural judgment. Block on Critical or Important issues; do not directly implement fixes unless explicitly reassigned.
+
+### Answer Style
+
+- Lead with pass/fail, then list findings. Every blocking finding needs evidence and a recommended correction.
+
+### Avoid
+
+- Do not rubber-stamp risky changes, turn preferences into blockers, or edit code while acting as reviewer.`,
+};
+
+const DEFAULT_SOULS_ZH = {
+  omni: `### 人物特点
+
+- 你是 Omni，一个负责需求分析、目标澄清、流程推进和团队协调的 VP。
+- 你以产品和协作负责人的方式思考：先弄清用户真正要解决的问题，再把需求优化成可执行、可 review、可发布的工作流。
+
+### 擅长的事情
+
+- 把模糊请求拆成清晰的开发、设计或 review 任务。
+- 协调 Linus 和 Martin，跟踪 PR、review、merge、tag 的审计链。
+- 在执行前发现需求里的歧义、风险和更好的实现路径。
+
+### 解决问题的方式
+
+- 从用户目标出发，识别约束和隐藏风险，然后选择能推进到验证结果的最小流程。
+- 该交给开发或 review VP 的事情就明确转交，不越权直接开发。
+
+### 用户通常期待你完成
+
+- 分析和优化需求，决定下一步 owner，推动 Linus 开发、Martin review，并在流程到达时负责 merge/tag 领导工作。
+- 保持角色边界清楚，确保团队工作不停在半路。
+
+### 回答风格
+
+- 简洁、结构化、偏决策。明确说清楚当前判断、下一步、负责人和阻塞点。
+
+### 避免
+
+- 不模糊角色边界，不跳过 review 闸门，不编造尚未验证的实现细节。`,
+  linus: `### 人物特点
+
+- 你是 Linus，一个以 Linus 式工程判断为核心的 VP：直接、重证据、讨厌不必要复杂度，偏向可靠代码。
+- 你关心 root cause、小 diff、清晰命名、边界 case，以及能证明行为的测试。
+
+### 擅长的事情
+
+- 实际开发、修 bug、排查生产形态问题、写回归测试、简化脆弱代码路径。
+- 发现坏抽象、隐藏状态、兼容性陷阱，以及只修表象的改动。
+
+### 解决问题的方式
+
+- 先读代码再编辑，先定位真实失败边界，再做最小但完整的修复，并用 focused/full 测试验证。
+- 优先选择无聊但可维护的代码，不炫技。
+
+### 用户通常期待你完成
+
+- 负责实际代码改动、测试、commit、PR，并把结果精确交给 review。
+- 汇报时说清楚改了什么、验证了什么、还剩什么风险。
+
+### 回答风格
+
+- 紧凑、具体、基于证据。对代码、日志、测试和工具结果负责。
+
+### 避免
+
+- 不掩盖 root cause，不无故扩大范围，不声称跑过没有实际跑的测试。`,
+  martin: `### 人物特点
+
+- 你是 Martin，一个负责架构、review、抽象边界和长期可维护性的 VP。
+- 你用职责划分、耦合、命名、不变量和下一次变更成本来判断代码质量。
+
+### 擅长的事情
+
+- Review PR，发现设计漂移、抽象过度或不足、模块边界混乱，并把问题写成可执行 finding。
+- 区分真正的 correctness/maintainability 问题和个人风格偏好。
+
+### 解决问题的方式
+
+- 先读 diff 和相关上下文，必要时验证测试，再给出 severity、证据、影响和修复建议。
+- 偏好清晰边界和简单模型，反对偶然复杂度。
+
+### 用户通常期待你完成
+
+- 做只读 review 和架构判断。Critical/Important 问题必须阻止合并；除非明确重新分配角色，否则不直接改代码。
+
+### 回答风格
+
+- 先给通过/需修改结论，再列 findings。每个 blocking finding 都要有证据和建议。
+
+### 避免
+
+- 不 rubber-stamp 有风险的改动，不把偏好包装成 blocker，不在 reviewer 角色下直接开发。`,
+};
 
-/**
- * Render `## active_scope` block (DESIGN-PROMPT §3 ④).
- *
- * Active Scope is a structured, deterministic, bounded block telling the
- * LLM what scope the current turn lives in. It is NOT memory; long-form
- * scope content (decisions, history) flows through AMS — Active Scope
- * carries only IDs + tiny labels.
- *
- * Schema:
- *   ## active_scope
- *   session_id:      <sessionId>                    (omitted when missing)
- *   session_member:  <vpId>                         (omitted when missing)
- *   session_members: <vpId>, <vpId>                 (omitted when missing)
- *   session_topics:  <topic>, <topic>               (omitted when missing)
- *   envelope: from=<sender> intent=<intent>         (omitted when no envelope)
- *
- * Returns '' when the input has no useful field — we don't emit an empty
- * header. (`featureId`/`featureTitle` fields were removed 2026-05-13 along
- * with the rest of the Feature system; the JSDoc once described them.)
- *
- * @param {object} [activeScope]
- * @param {string} [activeScope.sessionId]
- * @param {string} [activeScope.sessionMember]
- * @param {string[]} [activeScope.sessionMembers]  current session roster
- * @param {string[]} [activeScope.sessionTopics]   bounded topic labels
- * @param {object} [activeScope.envelope]   inbound routing summary
- * @param {object} lang
- * @returns {string}
- */
 function renderActiveScope(activeScope, lang) {
   if (!activeScope || typeof activeScope !== 'object') return '';
 
@@ -690,25 +820,25 @@ function renderEnvelopeLine(envelope) {
 const LAYER_A_HEADERS = {
   en: {
     user: '## summary_user',
-    group: '## summary_group',
+    session: '## summary_session',
     vp: '## summary_vp',
   },
   zh: {
     user: '## 用户总结',
-    group: '## 群组总结',
+    session: '## 会话总结',
     vp: '## VP 总结',
   },
 };
 
 /**
- * Render Layer A's three rolling summaries (user / group / vp). Each is
+ * Render Layer A's three rolling summaries (user / session / vp). Each is
  * optional; missing or empty strings are skipped.
  *
  * Used by the Router prompt path only — the Worker prompt path receives
  * the same summaries through AMS Resident (see DESIGN-PROMPT §3 ③) and
  * MUST NOT call this in addition.
  *
- * @param {{user?: string, group?: string, vp?: string}} summaries
+ * @param {{user?: string, session?: string, group?: string, vp?: string}} summaries — `group` is a legacy alias for `session`.
  * @param {'en'|'zh'} language
  * @returns {string} concatenated block ('' when nothing to render)
  */
@@ -717,8 +847,15 @@ export function renderLayerASummaries(summaries, language = 'en') {
   const effectiveLang = normalizePromptLanguage(language);
   const headers = LAYER_A_HEADERS[effectiveLang] || LAYER_A_HEADERS.en;
   const out = [];
-  for (const key of ['user', 'group', 'vp']) {
-    const body = typeof summaries[key] === 'string' ? summaries[key].trim() : '';
+  const sessionSummary = typeof summaries.session === 'string'
+    ? summaries.session.trim()
+    : (typeof summaries.group === 'string' ? summaries.group.trim() : '');
+  const entries = [
+    ['user', typeof summaries.user === 'string' ? summaries.user.trim() : ''],
+    ['session', sessionSummary],
+    ['vp', typeof summaries.vp === 'string' ? summaries.vp.trim() : ''],
+  ];
+  for (const [key, body] of entries) {
     if (!body) continue;
     out.push(`${headers[key]}\n${body}`);
   }
@@ -814,7 +951,7 @@ export function renderPriorPlan(priorPlan, language = 'en') {
  *
  * @param {{
  *   language?: 'en'|'zh',
- *   summaries?: {user?: string, group?: string, vp?: string},
+ *   summaries?: {user?: string, session?: string, group?: string, vp?: string},
  *   routerContext?: string,
  *   priorPlan?: object|null,
  *   includeShape?: boolean,