mio-agent-runtime 0.5.22 → 0.5.23

package/README.md

@@ -10,7 +10,7 @@ Mio Agent Runtime 的可发布 CLI 包。
 npm install -g mio-agent-runtime
 ```
 
-> 新机器 / 升级：请安装 `>=0.3.6`（Codex + OpenCode + Hermes 被动观察）。
+> 新机器 / 升级：请安装 `>=0.5.22`（5 host 被动观察 + trace 查询 + digest 价值管线）。
 > 旧版本只有主动 MCP 规则注入，没有「无感存/取」。安装后执行一次
 > `mio init && mio install <host>`。
 
@@ -43,12 +43,12 @@ mio --json evolution status Machine-readable evolution module health
 
 ## MCP Tools
 
-MCP server exposes 47 tools across 5 domains:
+MCP server exposes 46 tools across 5 domains:
 
-### Memory (10)
+### Memory (9)
 `mio.memory.query` · `mio.memory.record` · `mio.memory.archive` · `mio.memory.migrate` · `mio.memory.analyze` · `mio.experience.list` · `mio.experience.confirm` · `mio.experience.reuse` · `mio.policy.check`
 
-### Observer Pipeline (13)
+### Observer Pipeline (14)
 `mio.observer.world_model` · `mio.observer.trends` · `mio.observer.research` · `mio.observer.insights` · `mio.observer.status` · `mio.observer.collect` · `mio.observer.ferment` · `mio.observer.essays` · `mio.observer.dag` · `mio.observer.ingest` · `mio.observer.subscribe` · `mio.observer.digest`
 `mio.trace.query` — query the trace log (task outcomes, tool errors) by type/outcome/agent/project/time window; also available as `mio traces` on the CLI
 `mio.digest.generate` — aggregate recent data into an actionable digest (agent success rates, project activity, error hotspots, reuse evidence, suggestions); also available as `mio digest` on the CLI
@@ -59,8 +59,8 @@ MCP server exposes 47 tools across 5 domains:
 ### Creativity Engine (4)
 `mio.creativity.generate` · `mio.creativity.ferment` · `mio.creativity.list` · `mio.creativity.status`
 
-### Idea & Evolution (7)
-`mio.idea.generate` · `mio.idea.list` · `mio.evolution.suggest` · `mio.evolution.report` · `mio.evolution.strategies` · `mio.phase0.report`
+### Task, Agent & Evolution (15)
+`mio.task.route` · `mio.task.record_outcome` · `mio.agent.register` · `mio.agent.list` · `mio.agent.report` · `mio.host.capabilities` · `mio.evolution.status` · `mio.evolution.report` · `mio.evolution.shadow.record` · `mio.evolution.dual_write.record` · `mio.evolution.cutover.readiness` · `mio.evolution.cutover.apply` · `mio.evolution.migration.plan` · `mio.evolution.authority.plan` · `mio.phase0.report`
 
 ## Packages
 
@@ -77,6 +77,11 @@ Published to npm under `@akemi-mio` scope:
 | `@akemi-mio/evolution-strategy` | Evolution strategy engine |
 | `@akemi-mio/evolution-safety` | Safety guardrails |
 | `@akemi-mio/evolution-scheduler` | Evolution scheduler |
+| `@akemi-mio/analysis` | ModuleScanner — static analysis of package source trees (zero deps) |
+| `@akemi-mio/messaging` | ExternalMessageGateway — inbound message routing + notification dispatch via injected renderer/outbox (zero deps; telegram adapters stay host-side) |
+| `@akemi-mio/reasoning` | Reasoning planner — pure-function scoring for reasoning directives (injectable logger) |
+| `@akemi-mio/resource-control` | Resource budgets, background task runner, budget rebalancing (injectable runtime) |
+| `@akemi-mio/agent-persona` | Persona drift control, content classification, user behavior analysis (injectable logger) |
 | `@akemi-mio/creativity` | CreativityService — LLM-powered concept generation, conflict detection, merging, fermentation (TypeScript, depends on `@akemi-mio/core`) |
 | `@akemi-mio/observer` | ObserverService — multi-source data collection, trend analysis, deep research, world model, fermentation, DAG state machine, self-evolution engine (zero npm deps) |
 | `@akemi-mio/insight` | InsightService — LLM-based insight generation, conflict/drift/repetition/stalled-goal/friction detectors, presence service, insight scoring (zero npm deps) |
@@ -132,6 +137,21 @@ schema，`source=workbuddy-observer`）。
 - 已处理的文件偏移记录在 `MIO_HOME/observe-state.json`，不会重复写入。
 - 只回填最近 24h 内更新的会话，避免把历史记录灌入记忆库。
 
+### Context 块的写入范围（`MIO_CONTEXT`）
+
+`<workspace>/AGENTS.md`（或 `CLAUDE.md`）里的 `MIO_CONTEXT` 块最多保留
+6 条（`CONTEXT_MAX`），按时间倒序。两条规则保证真实历史不被挤掉：
+
+- **同一条内容只占一个位置**。重复写入只把它移到最前。旧版本会把同一行
+  反复插入，占满全部 6 个位置，把真实任务摘要整体挤出（存量
+  `observe-state.json` 里的重复条目会在下次写入时自动清理）。
+- **`mio digest --write-back` 只写回有数据的项目**。写回的是一行项目维度
+  摘要（`digest(7d): <项目> N 任务 M% 成功`），不是全局统计；`observe-state`
+  里没有对应 digest 数据的 workspace 会被直接跳过，不会被写入。旧版本把
+  同一条全局 headline 广播给所有历史项目，导致无关仓库（如 ComfyUI）被污染。
+
+需要强制指定目标时用 `--cwd <path>` 或 `--project <name>`；`--cwd` 优先。
+
 OpenCode 会话不依赖模型自觉：`mio observe` 通过 `opencode db` 增量读取
 `~/.local/share/opencode/opencode.db`（Windows 下该文件被 OpenCode 独占锁定，
 必须经 opencode 自带 CLI 读取），按 `part` 表的 rowid 做增量游标，只传输
@@ -169,6 +189,8 @@ packages/mio-cli
 │   ├── runtime-modules.js
 │   ├── creativity-engine.js
 │   ├── memory-store.js
+│   ├── retention.js
+│   ├── digest.js
 │   └── mio-intelligence-mcp/index.js
 └── package.json
 ```
@@ -178,7 +200,10 @@ query/record ranking (Latin token + CJK bigram scoring, project/global
 scope layers, reuse-evidence weighting). Both the MCP server
 (`mio.memory.query` / `mio.memory.record`) and the CLI
 (`mio recall` / `mio remember`) go through it, so rankings are identical
-across entry points.
+across entry points. `server/digest.js` aggregates traces/memory/reuse
+into actionable reports (also exposed as `mio.digest.generate`), and
+`server/retention.js` powers `mio prune` (age/expiry trimming with
+backups; `memory.jsonl` is never touched without explicit `--memory --yes`).
 
 ## Release verification
 

package/bin/mio.js

@@ -348,6 +348,13 @@ function digestCommand(args, useJson) {
     // Feed the actionable lines back into each active workspace context file
     // (AGENTS.md / CLAUDE.md MIO_CONTEXT block), so every agent passively
     // receives the digest's value on its next session.
+    //
+    // Targets are resolved per project, never broadcast. An earlier version
+    // wrote one global headline into Object.keys(state.contexts) -- every
+    // workspace ever observed -- so unrelated repos (ComfyUI, douyin_store)
+    // received the same line and their real history was pushed out of the
+    // six-slot block. Write only where the digest actually has data for that
+    // project, or where the caller explicitly asked with --cwd/--project.
     const stateFile = path.join(MIO_HOME, 'observe-state.json')
     let state = {}
     try {
@@ -355,27 +362,50 @@ function digestCommand(args, useJson) {
     } catch (_) {}
     state.contexts = state.contexts || {}
     const stamp = new Date().toLocaleString('zh-CN', { hour12: false })
-    const rate =
-      report.overview.tasks > 0
-        ? Math.round(((report.overview.byOutcome.success || 0) / report.overview.tasks) * 100)
-        : 0
-    const headline =
-      `digest(${days}d): 任务 ${report.overview.tasks} 条 成功率 ${rate}%, error ${report.overview.errorTraces}` +
-      (report.suggestions.length > 0 ? ' | ' + report.suggestions[0] : '')
-    const targetCwds = optionValue(flags, '--cwd')
-      ? [optionValue(flags, '--cwd')]
-      : Object.keys(state.contexts)
+
+    // Project name -> per-project digest line.
+    const projectLine = new Map()
+    for (const entry of report.projects || []) {
+      if (!entry || !entry.project) continue
+      const rate = entry.tasks > 0 ? Math.round(((entry.success || 0) / entry.tasks) * 100) : 0
+      projectLine.set(String(entry.project).toLowerCase(), `digest(${days}d): ${entry.project} ${entry.tasks} 任务 ${rate}% 成功`)
+    }
+
+    const explicitCwd = optionValue(flags, '--cwd')
+    const explicitProject = optionValue(flags, '--project')
+    let targetCwds = []
+    if (explicitCwd) {
+      targetCwds = [explicitCwd]
+    } else if (explicitProject) {
+      const wanted = String(explicitProject).toLowerCase()
+      targetCwds = Object.keys(state.contexts).filter(
+        (cwd) => path.basename(cwd).toLowerCase() === wanted,
+      )
+    } else {
+      // No explicit target: only write to workspaces this digest has data for.
+      // A workspace with no matching project gets nothing.
+      targetCwds = Object.keys(state.contexts).filter((cwd) =>
+        projectLine.has(path.basename(cwd).toLowerCase()),
+      )
+    }
+
     let written = 0
     for (const cwd of targetCwds) {
+      const line = projectLine.get(path.basename(cwd).toLowerCase())
+      if (!line) continue
       try {
-        observer.updateProjectContext(state, cwd, stamp + ' | ' + headline, 'CLAUDE.md')
-        observer.updateProjectContext(state, cwd, stamp + ' | ' + headline, 'AGENTS.md')
+        observer.updateProjectContext(state, cwd, stamp + ' | ' + line, 'CLAUDE.md')
+        observer.updateProjectContext(state, cwd, stamp + ' | ' + line, 'AGENTS.md')
         written += 1
       } catch (_) {}
     }
     fs.mkdirSync(MIO_HOME, { recursive: true })
     fs.writeFileSync(stateFile, JSON.stringify(state) + '\n', 'utf8')
-    report.writeBack = { workspaces: written, headline }
+    report.writeBack = {
+      workspaces: written,
+      skipped: Object.keys(state.contexts).length - written,
+      projects: [...projectLine.values()],
+    }
   }
 
   // Persist the report so it becomes part of the data asset itself.
@@ -388,7 +418,10 @@ function digestCommand(args, useJson) {
   console.log(report.markdown)
   console.log(`Report saved: ${reportFile}`)
   if (report.writeBack) {
-    console.log(`Write-back: ${report.writeBack.workspaces} workspace(s) <- "${report.writeBack.headline}"`)
+    console.log(
+      `Write-back: ${report.writeBack.workspaces} workspace(s), ${report.writeBack.skipped} skipped (no digest data)`,
+    )
+    for (const line of report.writeBack.projects) console.log(`  - ${line}`)
   }
 }
 

package/observe/observer.js

@@ -65,6 +65,30 @@ function stateFile(home) {
   return path.join(home, 'observe-state.json')
 }
 
+// Drop repeated lines from stored context lists, preserving the most recent
+// occurrence order. Older builds could write the same line into every slot of
+// a workspace block (a generated digest headline broadcast to all projects),
+// so existing observe-state files still carry duplicates until they are
+// rewritten. Normalising on load heals that data without a migration step.
+function dedupeContexts(contexts) {
+  const out = {}
+  for (const [cwd, list] of Object.entries(contexts)) {
+    if (!Array.isArray(list)) {
+      out[cwd] = []
+      continue
+    }
+    const seen = new Set()
+    const deduped = []
+    for (const line of list) {
+      if (seen.has(line)) continue
+      seen.add(line)
+      deduped.push(line)
+    }
+    out[cwd] = deduped
+  }
+  return out
+}
+
 function loadState(home) {
   try {
     const parsed = JSON.parse(fs.readFileSync(stateFile(home), 'utf8'))
@@ -74,7 +98,7 @@ function loadState(home) {
         files: parsed.files || {},
         buffers: parsed.buffers || {},
         openTurns: parsed.openTurns || {},
-        contexts: parsed.contexts || {},
+        contexts: dedupeContexts(parsed.contexts || {}),
         codexFiles: parsed.codexFiles || {},
         codexBuffers: parsed.codexBuffers || {},
         codexTurns: parsed.codexTurns || {},
@@ -291,7 +315,20 @@ const CONTEXT_MAX = 6
 function updateProjectContext(state, cwd, line, targetFile) {
   if (!cwd || !line) return
   const target = path.join(cwd, targetFile || 'AGENTS.md')
-  const list = state.contexts[cwd] || []
+  // Normalise the whole list before inserting, not just the incoming line.
+  // Callers may load state without going through loadState (the digest
+  // write-back reads the file directly), so stale duplicates have to be
+  // cleared here or they keep consuming slots and push real history out.
+  const list = []
+  const seen = new Set()
+  for (const item of state.contexts[cwd] || []) {
+    if (seen.has(item)) continue
+    seen.add(item)
+    list.push(item)
+  }
+  // Re-inserting an existing line moves it to the front instead of duplicating.
+  const existing = list.indexOf(line)
+  if (existing !== -1) list.splice(existing, 1)
   list.unshift(line)
   if (list.length > CONTEXT_MAX) list.pop()
   state.contexts[cwd] = list
@@ -1591,6 +1628,7 @@ module.exports = {
   workbuddyConfigDir,
   loadState,
   saveState,
+  dedupeContexts,
   updateProjectContext,
   runCycle,
   opencodeDbPath,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "mio-agent-runtime",
-  "version": "0.5.22",
+  "version": "0.5.23",
   "description": "Mio Agent Runtime installer and MCP intelligence layer",
   "license": "MIT",
   "type": "commonjs",
@@ -18,6 +18,7 @@
   ],
   "scripts": {
     "check": "node --check bin/mio.js && node --check adapters/host-utils.js && node --check adapters/codex.js && node --check adapters/opencode.js && node --check adapters/workbuddy.js && node --check adapters/hermes.js && node --check adapters/claude-code.js && node --check server/runtime-modules.js && node --check server/memory-store.js && node --check server/retention.js && node --check server/digest.js && node --check server/host-capabilities.js && node --check server/evolution-cutover.js && node --check server/mio-intelligence-mcp/index.js && node --check server/mio-intelligence-mcp/phase0.js && node --check observe/observer.js && node --check scripts/verify-packed-runtime.js",
+    "test": "node --test \"__tests__/*.test.js\" \"server/mio-intelligence-mcp/__tests__/*.test.js\"",
     "prepack": "npm run check",
     "verify:pack-install": "node scripts/verify-packed-runtime.js"
   },
@@ -34,5 +35,11 @@
     "@akemi-mio/evolution-scheduler": "^0.1.0",
     "@akemi-mio/observer": "^0.1.0",
     "@akemi-mio/insight": "^0.1.0"
-  }
+  },
+  "main": "index.js",
+  "directories": {
+    "doc": "docs"
+  },
+  "keywords": [],
+  "author": ""
 }

package/scripts/verify-packed-runtime.js

@@ -76,12 +76,13 @@ function npmCommand(platform = process.platform) {
 
 function npmInvocation(args, env = process.env, platform = process.platform, node = process.execPath) {
   if (env.npm_execpath) {
-    return { command: node, args: [env.npm_execpath, ...args] }
+    return { command: node, args: [env.npm_execpath, ...args], shell: false }
   }
-  if (platform === 'win32') {
-    return { command: 'cmd.exe', args: ['/d', '/s', '/c', 'npm', ...args] }
-  }
-  return { command: npmCommand(platform), args }
+  // Fall back to the npm shim on PATH instead of routing through cmd.exe.
+  // cmd.exe is unavailable in some sandboxes, and a bare .cmd/.bat shim cannot
+  // be spawned directly (Node requires shell:true for those on Windows), so
+  // signal the shell requirement to the runner instead of guessing there.
+  return { command: npmCommand(platform), args: [...args], shell: platform === 'win32' }
 }
 
 function packageDirName(name) {
@@ -95,9 +96,12 @@ function tarballFileName(name, version) {
 
 function readPackageJson(workspaceRoot, name) {
   const packageDir = path.join(workspaceRoot, 'packages', packageDirName(name))
+  // Windows editors may write package.json with a UTF-8 BOM; JSON.parse
+  // rejects the leading U+FEFF, so strip it before parsing.
+  const raw = fs.readFileSync(path.join(packageDir, 'package.json'), 'utf8').replace(/^\uFEFF/, '')
   return {
     packageDir,
-    manifest: JSON.parse(fs.readFileSync(path.join(packageDir, 'package.json'), 'utf8')),
+    manifest: JSON.parse(raw),
   }
 }
 
@@ -211,6 +215,11 @@ function run(command, args, options = {}) {
     env: options.env || process.env,
     encoding: 'utf8',
     stdio: options.stdio || ['ignore', 'pipe', 'pipe'],
+    // Windows .cmd/.bat shims (npm.cmd) cannot be spawned directly by
+    // execFileSync since Node's CVE-2024-27980 hardening; they need a shell.
+    // This only kicks in for the npm shim path — node-based invocations pass
+    // an absolute node executable and still run shell-free.
+    shell: options.shell === true,
   })
   return output ? output.trim() : ''
 }
@@ -296,14 +305,19 @@ function verifyPackedRuntimeInstall(input = {}) {
     run(pack.command, pack.args, {
       cwd: pkg.packageDir,
       stdio: ['ignore', 'pipe', 'pipe'],
+      shell: pack.shell,
     })
     packageArtifactEvidence(pkg)
   }
 
   const init = npmInvocation(['init', '-y'])
-  run(init.command, init.args, { cwd: plan.installDir, stdio: ['ignore', 'pipe', 'pipe'] })
+  run(init.command, init.args, { cwd: plan.installDir, stdio: ['ignore', 'pipe', 'pipe'], shell: init.shell })
   const install = npmInvocation(plan.install.args)
-  run(install.command, install.args, { cwd: plan.install.cwd, stdio: ['ignore', 'pipe', 'pipe'] })
+  run(install.command, install.args, {
+    cwd: plan.install.cwd,
+    stdio: ['ignore', 'pipe', 'pipe'],
+    shell: install.shell,
+  })
   const smokeEnv = {
     ...process.env,
     MIO_HOME: path.join(plan.installDir, '.mio-home'),
@@ -360,6 +374,7 @@ module.exports = {
   createVerifyPackedRuntimeOptions,
   createPackedRuntimeInstallPlan,
   packageArtifactEvidence,
+  readPackageJson,
   npmInvocation,
   npmCommand,
   parseVerifyPackedRuntimeArgs,

package/server/digest.js

@@ -186,6 +186,18 @@ function createDigest(options = {}) {
       suggestions,
     }
     report.markdown = renderMarkdown(report)
+
+    // Persist a machine-readable snapshot for consumers that should not
+    // recompute the aggregation — notably mio.task.route's agent-health signal.
+    try {
+      const digestDir = path.join(home, 'digest')
+      fs.mkdirSync(digestDir, { recursive: true })
+      const snapshot = { ...report }
+      delete snapshot.markdown
+      fs.writeFileSync(path.join(digestDir, 'latest.json'), JSON.stringify(snapshot) + '\n', 'utf8')
+    } catch (_) {
+      // best-effort persistence; the in-memory report is still returned
+    }
     return report
   }
 

package/server/mio-intelligence-mcp/index.js

@@ -1108,6 +1108,40 @@ function reportAgent(args = {}) {
   return { project, count: reports.length, reports }
 }
 
+const AGENT_HEALTH_FRESH_MS = 48 * 3600000
+
+function readAgentHealth() {
+  const file = path.join(dataDir, 'digest', 'latest.json')
+  if (!fs.existsSync(file)) return null
+  try {
+    const snapshot = JSON.parse(fs.readFileSync(file, 'utf8'))
+    const ts = Date.parse(snapshot.generatedAt || '')
+    const generatedAt = Number.isFinite(ts) ? snapshot.generatedAt : null
+    if (!Number.isFinite(ts) || Date.now() - ts > AGENT_HEALTH_FRESH_MS) {
+      return { stale: true, generatedAt, best: null, degraded: [], agents: [] }
+    }
+    const agents = Array.isArray(snapshot.agents) ? snapshot.agents : []
+    const reliable = agents.filter((a) => Number(a.total) >= 5)
+    const best =
+      reliable.length > 0
+        ? reliable.reduce((a, b) => (b.successRate > a.successRate ? b : a))
+        : null
+    const degraded = reliable
+      .filter((a) => a.successRate < 50)
+      .map((a) => ({ agent: a.agent, successRate: a.successRate, total: a.total }))
+    return {
+      stale: false,
+      generatedAt,
+      periodDays: snapshot.periodDays || null,
+      best: best ? { agent: best.agent, successRate: best.successRate, total: best.total } : null,
+      degraded,
+      agents: agents.map((a) => ({ agent: a.agent, total: a.total, successRate: a.successRate })),
+    }
+  } catch (_) {
+    return null
+  }
+}
+
 function routeTask(args = {}) {
   const task = String(args.task || '').trim()
   if (!task) throw new Error('task.route requires a non-empty task')
@@ -1223,6 +1257,26 @@ function routeTask(args = {}) {
       'No matching experience or memory found. Proceed with care and record the outcome to grow the experience base.'
   }
 
+  // Agent-health routing signal from the latest digest snapshot (value loop:
+  // collected task outcomes influence who should execute the next task).
+  const agentHealth = readAgentHealth()
+  let routingSignal = null
+  if (agentHealth && !agentHealth.stale && agentHealth.best) {
+    const parts = []
+    if (agentHealth.best.successRate >= 80) {
+      parts.push(
+        `prefer ${agentHealth.best.agent} (${agentHealth.best.successRate}% success over ${agentHealth.best.total} tasks)`,
+      )
+    }
+    for (const bad of agentHealth.degraded) {
+      parts.push(`avoid ${bad.agent} (${bad.successRate}% success over ${bad.total} tasks)`)
+    }
+    if (parts.length > 0) {
+      routingSignal = `Agent health (last ${agentHealth.periodDays || '?'}d): ${parts.join('; ')}.`
+      suggestion = `${suggestion} ${routingSignal}`
+    }
+  }
+
   return {
     task,
     project,
@@ -1230,9 +1284,11 @@ function routeTask(args = {}) {
     count: routes.length,
     routes,
     relatedMemories,
+    agentHealth,
     summary: {
       verifiedRoutes: routes.length,
       relatedMemories: relatedMemories.length,
+      routingSignal,
       suggestion,
     },
   }
@@ -1468,7 +1524,7 @@ const TOOLS = [
   },
   {
     name: 'mio.task.route',
-    description: 'Route an incoming task to the most relevant verified experiences and related memories, turning recall into decision influence (P1).',
+    description: 'Route an incoming task to the most relevant verified experiences and related memories, and include an agent-health signal (from the latest digest) saying which agents to prefer or avoid. Turns recall plus observed performance into decision influence (P1).',
     inputSchema: {
       type: 'object',
       properties: {