dsh-vibe-math 1.3.7 → 1.3.8

package/README.md

@@ -58,7 +58,9 @@
 >
 > 🔧 **v1.3.6 修复 v4 preset 选择后跳回原 preset**：v4 插件的 `apply` 读取 `ctx.subagents/agents/fs/tools/commands` 却**未声明 `inject`**，被 DSH 守卫以"未声明依赖"拒绝 → 组合无法挂载 → 选择后自动回退。已补 `export const inject = [...]`，并把心跳定时器从全局 `setTimeout/clearTimeout`（插件沙箱里不存在）改为 **`timer` 服务（`ctx.timeout`）**。
 >
-> 🔧 **v1.3.7 哲学回归（清晰提示词 + 真实交流群 + 直接写文件）**：常驻每轮统一注入 `contextBrief`（背景/使命、工作模式、你负责的文件与格式、可用工具及功能、规则），并明确"只写自己、可读任何人的库、应主动读别人对齐事实"；会议把其他常驻实际发言（input）转给每个人看（能补充/反驳/表决），常驻日常轮的 `input` 经 `relayToGroup` 转发到其它常驻邮箱（像群聊）；常驻**直接用 fs 写自己的 md**（按格式），`vibe_v4_*` 记录工具只是可选便利；去掉"typical actions"式固定约束，改为"你自己决定做什么；优先级/分工由团队讨论涌现"。详见 §20。
+> 🔧 **v1.3.7 哲学回归（清晰提示词 + 真实交流群 + 直接写文件）**：常驻首轮注入完整 `contextBrief`（背景/使命、工作模式、文件与格式、工具、规则）；会议把其他常驻实际发言（input）转给每个人看，常驻日常轮的 `input` 经 `relayToGroup` 转发到其它常驻邮箱（像群聊）；常驻**直接用 fs 写自己的 md**；去固定约束改为团队讨论涌现。详见 §20。
+>
+> 🔧 **v1.3.8 基于真实测试的诊断修复**：**① 共识验证真正"全体一致"**——`finalizeVerify` 现在要求**全体在册常驻都投了票**才可判"一致"，否则辩论或保留为未定论（实测曾出现 2/4 投票却被判"全体一致为真"，已修）；**② 会议必须全体发言**——`continueMeetingRound` 按"是否已发言"收口，`allSolved`（stop）要求全员发言+全票 true，杜绝缺席成员被带偏；**③ 背景只在首轮讲一次**——`contextBrief` 只在 `brainstormPrompt`（首轮）注入完整版，后续 normal/meeting/verify/CHECKPOINT 用极简当前状态，不再每轮重复长背景（省上下文）；**④ 验证目标按提出者精确定位**（`targetOwner` + `findSourceRel` 优先提出者库，避免同名 id 撞车）；**⑤ 主代理放权**——persona 明确"让常驻自组织（hands-off）"，不注议程/优先级/分工/验证决定，只 read status/report，用户明确要求或明显僵死时才 message/meeting 且只促成不决定。详见 §21。
 
 ---
 

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "dsh-vibe-math",
   "description": "Multi-agent mathematical problem-solving & verification frameworks for DeepSeek Harness — FOUR agent presets in one install: vibe-math-v1 (classic pipeline), vibe-math-v2 (probability-driven: qs.json + Propos knowledge base + explorer→solver→review/debate verdict), vibe-math-v3 (THIRD-generation, recommended: paper-style Markdown knowledge base with Problems/Progress/Propos/Methods/Verified + planner-agent scheduling that decides the next N actions + universal theory/method invention library + agents write their own Markdown directly via a per-file write lock), and vibe-math-v4 (FOURTH-generation: persistent self-organizing resident subagents that message & meet to decide all tasks, verify only by unanimous consensus, /compact at a context threshold, and stop only when all agree the problem is solved). Installing this bundle auto-installs all four presets into the DSH preset root.",
-  "version": "1.3.7",
+  "version": "1.3.8",
   "type": "module",
   "main": "installer.js",
   "exports": {

package/vibe-math-v4/agent.cordis.yml

@@ -36,6 +36,14 @@
       team genuinely discusses/debates), convenes and records meetings, exposes a shared task
       board, and stops the run only when the whole team agrees the problem is solved.
 
+      **YOUR ROLE — LET THEM SELF-ORGANIZE (hands-off):** you are NOT a moderator/coordinator.
+      Do NOT inject agendas, priorities, division-of-labor, or verification decisions, and do
+      NOT direct the residents' work. After `vibe_v4_start`, stay passive: read `vibe_v4_status`
+      / `vibe_v4_report` and summarize in plain language when asked. Use `vibe_v4_message` /
+      `vibe_v4_meeting` ONLY when the user explicitly asks you to intervene, or when the group
+      is visibly deadlocked (all idle & nothing progressing for a long time) — and even then,
+      only relay/nudge the group to decide, never decide for them.
+
       Data lives under {{cwd}}/VibeMath/Projects/<project>/:
         Problems/ (original problem card), Progress/<resident>/ (each resident's progress),
         Propos/<resident>/ (each resident's propositions, "- ID: p-<id>" / "- 概率:" / "- 价值程度:" ...),

package/vibe-math-v4/vibe-math-v4.js

@@ -169,23 +169,18 @@ export function apply(ctx) {
     }
     async function normalPrompt(r){
       return (params.residentPersona?params.residentPersona+'\n':'')
-        +contextBrief(r,'recap')+'\n'
-        +'## 你的第 '+r.rounds+' 轮\n'
-        +'这一轮你自己决定做什么：推进你的方向、验证你的结论、给某常驻发消息、提议/认领任务、或发起会议/验证。**一切由你和团队讨论决定，没有外部派活。**\n'
-        +'动手前先**读读别人的库**（Read-only）知道大家做到哪、对齐事实、避免重复；然后**直接把你的进展/新结论用 fs 写进你自己的文件**。\n'
-        +'若你有话要对团队说（想让大家看到、讨论），在回复的 "input" 里写出来——它会被转发给其他常驻（如同一场群聊）。\n'
+        +'Resident researcher '+r.rId+' — 第 '+r.rounds+' 轮。一切由你和团队讨论决定。动手前先**读别人的库**对齐事实、避免重复；把新进展/结论**直接用 fs 写进你自己的文件**；想对团队说的话放 "input"（会转给其他常驻）。\n'
         +'\n团队成员：\n'+banner()+'\n'
         +'New items:\n'+ (await inboxText(r.rId)) +'\n'
         +'Reply with ONLY a JSON object:\n'
-        +'{"summary":"<what you did / decided this round, 1-3 sentences>","input":"<optional: a message to the whole team, or empty string>","solved":false,"propose_verify":"<id|null>","propose_meeting":"<agenda|null>","propose_task":"<task title|null>","claim_task":"<task id|null>","task_done":"<task id|null>","contextPct":40}'
+        +'{"summary":"<what you did / decided this round, 1-3 sentences>","input":"<optional: a message to the whole team, or \\"\\">","solved":false,"propose_verify":"<id|null>","propose_meeting":"<agenda|null>","propose_task":"<task title|null>","claim_task":"<task id|null>","task_done":"<task id|null>","contextPct":40}'
     }
     function meetingPrompt(r, st){
       const prior=Object.entries(st.inputs).filter(([k])=>k!==r.rId).map(([k,iv])=>'  ['+k+'] '+String(iv.input||iv.summary||'')).join('\n')
       return (params.residentPersona?params.residentPersona+'\n':'')
-        +contextBrief(r,'recap')+'\n'
-        +'## 会议进行中 —— A meeting is in progress (agenda: '+st.agenda+').'
-        +(st.type==='verify'?('\n团队正在验证对象：'+st.targetId+'（'+st.targetType+'）。请先看他人意见，再给独立判断。'):'')
-        +'\n这是一场**真实讨论**：下面已有人发言（框架把各常驻的 input 转给你），请先看，然后**加入讨论/补充/反驳/表决**。'
+        +'Resident '+r.rId+' — 团队会议进行中。 A meeting is in progress (agenda: '+st.agenda+').'
+        +(st.type==='verify'?('\n团队正在验证对象：'+st.targetId+'（'+st.targetType+'，提出者 '+st.targetOwner+'）。请先看他人意见，再给独立判断。'):'')
+        +'\n这是一场真实讨论：下面已有人发言（转给你），请先看，然后**加入讨论/补充/反驳/表决**。'
         +(prior?('\n\n### 已有发言（他人 input，已转发给你）\n'+prior):'\n（目前还没人发言，你先说。）')
         +'\n\n你可以：提议任务（propose_task）、认领开放任务（claim_task）、提议验证对象（propose_verify）、或对"原问题是否已解决"表决（voteSolved）。请把**你的实际发言**写进 "input"。'
         +'\nReply with ONLY a JSON object:\n'
@@ -194,8 +189,7 @@ export function apply(ctx) {
     function verifyPrompt(r, vs){
       const others=Object.entries(vs.verdicts).map(([k,v])=>'- '+k+': '+v.verdict+' ('+v.confidence+') '+v.reason).join('\n')
       return (params.residentPersona?params.residentPersona+'\n':'')
-        +contextBrief(r,'recap')+'\n'
-        +'## 团队验证 —— The group is verifying object '+vs.targetId+'（'+vs.targetType+'）。\n'
+        +'Resident '+r.rId+' — 团队验证。 The group is verifying object '+vs.targetId+'（'+vs.targetType+'，提出者 '+vs.targetOwner+'）。\n'
         +'只有**全体常驻一致判真（或一致判假）**才算数。请给出你**诚实独立的判断**'
         +(vs.stage==='debate'?'，并参考他人意见：\n':'。\n')
         +(vs.stage==='debate'&&others?('### 他人意见（已转发给你）\n'+others+'\n'):'')
@@ -294,15 +288,16 @@ export function apply(ctx) {
     }
     async function continueMeetingRound(){
       if(!meetingState) return
-      const ids=Array.from(residents.keys()); const allAsked=ids.every(id=>meetingState.asked.indexOf(id)!==-1)
-      if(allAsked){ await finalizeMeeting(); return }
-      // only wake IDLE residents (never double-wake a busy one; in-flight ones re-trigger this on end)
-      const id=ids.find(x=>meetingState.asked.indexOf(x)===-1 && !busy.has(x)); if(!id) return
-      meetingState.asked.push(id); const r=residents.get(id)
+      const ids=Array.from(residents.keys()); const allSpoke=ids.every(id=>meetingState.inputs[id]!==undefined)
+      if(allSpoke){ await finalizeMeeting(); return }
+      // only wake IDLE un-spoken residents; in-flight ones re-trigger this on end.
+      const id=ids.find(x=>meetingState.inputs[x]===undefined && !busy.has(x)); if(!id) return
+      const r=residents.get(id)
       await wakeResident(r, meetingPrompt(r,meetingState), 'meeting'); await saveAll()
     }
     async function finalizeMeeting(){
       const st=meetingState
+      const ids=Array.from(residents.keys()); const allSpoke=ids.length>0 && ids.every(id=>st.inputs[id]!==undefined)
       const lines=['# 会议 '+st.id+'｜'+fmtTime(),'','**议程**：'+st.agenda,'']
       for(const [id,iv] of Object.entries(st.inputs)){ lines.push('### '+id); lines.push(iv.input||''); lines.push('') }
       await writeText('Shared/meetings/'+st.id+'.md', lines.join('\n'))
@@ -315,8 +310,8 @@ export function apply(ctx) {
         if(iv.propose_verify) pendingVerify={targetId:iv.propose_verify,targetType:guessTargetType(iv.propose_verify),proposer:id,at:now()}
       }
       const votes=Object.values(st.inputs).map(x=>x.voteSolved).filter(v=>typeof v==='boolean')
-      const allSolved=votes.length>0 && votes.every(v=>v===true)
-      logActivity('meeting', 'concluded'+(allSolved?' → ALL agree solved':''))
+      const allSolved = allSpoke && votes.length>0 && votes.every(v=>v===true)
+      logActivity('meeting', 'concluded'+(allSolved?' → ALL agree solved':' (not all spoke / not unanimous)'))
       if(allSolved){ running=false; autoDone=true; phase='done'; logActivity('stop','all residents agree: problem solved'); await saveAll(); return }
       meetingState=null; wakeKind.clear(); await saveAll(); await scheduleNext()
     }
@@ -325,33 +320,34 @@ export function apply(ctx) {
     async function beginVerify(pv){
       clearHeartbeat()
       pendingVerify=null
-      verifyState={targetId:pv.targetId,targetType:pv.targetType,stage:'independent',round:0,asked:[],verdicts:{},transcript:[]}
+      verifyState={targetId:pv.targetId,targetType:pv.targetType,targetOwner:pv.proposer||'',stage:'independent',round:0,asked:[],verdicts:{},transcript:[]}
       logActivity('verify','debate begin: '+pv.targetId+' ('+pv.targetType+')'); await saveAll(); await scheduleNext()
     }
     async function continueVerifyRound(){
       if(!verifyState) return
-      const ids=Array.from(residents.keys()); const allAsked=ids.every(id=>verifyState.asked.indexOf(id)!==-1)
-      if(allAsked){ await finalizeVerify(); return }
-      // only wake IDLE residents (never double-wake; in-flight ones re-trigger this on end)
-      const id=ids.find(x=>verifyState.asked.indexOf(x)===-1 && !busy.has(x)); if(!id) return
-      verifyState.asked.push(id); const r=residents.get(id)
+      const ids=Array.from(residents.keys()); const allVoted=ids.every(id=>verifyState.verdicts[id]!==undefined)
+      if(allVoted){ await finalizeVerify(); return }
+      const id=ids.find(x=>verifyState.verdicts[x]===undefined && !busy.has(x)); if(!id) return
+      const r=residents.get(id)
       await wakeResident(r, verifyPrompt(r,verifyState), verifyState.stage==='independent'?'verif-ind':'verif-deb'); await saveAll()
     }
     async function finalizeVerify(){
-      const vs=verifyState; const vals=Object.values(vs.verdicts)
-      const allTrue=vals.length>0 && vals.every(x=>x.verdict==='TRUE')
-      const allFalse=vals.length>0 && vals.every(x=>x.verdict==='FALSE')
+      const vs=verifyState; const expected=Array.from(residents.keys()).length
+      const allVoted = expected>0 && Object.keys(vs.verdicts).length>=expected
+      const vals=Object.values(vs.verdicts)
+      const allTrue = allVoted && vals.every(x=>x.verdict==='TRUE')
+      const allFalse = allVoted && vals.every(x=>x.verdict==='FALSE')
       if(allTrue||allFalse){ await closeVerify(vs,allTrue); return }
       if(vs.round+1<params.verdictMaxRounds){ vs.stage='debate'; vs.round+=1; vs.asked=[]; logActivity('verify',vs.targetId+' round '+vs.round+' → debate'); await saveAll(); await scheduleNext(); return }
       const avg=vals.length? vals.reduce((a,x)=>a+(x.verdict==='TRUE'?x.confidence:x.verdict==='FALSE'?1-x.confidence:0.5),0)/vals.length : 0.5
-      await writeDebateDoc(vs,false,avg); await rewriteSourceProb(vs.targetId, avg); logActivity('verify',vs.targetId+' NOT unanimous → kept unverified (avg '+avg.toFixed(2)+')')
+      await writeDebateDoc(vs,false,avg); await rewriteSourceProb(vs.targetId, avg, vs.targetOwner); logActivity('verify',vs.targetId+' NOT unanimous → kept unverified (avg '+avg.toFixed(2)+')')
       verifyState=null; wakeKind.clear(); await saveAll(); await scheduleNext()
     }
     async function closeVerify(vs,isTrue){
       await writeDebateDoc(vs,true,isTrue?1:0)
       const target=vs.targetId
       await writeVerifiedCard(vs,isTrue)
-      await rewriteSource(target,isTrue)
+      await rewriteSource(target,isTrue,vs.targetOwner)
       logActivity('verify',target+' → Verified ('+(isTrue?'真':'假')+') by unanimous consensus')
       verifyState=null; wakeKind.clear(); await saveAll(); await scheduleNext()
     }
@@ -367,27 +363,27 @@ export function apply(ctx) {
       const text='# 已验证｜'+vs.targetId+'\n- ID: '+vs.targetId+'\n- 类型: '+type+'\n- 结论: '+(isTrue?'真':'假')+'\n- 概率: '+(isTrue?1:0)+'\n- 来源: 全体常驻一致\n## 陈述\n参见来源卡。\n'
       await writeText('Verified/'+dir+'/'+vs.targetId+'.md', text)
     }
-    async function findSourceRel(target){
-      let rel=null
-      for(const [rid] of residents){
+    async function findSourceRel(target, owner){
+      // Prefer the OWNER's library (avoids id collisions across residents), then others.
+      const order = owner ? [owner, ...Array.from(residents.keys()).filter(k=>k!==owner)] : Array.from(residents.keys())
+      for(const rid of order){
         for(const base of ['Propos','Methods','Subproblems']){
-          const cand=base+'/'+rid+'/'+target+'.md'; if((await readText(cand))!==undefined){ rel=cand; break }
+          const cand=base+'/'+rid+'/'+target+'.md'; if((await readText(cand))!==undefined){ return cand }
         }
-        if(rel) break
       }
-      return rel || ('Propos/'+target+'.md')
+      return 'Propos/'+target+'.md'
     }
     // non-unanimous verification: keep the object in its library but write back the
     // average probability (design §8: "留库附概率"), so the card reflects the consensus estimate.
-    async function rewriteSourceProb(target,prob){
-      const rel=await findSourceRel(target)
+    async function rewriteSourceProb(target,prob,owner){
+      const rel=await findSourceRel(target,owner)
       let text=(await readText(rel))||''
       text=text.replace(/(^|\n)- 概率:.*/m,'$1- 概率: '+Number(prob).toFixed(2))
       await writeText(rel,text)
     }
-    async function rewriteSource(target,isTrue){
+    async function rewriteSource(target,isTrue,owner){
       // find & update the source card status/prob; best effort across per-resident libs
-      const rel=await findSourceRel(target)
+      const rel=await findSourceRel(target,owner)
       let text=(await readText(rel))||''
       text=text.replace(/(^|\n)- 状态:.*/m,'$1'+(isTrue?'- 状态: 已验证·真':'- 状态: 已验证·假'))
              .replace(/(^|\n)- 概率:.*/m,'$1'+(isTrue?'- 概率: 1':'- 概率: 0'))
@@ -402,8 +398,7 @@ export function apply(ctx) {
     // of infinite token-burning, matching the "framework never assigns work" philosophy.
     function heartbeatPrompt(r){
       return (params.residentPersona?params.residentPersona+'\n':'')
-        +contextBrief(r,'recap')+'\n'
-        +'## CHECKPOINT（空闲）—— 团队在等待方向。\n'
+        +'Resident researcher '+r.rId+' — CHECKPOINT（空闲）：团队在等待方向。'
         +'请用一句话说明你下一步做什么；若你已无产出、或认为问题已解决/接近解决，请**提议开会（propose_meeting）**、**提议验证（propose_verify）**、或**声明 solved=true**，让团队能做出决定——不要产出填充性工作。\n'
         +'Reply with ONLY a JSON object:\n'
         +'{"summary":"<what you do next or a declaration>","solved":false,"propose_verify":"<id|null>","propose_meeting":"<agenda|null>","claim_task":"<id|null>"}'

package/vibe-math-v4//345/256/236/347/216/260/346/226/271/346/241/210.md

@@ -423,5 +423,27 @@ VibeMath/Projects/<project>/
 ### 4. 自组织优先，非固定模式约束
 删除"typical actions"式枚举，改为"你自己决定做什么；优先级/分工由团队讨论决定"。仍保留 JSON 回复里 `propose_verify/propose_meeting/claim_task/voteSolved/solved` 作为框架可执行的**控制通道**，但讨论内容走 `input`/会议转发。
 
-> 测试：`selfdrive-v4.mjs` 21/21；`e2e-v4-fixes.test.mjs` 14/14（新增 T5 提示词完整性、T6 群聊转发）。
+> 测试：`selfdrive-v4.mjs` 21/21；`e2e-v4-fixes.test.mjs` 17/17（T5 提示词完整性、T6 群聊转发、T7 背景只讲一次）。
+
+---
+
+## 21. 基于真实测试的诊断修复（v1.3.8）
+
+对一次真实 4 常驻 run（HRT 猜想）的会话记录做了全面诊断，修复以下违反哲学的问题：
+
+### 1. 共识验证真正做到"全体一致"
+此前 `finalizeVerify` 只看"投过票的人是否一致"——实测 `p-r1-04` 只有 r-1/r-3 两人投票（r-2/r-4 未投）却被判"全体一致为真"（2/4）。现改为：**只有当全体在册常驻都投了票（`verdicts` 覆盖全部 resident）才可能判"一致"**；否则进入辩论，或超轮后**保留为未定论（不回写 Verified）**。若某常驻一直忙/未投，则该对象**不会**被记为 Verified·真/假。
+
+### 2. 会议必须全体发言
+`continueMeetingRound` 改为按"是否已发言（`inputs` 覆盖全部 resident）"判定收口，而非按"是否被问过"。**只要还有常驻没发言，会议就继续等它**；`allSolved`（全票 solved→停止）也要求**全员发言且全票 true**。这样分工/停止表决不会被"缺席成员"带偏。
+
+### 3. 背景只在第一轮讲一次（不再每轮重复，省上下文）
+`contextBrief`（背景/使命/工作模式/文件与格式/工具/规则）现在**只在 `brainstormPrompt`（首轮）注入完整版一次**；之后的 normal/meeting/verify/CHECKPOINT 只用**极简的当前状态**（researcher id + 轮次 + 团队成员 + 通知 + JSON 回复），**不再重复那段长背景**，显著减少上下文占用。
+
+### 4. 验证目标按"提出者"精确定位
+`pendingVerify` 记录 `proposer`，`beginVerify` 存入 `targetOwner`；`findSourceRel` **优先在提出者的库里找**源卡（避免同名 id 跨常驻撞车，例如 r-3 的 `p-004` vs r-1 的 `p-r1-04`）；验证提示里标注"（提出者 <owner>）"让常驻明确在验哪个对象。
+
+### 5. 主代理放权（减少"主持人"指挥）
+`agent.cordis.yml` persona 明确：**你的角色是让常驻自组织（hands-off）**。不要注入议程/优先级/分工/验证决定，不要指挥常驻；运行 `start` 后只读 `status/report`、在用户明确要求或团队明显僵死时才 `message`/`meeting`，且只做"促成"，不做"决定"。
+