@goodandready/dsh-moa 0.2.38 → 0.2.39

package/README.md

@@ -140,7 +140,7 @@ Configure your models in `Settings → Plugins → Mixture of Agents`:
 Inspect line-by-line differences between candidate proposals and the curator's synthesized deliverable directly in the UI. Features file selection, delta line highlights (added, removed, same), and unified diff rendering via a zero-dependency in-memory LCS algorithm.
 
 ### 8. Pre-Promotion Git Checkpoints (`dsh-time-machine`)
-Before promoting any winning candidate files over the workspace root, `dsh-moa` invokes the local `dsh-time-machine` service to create a shadow Git checkpoint (`moa-pre-promotion: candidate-N`). If `dsh-time-machine` is absent or unreachable, file promotion proceeds seamlessly via best-effort fallback.
+Before promoting any winning candidate files over the workspace root, `dsh-moa` invokes the local `dsh-time-machine` service to create a shadow Git checkpoint (`moa-pre-promotion: candidate-N`). If `dsh-time-machine` is absent or unreachable, file promotion is blocked (fail-closed) to protect uncheckpointed workspace files against unintended overwrites (override with `force: true` only when automated checkpoints are explicitly unneeded).
 
 ### 9. Automated Test Execution Gate
 When `test_gate_enabled: true` and `test_command` (e.g. `npm test` or `pytest`) are configured, candidate code is executed in an ephemeral sandbox overlay. The judge receives concrete test outcomes, durations, and output logs to ground decisions in objective verification.

package/README.ru.md

@@ -140,7 +140,7 @@ graph TD
 Позволяет визуально сопоставлять построчные различия между решениями моделей-советников и итоговым синтезом судьи прямо в интерфейсе. Поддерживает выбор конкретного файла, подсветку добавленных/удаленных строк и вычисление дельт с помощью легковесного алгоритма LCS без внешних зависимостей.
 
 ### 8. Теневые Git-чекпоинты (`dsh-time-machine`)
-Перед промоушном файлов победителя в корень рабочей области `dsh-moa` автоматически обращается к сервису `dsh-time-machine` для создания снимка (`moa-pre-promotion: candidate-N`). Если плагин недоступен, промоушн штатно продолжается по схеме best-effort.
+Перед промоушном файлов победителя в корень рабочей области `dsh-moa` автоматически обращается к сервису `dsh-time-machine` для создания снимка (`moa-pre-promotion: candidate-N`). Если плагин недоступен или создание снимка завершилось ошибкой, перезапись файлов блокируется во избежание потери несохранённых данных (переопределяется параметром `force: true`, только если создание снимков намеренно отключено).
 
 ### 9. Автоматический шлюз выполнения тестов (Automated Test Execution Gate)
 При включении `test_gate_enabled: true` и указании тестовой команды (`test_command`, например `npm test` или `pytest`) каждый кандидат запускается в изолированном оверлейном окружении. Судья получает объективный вердикт с кодами возврата, временем исполнения и логами тестов до вынесения итогового решения.

package/README.zh.md

@@ -140,7 +140,7 @@ graph TD
 直接在界面中可视化对比顾问方案之间以及顾问与裁判综合方案之间的逐行差异。支持文件筛选、增删改高亮，基于纯 JavaScript LCS 差异算法，零第三方依赖。
 
 ### 8. 提升前 Git 快照检查点 (`dsh-time-machine`)
-在将胜出候选者的文件提升并覆盖至项目根目录前，`dsh-moa` 自动调用本地 `dsh-time-machine` 建立影子 Git 检查点 (`moa-pre-promotion: candidate-N`)。如未安装则静默回退，不阻断提升流程。
+在将胜出候选者的文件提升并覆盖至项目根目录前，`dsh-moa` 自动调用本地 `dsh-time-machine` 建立影子 Git 检查点 (`moa-pre-promotion: candidate-N`)。如 `dsh-time-machine` 未安装或无法访问，文件提升将被阻断以防止未备份的项目文件被意外覆盖（仅在明确不需要检查点时可通过 `force: true` 覆盖）。
 
 ### 18. 平滑降级与本地容灾恢复 (Local Fallback Resilience)
 当 `local_fallback_enabled: true` 时，因网络中断、超时或速率限制（429/500）而失败的候选模型会自动使用配置的本地模型（Ollama / MiniPC）恢复。每个后备模型都拥有独立的超时执行配额与关联主 turn signal 的隔离取消控制器，防止云端超时耗尽本地恢复时间。同时严格检测终止错误与流中断，确保安全容灾且杜绝已取消 turn 的延迟文件提升。

package/lib/index.js

@@ -36,15 +36,18 @@ export function apply(ctx, config) {
   if (ctx.logger) setLogger(ctx.logger)
   refreshCatalogInBackground()
 
-  let localConfig = plainConfig(config ?? {})
-  const getConfig = () => localConfig
+  let localOverrides = null
+  const getSourceConfig = () => {
+    const liveSource = plainConfig(ctx.config ?? config ?? {})
+    return localOverrides ? { ...liveSource, ...localOverrides } : liveSource
+  }
   const live = () => {
     try {
-      const raw = getConfig() ?? {}
+      const raw = getSourceConfig() ?? {}
       const validated = plainConfig(Config(plainConfig(raw)))
-      return plainConfig(validated) ?? localConfig
+      return plainConfig(validated) ?? raw
     } catch {
-      return localConfig
+      return getSourceConfig()
     }
   }
 
@@ -72,17 +75,13 @@ export function apply(ctx, config) {
 
   const saveConfig = async (newFields) => {
     const plain = plainConfig(newFields)
-    localConfig = { ...localConfig, ...plain }
-    try {
-      if (settingsService && typeof settingsService.replace === 'function') {
-        await settingsService.replace(NS, plain)
-      } else if (settingsService && typeof settingsService.update === 'function') {
-        await settingsService.update(NS, plain)
-      }
-    } catch (err) {
-      logger.warn('[dsh-moa] Settings persistence warning:', err?.message || err)
+    if (settingsService && typeof settingsService.replace === 'function') {
+      await settingsService.replace(NS, plain)
+    } else if (settingsService && typeof settingsService.update === 'function') {
+      await settingsService.update(NS, plain)
     }
-    return localConfig
+    localOverrides = { ...(localOverrides || {}), ...plain }
+    return live()
   }
 
   const callLlm = createLlmCaller(ctx)
@@ -110,6 +109,8 @@ export function apply(ctx, config) {
     saveConfig,
     Config,
     plainConfig,
+    webServerPort: ctx.webServer?.port,
+    timeMachineEngine: ctx.get?.('timeMachineEngine') || ctx.timeMachineEngine,
   })
 
   // Session turn interceptor: route turn to real aggregator model & stream MoA synthesis
@@ -166,14 +167,15 @@ export function apply(ctx, config) {
       if (userText.startsWith('/moa')) {
         const cfg = live()
         if (cfg.enabled !== false) {
-          const parsed = parseMoACommand(userText, cfg.presets || []) || { prompt: userText, presetName: cfg.default_preset || 'default' }
-          const isExplicitPreset = userText.includes('--preset') || userText.includes('-p ')
+          const effectiveDefault = cfg.default_preset || 'default'
+          const parsed = parseMoACommand(userText, cfg.presets || [], effectiveDefault) || { prompt: userText, presetName: effectiveDefault, isExplicit: false }
+          const isExplicitPreset = Boolean(parsed.isExplicit || userText.includes('--preset') || userText.includes('-p '))
           if (!isExplicitPreset && (cfg.smart_routing_enabled || cfg.presets?.some((p) => p.smart_routing_enabled))) {
             try {
               const routed = await resolvePresetForPrompt({
                 prompt: parsed.prompt || userText,
                 presets: cfg.presets || [],
-                defaultPreset: parsed.presetName || cfg.default_preset || 'default',
+                defaultPreset: parsed.presetName || effectiveDefault,
                 routingModel: cfg.smart_routing_model,
                 callLlm,
                 enabled: true,
@@ -185,7 +187,12 @@ export function apply(ctx, config) {
               // fallback silently
             }
           }
-          const targetPreset = (cfg.presets || []).find((p) => p.name === parsed.presetName) || cfg.presets?.[0]
+          const targetPreset = (cfg.presets || []).find((p) => p.name === (parsed.presetName || effectiveDefault))
+            || (!isExplicitPreset ? (cfg.presets || []).find((p) => p.name === effectiveDefault) || cfg.presets?.[0] : null)
+          if (!targetPreset) {
+            logger.warn(`[dsh-moa] Unknown explicit preset "${parsed.presetName}" in /moa command.`)
+            return next()
+          }
           const aggProvider = targetPreset?.aggregator?.provider || ''
           const aggModel = targetPreset?.aggregator?.model || ''
           const sessCwd = event.cwd || event.session?.cwd || (event.session?.path ? path.dirname(event.session.path) : undefined) || process.cwd()

package/lib/moa-budget.js

@@ -31,6 +31,7 @@ export function estimateCandidateRunCost(slots = [], prices = {}, promptTokens =
 
 /**
  * Applies budget guardrails before initiating parallel candidate execution.
+ * Evaluates candidate runs, peer critique (Round 2) and downstream synthesis phases.
  */
 export function applyBudgetGuardrails({
   references = [],
@@ -39,6 +40,11 @@ export function applyBudgetGuardrails({
   action = 'trim', // 'trim' | 'abort'
   prices = {},
   promptLength = 1000,
+  messages = [],
+  maxTokens = 2048,
+  peerCritiqueEnabled = false,
+  aggregator = null,
+  judgeModels = [],
 }) {
   if (!enabled || typeof maxBudgetUsd !== 'number' || maxBudgetUsd <= 0 || references.length === 0) {
     return {
@@ -49,15 +55,52 @@ export function applyBudgetGuardrails({
     }
   }
 
-  const promptTokens = Math.max(50, Math.round(promptLength / 4))
-  const { totalEstimatedUsd, slotEstimates } = estimateCandidateRunCost(references, prices, promptTokens)
+  // Calculate full input tokens from userPrompt length + messages context
+  let contextChars = Number(promptLength) || 1000
+  if (Array.isArray(messages) && messages.length > 0) {
+    for (const m of messages) {
+      if (typeof m?.content === 'string') {
+        contextChars += m.content.length
+      } else if (Array.isArray(m?.content)) {
+        for (const part of m.content) {
+          if (part?.text) contextChars += part.text.length
+        }
+      }
+    }
+  }
+  const promptTokens = Math.max(50, Math.round(contextChars / 4))
+  const expectedOutputTokens = typeof maxTokens === 'number' && maxTokens > 0 ? maxTokens : 2048
+
+  // Base candidate cost per slot (Round 1)
+  const { totalEstimatedUsd, slotEstimates } = estimateCandidateRunCost(references, prices, promptTokens, expectedOutputTokens)
+
+  // Estimate fixed cost for downstream phases (Aggregator / Judge panel)
+  let downstreamUsd = 0
+  if (aggregator?.provider && aggregator?.model) {
+    const aggInput = promptTokens + expectedOutputTokens * references.length
+    const aggCost = estimateTokenCost(aggregator, { inputTokens: aggInput, outputTokens: expectedOutputTokens }, prices)
+    downstreamUsd += aggCost?.costUsd || 0
+  }
+  if (Array.isArray(judgeModels) && judgeModels.length > 0) {
+    for (const jm of judgeModels) {
+      if (jm?.provider && jm?.model) {
+        const jInput = promptTokens + expectedOutputTokens * references.length
+        const jCost = estimateTokenCost(jm, { inputTokens: jInput, outputTokens: expectedOutputTokens }, prices)
+        downstreamUsd += jCost?.costUsd || 0
+      }
+    }
+  }
+
+  // Multiplier if peer critique (Round 2) is enabled: each candidate runs another turn with input from all peers
+  const roundMultiplier = peerCritiqueEnabled ? 2 : 1
+  const fullEstimatedUsd = Number(((totalEstimatedUsd * roundMultiplier) + downstreamUsd).toFixed(6))
 
-  if (totalEstimatedUsd <= maxBudgetUsd) {
+  if (fullEstimatedUsd <= maxBudgetUsd) {
     return {
       allowed: true,
       references,
       action: 'pass',
-      estimatedCostUsd: totalEstimatedUsd,
+      estimatedCostUsd: fullEstimatedUsd,
     }
   }
 
@@ -66,9 +109,9 @@ export function applyBudgetGuardrails({
       allowed: false,
       references: [],
       action: 'abort',
-      estimatedCostUsd: totalEstimatedUsd,
+      estimatedCostUsd: fullEstimatedUsd,
       maxBudgetUsd,
-      reason: `Estimated MoA run cost ($${totalEstimatedUsd.toFixed(4)}) exceeds configured max budget ($${maxBudgetUsd.toFixed(4)}).`,
+      reason: `Estimated MoA run cost ($${fullEstimatedUsd.toFixed(4)}) exceeds configured max budget ($${maxBudgetUsd.toFixed(4)}).`,
     }
   }
 
@@ -78,9 +121,23 @@ export function applyBudgetGuardrails({
   let cumulative = 0
 
   for (const item of sorted) {
-    if (cumulative + item.costUsd <= maxBudgetUsd || kept.length === 0) {
+    const candidateCost = item.costUsd * roundMultiplier
+    if (cumulative + candidateCost <= maxBudgetUsd) {
       kept.push(item.slot)
-      cumulative += item.costUsd
+      cumulative += candidateCost
+    }
+  }
+
+  if (kept.length === 0) {
+    return {
+      allowed: false,
+      references: [],
+      action: 'trim',
+      originalCount: references.length,
+      trimmedCount: 0,
+      estimatedCostUsd: Number((sorted[0]?.costUsd * roundMultiplier || 0).toFixed(6)),
+      maxBudgetUsd,
+      reason: `Cannot fit any candidate model within budget ($${maxBudgetUsd.toFixed(4)}). Cheapest candidate requires $${((sorted[0]?.costUsd || 0) * roundMultiplier).toFixed(4)}.`,
     }
   }
 

package/lib/moa-parser.js

@@ -40,7 +40,7 @@ export function parseRecommendedAssembler(judgeText, defaultIndex = 1, candidate
 /**
  * Parses a `/moa [preset] <prompt>` command invocation.
  */
-export function parseMoACommand(text, presets = []) {
+export function parseMoACommand(text, presets = [], defaultPreset = undefined) {
   if (typeof text !== 'string' || !text.startsWith('/moa')) {
     return null
   }
@@ -59,8 +59,9 @@ export function parseMoACommand(text, presets = []) {
   const remainder = text.slice(4).trim()
   if (!remainder) {
     return {
-      presetName: 'default',
+      presetName: defaultPreset,
       prompt: '',
+      isExplicit: false,
     }
   }
 
@@ -69,6 +70,16 @@ export function parseMoACommand(text, presets = []) {
     return {
       presetName: matchPreset[1],
       prompt: matchPreset[2] || '',
+      isExplicit: true,
+    }
+  }
+
+  const matchShortPreset = /^-p\s+([a-zA-Z0-9_-]+)\s*(.*)/s.exec(remainder)
+  if (matchShortPreset) {
+    return {
+      presetName: matchShortPreset[1],
+      prompt: matchShortPreset[2] || '',
+      isExplicit: true,
     }
   }
 
@@ -80,13 +91,15 @@ export function parseMoACommand(text, presets = []) {
       return {
         presetName: matched.name,
         prompt: parts.slice(1).join(' ').trim(),
+        isExplicit: true,
       }
     }
   }
 
   return {
-    presetName: 'default',
+    presetName: defaultPreset,
     prompt: remainder,
+    isExplicit: false,
   }
 }
 

package/lib/moa-runner.js

@@ -34,25 +34,11 @@ export async function runMoAPipeline({
   liveCanvas = null,
   onStreamDelta = null,
   signal = null,
-  testGateEnabled = null,
-  testCommand = null,
-  execFn = null,
-  webServerPort = null,
-  timeMachineEngine = null,
-  timeMachineUrl = null,
-  force = isTestEnvironment() ? true : false,
-  checkpoint = true,
-  checkpointFn = null,
-  runId: customRunId = null,
+  testGateEnabled = null, testCommand = null, execFn = null,
+  webServerPort = null, timeMachineEngine = null, timeMachineUrl = null,
+  force = isTestEnvironment() ? true : false, checkpoint = true, checkpointFn = null, runId: customRunId = null,
 }) {
-  const promoteOpts = {
-    webServerPort,
-    timeMachineEngine,
-    timeMachineUrl,
-    force: Boolean(force),
-    checkpoint,
-    checkpointFn,
-  }
+  const promoteOpts = { webServerPort, timeMachineEngine, timeMachineUrl, force: Boolean(force), checkpoint, checkpointFn }
   if (signal?.aborted) return { error: new Error('Turn aborted') }
   const startTime = Date.now()
   // 1. Resolve configurations
@@ -81,6 +67,11 @@ export async function runMoAPipeline({
     action: preset?.budget_action || 'trim',
     prices,
     promptLength: userPrompt?.length || 1000,
+    messages,
+    maxTokens,
+    peerCritiqueEnabled: isPeerCritiqueEnabled,
+    aggregator: primaryJudge,
+    judgeModels: preset?.multi_judge_enabled ? preset?.judge_models : [],
   })
   if (!budgetCheck.allowed) {
     return {
@@ -358,7 +349,7 @@ export async function runMoAPipeline({
         }))
       const r2Prompt = buildPeerCritiquePrompt(userPrompt, cand.text, opponents, isBlindEvaluation)
       try {
-        const slot = referenceModels[cand.index - 1] || { provider: cand.provider, model: cand.model }
+        const slot = cand.slot || { provider: cand.provider, model: cand.model }
         const res = await callWithTransientRetry(callLlm, {
           provider: slot.provider,
           model: slot.model,
@@ -523,7 +514,7 @@ export async function runMoAPipeline({
   const durationMs = Date.now() - startTime
   const usageSummary = summarizeMoAUsage(referenceOutputs, aggUsage)
   const winningRef = referenceOutputs[winningIndex - 1]
-  const winnerModel = winningRef?.label || slotLabel(referenceModels[0])
+  const winnerModel = winningRef?.label || slotLabel(effectiveReferences[0] || referenceModels[0])
   const finalAggLabel = slotLabel(chosenJudge)
 
   // Feature 9: Benchmark & Post-Mortem PR report

package/lib/routes.js

@@ -40,7 +40,7 @@ function readBody(req) {
   })
 }
 
-export function registerMoaRoutes(ctx, { live, callLlm, liveCanvas, saveConfig, Config, plainConfig }) {
+export function registerMoaRoutes(ctx, { live, callLlm, liveCanvas, saveConfig, Config, plainConfig, webServerPort, timeMachineEngine }) {
 
 ctx.effect(() => {
     return ctx.webServer.register({
@@ -220,7 +220,12 @@ ctx.effect(() => {
 
             // Persist via saveConfig (in-memory and DSH profile settings)
             if (typeof saveConfig === 'function') {
-              await saveConfig(validated)
+              try {
+                await saveConfig(validated)
+              } catch (perr) {
+                writeJson(res, 500, { ok: false, error: 'Settings persistence failed: ' + (perr?.message || String(perr)) })
+                return
+              }
             }
 
             const unwrapped = typeof plainConfig === 'function' ? plainConfig(validated) : validated
@@ -563,13 +568,12 @@ ctx.effect(() => {
           const preset = (cfg.presets || []).find((p) => p.name === presetName) || cfg.presets?.[0]
 
           const result = await runMoAPipeline({
-            userPrompt: prompt,
-            messages: payload.messages || [],
-            preset,
-            callLlm,
-            cwd: payload.cwd || process.cwd(),
-            prices: cfg.prices || {},
-            liveCanvas,
+            userPrompt: prompt, messages: payload.messages || [], preset, callLlm,
+            cwd: payload.cwd || process.cwd(), prices: cfg.prices || {}, liveCanvas,
+            webServerPort: webServerPort || ctx.webServer?.port,
+            timeMachineEngine: timeMachineEngine || ctx.get?.('timeMachineEngine') || ctx.timeMachineEngine,
+            timeMachineUrl: payload.timeMachineUrl, force: payload.force,
+            checkpoint: payload.checkpoint, checkpointFn: payload.checkpointFn,
           })
 
           if (result?.skippedFiles > 0) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@goodandready/dsh-moa",
-  "version": "0.2.38",
+  "version": "0.2.39",
   "description": "Mixture of Agents (MoA) plugin for DeepSeek Harness with /moa slash command",
   "type": "module",
   "main": "lib/index.js",