mio-agent-runtime 0.5.22 → 0.6.0

package/bin/mio.js

@@ -21,8 +21,21 @@ const observer = require('../observe/observer.js')
 const { getEvolutionStatus, formatEvolutionStatusText } = require('../server/runtime-modules.js')
 const { createEvolutionCutoverTools } = require('../server/evolution-cutover.js')
 const { createMemoryStore } = require('../server/memory-store.js')
+const { createExperienceStore } = require('../server/experience-store.js')
+const { createPolicyStore } = require('../server/policy-store.js')
+const { CreativityEngine } = require('../server/creativity-engine.js')
+const { createAgentStore } = require('../server/agent-store.js')
+const { createInsightStore } = require('../server/insight-store.js')
+const { createObserverStore } = require('../server/observer-store.js')
+const { loadPhase0, renderPhase0Markdown } = require('../server/mio-intelligence-mcp/phase0.js')
+const { listHostCapabilities } = require('../server/host-capabilities.js')
+const { createTaskStore } = require('../server/task-store.js')
+const { chatJson, llmConfig, isLlmConfigured } = require('../server/llm-client.js')
+const { createQueryLog } = require('../server/query-log.js')
+const { createSubscriptionStore, subscribeKey } = require('../server/subscription-store.js')
 const { createRetention } = require('../server/retention.js')
 const { createDigest } = require('../server/digest.js')
+const { createEvolutionReport, formatEvolutionReportText } = require('../server/evolution-report.js')
 
 function appendJsonl(file, value) {
   fs.mkdirSync(path.dirname(file), { recursive: true })
@@ -96,6 +109,39 @@ function parseNumberOption(args, name) {
   return value
 }
 
+// mio evolution report shares one implementation with mio.evolution.report
+// (see ../server/evolution-report.js), so CLI and MCP cannot report different numbers.
+function cliEvolutionReport() {
+  return createEvolutionReport({ dataDir: MIO_HOME, projectName })
+}
+
+const EVOLUTION_REPORT_PERIODS = new Set(['24h', '7d', '30d', 'all'])
+
+function evolutionReportCommand(args, useJson) {
+  const flags = args.slice(2)
+  const period = optionValue(flags, '--period')
+  if (period !== undefined && !EVOLUTION_REPORT_PERIODS.has(period)) {
+    console.error(`--period must be one of: ${[...EVOLUTION_REPORT_PERIODS].join(', ')}`)
+    process.exitCode = 1
+    return
+  }
+
+  let result
+  try {
+    result = cliEvolutionReport().report({
+      project: optionValue(flags, '--project'),
+      period,
+    })
+  } catch (error) {
+    console.error(error.message || error)
+    process.exitCode = 1
+    return
+  }
+
+  if (useJson) return jsonOrText(result, true)
+  console.log(formatEvolutionReportText(result))
+}
+
 function evolutionStatus(useJson) {
   const payload = getEvolutionStatus()
   if (useJson) {
@@ -148,6 +194,7 @@ function printDualWriteRecord(record, useJson) {
 
 async function evolutionCommand(args, useJson) {
   if (args[1] === 'status') return evolutionStatus(useJson)
+  if (args[1] === 'report') return evolutionReportCommand(args, useJson)
   try {
     if (args[1] === 'shadow' && args[2] === 'record') {
       return printShadowRecord(
@@ -218,7 +265,7 @@ async function evolutionCommand(args, useJson) {
     process.exitCode = 1
     return
   }
-  console.error('Usage: mio evolution status|shadow record|dual-write record|cutover readiness|cutover apply|authority plan|migration plan')
+  console.error('Usage: mio evolution status|report|shadow record|dual-write record|cutover readiness|cutover apply|authority plan|migration plan')
   process.exitCode = 1
 }
 
@@ -238,6 +285,54 @@ function cliMemoryStore() {
   })
 }
 
+// policy.check deliberately reads the *global* MIO_HOME rather than a
+// per-project directory, so its history matches `mio recall` and `mio traces`.
+// It shares the memory store's tokenizer and scoring, so risk evidence and
+// related-memory ranking line up with `mio.memory.query`.
+function cliPolicyStore() {
+  const memoryStore = cliMemoryStore()
+  return createPolicyStore({
+    dataDir: MIO_HOME,
+    projectName,
+    memoryStore,
+  })
+}
+
+// The creativity engine is the single implementation behind mio.creativity.*;
+// the CLI points it at the global MIO_HOME so a terminal `mio creativity list`
+// sees the same hypotheses the MCP server would. generate/ferment need an LLM,
+// so they use the shared client -- configured with LLM_API_URL / LLM_KEY /
+// LLM_CHAT_MODEL, the same environment the MCP server reads. A local Ollama
+// works: LLM_API_URL=http://localhost:11434/v1/chat/completions
+function cliCreativityEngine() {
+  const creativityDir = path.join(MIO_HOME, 'creativity')
+  return new CreativityEngine(creativityDir, chatJson)
+}
+
+// Observed-agent telemetry (agents.jsonl + traces/memory/reuse cross-reference)
+// shares one store with the MCP server, so CLI and MCP report identical agents.
+function cliAgentStore() {
+  return createAgentStore({ dataDir: MIO_HOME, projectName })
+}
+
+// Task routing shares one store with mio.task.route. persistQuery stays false:
+// the MCP call feeds the query log for auto-claim, but a terminal inspection
+// must not write to queries.jsonl every time it runs.
+function cliTaskStore() {
+  // Shares one query-log instance with nothing else here, but created from the
+  // same module the MCP server uses, so queries.jsonl has a single reader/writer
+  // implementation. cliMemoryStore deliberately does NOT get one: a terminal
+  // recall is never followed by a task_outcome, so recording those queries
+  // would only add noise.
+  return createTaskStore({
+    dataDir: MIO_HOME,
+    projectName,
+    memoryStore: cliMemoryStore(),
+    queryLog: createQueryLog({ dataDir: MIO_HOME }),
+    agentId: () => 'cli',
+  })
+}
+
 function splitTagsOption(args, name) {
   const raw = optionValue(args, name)
   if (raw === undefined) return []
@@ -348,6 +443,13 @@ 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 +457,50 @@ 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 +513,10 @@ 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}`)
   }
 }
 
@@ -451,102 +579,1763 @@ function rememberCommand(args, useJson) {
   console.log(`Recorded ${record.id} (kind=${record.kind}, project=${record.project || 'global'}, scope=${record.scope})`)
 }
 
-function readConfig() {
-  const fallback = { version: 1, home: MIO_HOME, agents: {} }
-  if (!fs.existsSync(CONFIG_FILE)) return fallback
-  try {
-    return { ...fallback, ...JSON.parse(fs.readFileSync(CONFIG_FILE, 'utf8')) }
-  } catch (_) {
-    return fallback
+// Every flag that consumes the next token. A positional directly after one of
+// these is that flag's value, not an id.
+const VALUE_FLAGS = new Set([
+  '--ids', '--reason', '--project', '--limit', '--scope', '--keep',
+  '--by', '--notes', '--source-agent', '--target-agent', '--experience-id', '--status',
+])
+
+// Accept ids both as `--ids a,b,c` and as bare positionals
+// (`mio memory archive mem_a mem_b`).
+function collectIds(flags) {
+  const ids = []
+  for (let i = 0; i < flags.length; i += 1) {
+    const token = flags[i]
+    if (token.startsWith('--')) continue
+    const previous = flags[i - 1]
+    if (previous && VALUE_FLAGS.has(previous)) continue
+    // Ids never contain commas, so a comma-separated positional is unambiguous.
+    for (const id of token.split(',')) {
+      const trimmed = id.trim()
+      if (trimmed) ids.push(trimmed)
+    }
+  }
+  const raw = optionValue(flags, '--ids')
+  if (raw !== undefined) {
+    for (const id of raw.split(',')) {
+      const trimmed = id.trim()
+      if (trimmed) ids.push(trimmed)
+    }
   }
+  return [...new Set(ids)]
 }
 
-function writeConfig(config) {
-  ensureHome()
-  fs.writeFileSync(CONFIG_FILE, `${JSON.stringify(config, null, 2)}\n`, 'utf8')
+function printMemoryAnalyze(result) {
+  const project = result.project || 'all'
+  console.log(
+    `Memory analyze: ${result.total} active, ${result.archived} archived (project=${project})`,
+  )
+  console.log(`layers: project=${result.layers.project} global=${result.layers.global}`)
+  const kinds = Object.entries(result.byKind)
+  if (kinds.length > 0) {
+    console.log(`by kind: ${kinds.map(([kind, count]) => `${kind}=${count}`).join(', ')}`)
+  }
+
+  if (result.duplicates.length > 0) {
+    console.log(`\nDuplicate groups: ${result.duplicates.length}`)
+    for (const group of result.duplicates) {
+      console.log(`- ${group.size} record(s): ${group.records.map((record) => record.id).join(', ')}`)
+      const preview = String(group.records[0].content || '')
+      if (preview) {
+        console.log(`  ${preview.length > 80 ? `${preview.slice(0, 80)}...` : preview}`)
+      }
+    }
+  }
+
+  if (result.lowQuality.length > 0) {
+    console.log(`\nLow quality: ${result.issues.lowQuality} record(s)`)
+    for (const record of result.lowQuality) {
+      console.log(`- ${record.id} [${record.kind || 'note'}] ${record.issues.join(', ')}`)
+    }
+  }
+
+  if (result.issues.duplicatesSkipped) {
+    console.log('\nDuplicate scan skipped: too many active records to compare pairwise.')
+  }
+
+  if (result.suggestions.length > 0) {
+    console.log('')
+    for (const suggestion of result.suggestions) console.log(`suggestion: ${suggestion}`)
+  }
 }
 
-function jsonOrText(payload, useJson) {
-  if (useJson) {
-    console.log(JSON.stringify(payload, null, 2))
+// Merging divergent content is refused by default; when that happens the
+// caller needs to see *what* differs before deciding, so print the candidates.
+function printMemoryMerge(result) {
+  if (!result.merged) {
+    console.log(`No merge applied (project=${result.project || 'current'}): ${result.reason}`)
+    if (result.divergent.length > 0) {
+      console.log('\nDivergent content — inspect before choosing a survivor:')
+      for (const record of result.divergent) {
+        console.log(`- ${record.id} (${record.contentLength} chars, ${record.timestamp || 'no timestamp'})`)
+        console.log(`    ${record.content.split('\n')[0].slice(0, 110)}`)
+      }
+    }
+    if (result.notFound.length > 0) console.log(`\nnot found: ${result.notFound.join(', ')}`)
+    if (result.outOfScope && result.outOfScope.length > 0) {
+      console.log(`\nbelongs to another project: ${result.outOfScope.join(', ')}`)
+    }
+    if (result.hint) console.log(`\n${result.hint}`)
+    process.exitCode = 1
     return
   }
-  return payload
+
+  // A merge that archived nothing is a no-op: every requested id was already
+  // archived (i.e. these were merged before). Report it as failure the same way
+  // archive does, so scripts do not read success out of an idempotent re-run.
+  if (result.archivedCount === 0) {
+    console.log(`No merge applied (project=${result.project || 'current'}): nothing left to merge`)
+    if (result.skipped.length > 0) {
+      console.log(`already archived (previously merged): ${result.skipped.join(', ')}`)
+      console.log(`superseded by: ${result.survivor.id}`)
+    }
+    if (result.notFound.length > 0) console.log(`not found: ${result.notFound.join(', ')}`)
+    process.exitCode = 1
+    return
+  }
+
+  console.log(
+    `Merged ${result.archivedCount} record(s) into ${result.survivor.id} (project=${result.project || 'current'})`,
+  )
+  console.log(`survivor: ${result.survivor.contentLength} chars, ${result.survivor.timestamp || 'no timestamp'}`)
+  if (result.divergentContent) console.log('note: content differed across members; the survivor body was chosen explicitly')
+  console.log(`supersedes: ${result.survivor.supersedes.join(', ')}`)
+  if (result.skipped.length > 0) console.log(`already archived: ${result.skipped.join(', ')}`)
+  if (result.notFound.length > 0) console.log(`not found: ${result.notFound.join(', ')}`)
+  console.log(`Undo with: mio memory restore --ids ${result.archived.join(',')}`)
 }
 
-function initCommand() {
-  ensureHome()
-  const config = readConfig()
-  if (!config.createdAt) config.createdAt = new Date().toISOString()
-  writeConfig(config)
-  console.log(`MIO_HOME ready: ${MIO_HOME}`)
+function memoryUsage() {
+  console.log(`Usage:
+  mio memory analyze                        Report duplicates, low-quality records and kind histogram
+  mio memory archive --ids a,b              Archive records (soft delete; hidden from recall/analyze)
+  mio memory restore --ids a,b              Un-archive previously archived records
+  mio memory forget --ids a,b               PERMANENTLY delete records (--yes required; writes an audit entry)
+  mio memory merge --ids a,b                Merge duplicates: one survives, the rest are archived
+  mio memory migrate --ids a,b --scope global   Move records between project and global layers
+
+Options:
+  --ids a,b,c        Memory record ids (also accepted as bare positionals)
+  --keep id          Merge: which record's content survives (default: newest)
+  --allow-divergent  Merge: permit merging records whose content differs (requires --keep)
+  --scope global|project   Target layer for migrate
+  --project name     Project filter (defaults to current directory name)
+  --reason text      Optional note stored on the record
+  --limit n          Max duplicate groups / low-quality rows to show (analyze, 1-20)
+  --yes              Apply archive/merge (without it they only preview)
+  --json             Machine-readable output
+`)
 }
 
-function statusCommand(useJson) {
-  const config = readConfig()
-  const codexRestart = adapters.codex.restartWarningMessage()
-  const opencodeRestart = adapters.opencode.restartWarningMessage()
-  const installed = {
-    codex: adapters.codex.isInstalled(),
-    opencode: adapters.opencode.isInstalled(),
-    workbuddy: adapters.workbuddy.isInstalled(),
-    hermes: adapters.hermes.isInstalled(),
-    claude: adapters.claude.isInstalled(),
+function memoryCommand(args, useJson) {
+  const sub = args[1]
+  const flags = args.slice(2)
+
+  if (!sub || sub === 'help' || sub === '--help' || sub === '-h') {
+    memoryUsage()
+    if (!sub) process.exitCode = 1
+    return
   }
-  const payload = {
-    version: config.version,
-    home: MIO_HOME,
-    serverScript: SERVER_SCRIPT,
-    serverScriptExists: fs.existsSync(SERVER_SCRIPT),
-    agents: config.agents || {},
-    codexInstalled: installed.codex,
-    opencodeInstalled: installed.opencode,
-    workbuddyInstalled: installed.workbuddy,
-    hermesInstalled: installed.hermes,
-    claudeInstalled: installed.claude,
-    codexNeedsRestart: Boolean(codexRestart),
-    opencodeNeedsRestart: Boolean(opencodeRestart),
-    codexRestartHint: codexRestart,
-    opencodeRestartHint: opencodeRestart,
-    hermesObserverReady: fs.existsSync(observer.hermesDbPath()),
-    claudeObserverReady: fs.existsSync(observer.claudeProjectsDir()),
-    opencodeObserverReady: fs.existsSync(observer.opencodeDbPath()),
-    observerRunning: observer.isRunning(MIO_HOME),
-    observerPid: observer.isRunning(MIO_HOME) ? observer.readPid(MIO_HOME) : null,
+  if (!['analyze', 'archive', 'restore', 'forget', 'merge', 'migrate'].includes(sub)) {
+    console.error(`Unknown memory subcommand: ${sub}`)
+    memoryUsage()
+    process.exitCode = 1
+    return
   }
-  if (useJson) {
-    jsonOrText(payload, true)
+
+  const store = cliMemoryStore()
+
+  if (sub === 'analyze') {
+    let result
+    try {
+      result = store.analyzeMemory({
+        project: optionValue(flags, '--project'),
+        limit: parseNumberOption(flags, '--limit'),
+      })
+    } catch (error) {
+      console.error(error.message || error)
+      process.exitCode = 1
+      return
+    }
+    if (useJson) return jsonOrText(result, true)
+    return printMemoryAnalyze(result)
+  }
+
+  const ids = collectIds(flags)
+  if (ids.length === 0) {
+    console.error(`mio memory ${sub} requires --ids <id,id,...> (or bare ids)`)
+    process.exitCode = 1
     return
   }
-  console.log(`MIO_HOME: ${MIO_HOME}`)
-  console.log(`MCP Server: ${SERVER_SCRIPT}`)
-  console.log(`Codex MCP: ${installed.codex ? 'installed' : 'not installed'}`)
-  if (codexRestart) console.log(`Codex app: running with an older Mio config - fully restart Codex to load it`)
-  console.log(`OpenCode MCP: ${installed.opencode ? 'installed' : 'not installed'}`)
-  if (opencodeRestart) console.log(`OpenCode Desktop: running with an older Mio config - restart it to load it`)
-  console.log(`OpenCode observer: ${fs.existsSync(observer.opencodeDbPath()) ? 'ready' : 'no database'}`)
-  console.log(`WorkBuddy MCP: ${installed.workbuddy ? 'installed' : 'not installed'}`)
-  console.log(`Hermes MCP: ${installed.hermes ? 'installed' : 'not installed'}`)
-  console.log(`Hermes observer: ${fs.existsSync(observer.hermesDbPath()) ? 'ready' : 'no database'}`)
-  console.log(`Claude Code MCP: ${installed.claude ? 'installed' : 'not installed'}`)
-  console.log(`Claude Code observer: ${fs.existsSync(observer.claudeProjectsDir()) ? 'ready' : 'no transcripts'}`)
-  console.log(`Observer: ${observer.isRunning(MIO_HOME) ? 'running (pid ' + observer.readPid(MIO_HOME) + ')' : 'not running'}`)
+  if (sub === 'merge' && ids.length < 2) {
+    console.error('mio memory merge requires at least 2 ids (nothing to merge otherwise)')
+    process.exitCode = 1
+    return
+  }
+
+  if ((sub === 'archive' || sub === 'merge' || sub === 'forget') && !flagPresent(flags, '--yes')) {
+    // Archiving hides records from recall, merging additionally rewrites the
+    // survivor's supersedes list, and forgetting deletes outright. Preview first
+    // so a mistyped id cannot silently destroy data.
+    const previewProject = optionValue(flags, '--project') || projectName() || 'current'
+    if (useJson) {
+      jsonOrText(
+        {
+          preview: true,
+          applied: false,
+          project: previewProject,
+          ids,
+          keep: optionValue(flags, '--keep') || null,
+          reversible: sub !== 'forget',
+          hint:
+            sub === 'forget'
+              ? 'PERMANENT: re-run with --yes to delete. Consider mio memory archive instead.'
+              : 'Re-run with --yes to apply.',
+        },
+        true,
+      )
+    } else if (sub === 'forget') {
+      // Show *what* is about to disappear, not just the ids: this is the one
+      // memory operation that cannot be undone.
+      const all = readJsonl(cliMemoryStore().memoryPath)
+      const idSet = new Set(ids)
+      const matched = all.filter((r) => idSet.has(r.id))
+      console.log(`Forget preview: ${ids.length} id(s) (project=${previewProject})`)
+      for (const record of matched) {
+        console.log(`  ${record.id}  ${String(record.content || '').replace(/\s+/g, ' ').slice(0, 80)}`)
+      }
+      for (const id of ids) {
+        if (!matched.some((r) => r.id === id)) console.log(`  ${id}  (not found)`)
+      }
+      console.log('')
+      console.log('WARNING: this permanently deletes the record(s) and cannot be undone.')
+      console.log('An audit entry (excerpt only) is written to memory-forget-audit.jsonl.')
+      console.log(`Reversible alternative: mio memory archive --ids ${ids.join(',')} --yes`)
+      console.log('Re-run with --yes to delete permanently.')
+    } else {
+      // Archive keeps its historical "Archive preview" wording; merge is newer
+      // and uses a lowercase verb to read naturally mid-sentence.
+      const label = sub === 'merge' ? 'merge preview' : 'Archive preview'
+      console.log(`${label}: ${ids.length} id(s) (project=${previewProject})`)
+      console.log(`  ${ids.join(', ')}`)
+      if (sub === 'merge') {
+        const keep = optionValue(flags, '--keep')
+        console.log(`  survivor: ${keep || 'newest by timestamp (default)'}`)
+      }
+      console.log(`Re-run with --yes to apply. Undo with: mio memory restore --ids <ids>`)
+    }
+    process.exitCode = 1
+    return
+  }
+
+  let result
+  try {
+    if (sub === 'archive' || sub === 'restore') {
+      result = store.archiveMemory({
+        ids,
+        project: optionValue(flags, '--project'),
+        restore: sub === 'restore',
+        reason: optionValue(flags, '--reason'),
+      })
+    } else if (sub === 'merge') {
+      result = store.mergeMemory({
+        ids,
+        project: optionValue(flags, '--project'),
+        keep: optionValue(flags, '--keep'),
+        allowDivergent: flagPresent(flags, '--allow-divergent'),
+        reason: optionValue(flags, '--reason'),
+      })
+    } else if (sub === 'forget') {
+      result = store.forgetMemory({
+        ids,
+        project: optionValue(flags, '--project'),
+        reason: optionValue(flags, '--reason'),
+        by: 'cli',
+      })
+    } else {
+      result = store.migrateMemory({
+        ids,
+        scope: optionValue(flags, '--scope'),
+        project: optionValue(flags, '--project'),
+        reason: optionValue(flags, '--reason'),
+      })
+    }
+  } catch (error) {
+    console.error(error.message || error)
+    process.exitCode = 1
+    return
+  }
+
+  if (useJson) return jsonOrText(result, true)
+
+  if (sub === 'merge') return printMemoryMerge(result)
+
+  if (sub === 'forget') {
+    console.log(`Forgot ${result.forgottenCount} record(s) permanently (project=${result.project || 'current'})`)
+    if (result.forgotten.length > 0) console.log(`deleted: ${result.forgotten.join(', ')}`)
+    if (result.notFound.length > 0) console.log(`not found: ${result.notFound.join(', ')}`)
+    console.log(`audit: ${result.auditPath}`)
+    // Everything missing is a typo; do not let a no-op read as success.
+    if (result.forgottenCount === 0) {
+      process.exitCode = 1
+      return
+    }
+    // No undo is possible -- say so, and point at what should have been used.
+    console.log('This cannot be undone. (Next time consider `mio memory archive`, which is reversible.)')
+    return
+  }
+
+  if (sub === 'migrate') {
+    const target = result.scope === 'global' ? 'global' : `project=${result.project || 'current'}`
+    console.log(`Migrated ${result.updatedCount} record(s) to ${target}`)
+    if (result.updated.length > 0) console.log(`updated: ${result.updated.join(', ')}`)
+    if (result.unchanged.length > 0) console.log(`already ${result.scope}: ${result.unchanged.join(', ')}`)
+    if (result.notFound.length > 0) console.log(`not found: ${result.notFound.join(', ')}`)
+    // Every requested id missing is almost certainly a typo; signal it so
+    // scripts do not treat a no-op as success.
+    if (result.updatedCount === 0 && result.unchanged.length === 0) {
+      process.exitCode = 1
+      return
+    }
+    if (result.updated.length > 0) {
+      const back = result.scope === 'global' ? 'project' : 'global'
+      console.log(`Undo with: mio memory migrate --ids ${result.updated.join(',')} --scope ${back}`)
+    }
+    return
+  }
+
+  const verb = sub === 'restore' ? 'Restored' : 'Archived'
+  console.log(`${verb} ${result.archivedCount} record(s) (project=${result.project || 'current'})`)
+  if (result.archived.length > 0) console.log(`${sub}d: ${result.archived.join(', ')}`)
+  if (result.skipped.length > 0) console.log(`already ${sub}d: ${result.skipped.join(', ')}`)
+  if (result.notFound.length > 0) console.log(`not found: ${result.notFound.join(', ')}`)
+  if (result.archivedCount === 0) {
+    // Nothing changed: either the ids do not exist or they were already in the
+    // target state. Either way the caller's intent was not fulfilled, so say so
+    // instead of silently exiting 0.
+    console.error(`No records ${sub}d. Check the ids, or run: mio memory analyze`)
+    process.exitCode = 1
+    return
+  }
+  if (sub === 'archive' && result.archived.length > 0) {
+    console.log(`Undo with: mio memory restore --ids ${result.archived.join(',')}`)
+  }
 }
 
-function agentsCommand(useJson) {
-  const config = readConfig()
-  const agents = config.agents || {}
-  if (useJson) {
-    jsonOrText(agents, true)
+// The observer auto-claims experience reuse, but an unconfirmed auto_claim
+// never counts as `verified` and so never feeds memory ranking or task
+// routing. In practice most claims sat unconfirmed forever because confirming
+// was only possible through the MCP server.
+function cliExperienceStore() {
+  return createExperienceStore({
+    dataDir: MIO_HOME,
+    projectName,
+  })
+}
+
+function experienceUsage() {
+  console.log(`Usage:
+  mio experience list                            List reuse records (--status/--target-agent/--project/--limit)
+  mio experience confirm --ids a,b               Confirm auto-claimed reuse (bulk supported)
+  mio experience reuse --source-agent A --target-agent B --experience-id X   Record a reuse manually
+
+Options:
+  --ids a,b,c            Record ids (also accepted as bare positionals; confirm only)
+  --status pending|confirmed|verified|auto_claim|agent_report|all   Filter for list (default all)
+  --target-agent name    Only records reused by this agent
+  --experience-id id     The memory/experience that was reused
+  --source-agent name    Agent the experience came from
+  --reuse --behavior-changed --outcome-improved   Booleans describing the reuse
+  --outcome-improved     Confirm: accept the claim (--no-improved rejects it)
+  --by name              Who is confirming (default: cli)
+  --notes text           Optional note
+  --force                Re-confirm records that are already confirmed
+  --project name         Project filter (defaults to current directory name)
+  --json                 Machine-readable output
+`)
+}
+
+function printExperienceList(result) {
+  console.log(
+    `Experience reuse: ${result.count} shown of ${result.total} (project=${result.project || 'all'}, status=${result.status})`,
+  )
+  if (result.count === 0) {
+    console.log('No reuse records match.')
+    return
+  }
+  result.records.forEach((record, index) => {
+    const state = record.confirmed ? 'confirmed' : 'pending'
+    const badge = record.reuse && record.behaviorChanged && record.outcomeImproved ? 'verified' : state
+    const arrow = `${record.sourceAgent || '?'} -> ${record.targetAgent || '?'}`
+    const flags = [
+      record.reuse ? 'reused' : null,
+      record.behaviorChanged ? 'changed' : null,
+      record.outcomeImproved ? 'improved' : null,
+    ]
+      .filter(Boolean)
+      .join(',')
+    console.log(`${index + 1}. [${badge}] ${record.experienceId}  ${arrow}`)
+    console.log(
+      `   ${flags || 'no effect flags'} | ${(record.timestamp || '').slice(0, 19)} | ${record.id}`,
+    )
+    if (record.notes) console.log(`   note: ${record.notes}`)
+  })
+  // Only auto_claims are confirmable — experience.confirm rejects anything else.
+  const pending = result.records
+    .filter((record) => record.source === 'auto_claim' && !record.confirmed)
+    .map((record) => record.id)
+  if (pending.length > 0) {
+    console.log('')
+    console.log(
+      `${pending.length} unconfirmed auto-claim(s) shown. Unconfirmed auto-claims never count as verified, so they do not affect ranking or routing.`,
+    )
+    console.log(`Confirm with: mio experience confirm --ids ${pending.join(',')}`)
+  }
+}
+
+function experienceCommand(args, useJson) {
+  const sub = args[1]
+  const flags = args.slice(2)
+
+  if (!sub || sub === 'help' || sub === '--help' || sub === '-h') {
+    experienceUsage()
+    if (!sub) process.exitCode = 1
+    return
+  }
+  if (!['list', 'confirm', 'reuse'].includes(sub)) {
+    console.error(`Unknown experience subcommand: ${sub}`)
+    experienceUsage()
+    process.exitCode = 1
+    return
+  }
+
+  const store = cliExperienceStore()
+
+  if (sub === 'list') {
+    let result
+    try {
+      result = store.listReuse({
+        status: optionValue(flags, '--status'),
+        targetAgent: optionValue(flags, '--target-agent'),
+        project: optionValue(flags, '--project'),
+        limit: parseNumberOption(flags, '--limit'),
+      })
+    } catch (error) {
+      console.error(error.message || error)
+      process.exitCode = 1
+      return
+    }
+    if (useJson) return jsonOrText(result, true)
+    return printExperienceList(result)
+  }
+
+  let result
+  try {
+    if (sub === 'confirm') {
+      const ids = collectIds(flags)
+      if (ids.length === 0) {
+        console.error('mio experience confirm requires --ids <id,id,...> (or bare ids)')
+        process.exitCode = 1
+        return
+      }
+      // Absent = keep whatever the auto-claim recorded; --no-improved rejects it.
+      const improved = flagPresent(flags, '--no-improved')
+        ? false
+        : flagPresent(flags, '--outcome-improved')
+          ? true
+          : undefined
+      result = store.confirmReuse({
+        ids,
+        outcomeImproved: improved,
+        confirmedBy: optionValue(flags, '--by') || 'cli',
+        notes: optionValue(flags, '--notes'),
+        force: flagPresent(flags, '--force'),
+      })
+    } else {
+      result = store.recordReuse({
+        sourceAgent: optionValue(flags, '--source-agent'),
+        targetAgent: optionValue(flags, '--target-agent'),
+        experienceId: optionValue(flags, '--experience-id'),
+        reuse: flagPresent(flags, '--reuse'),
+        behaviorChanged: flagPresent(flags, '--behavior-changed'),
+        outcomeImproved: flagPresent(flags, '--outcome-improved'),
+        project: optionValue(flags, '--project'),
+        notes: optionValue(flags, '--notes'),
+      })
+    }
+  } catch (error) {
+    console.error(error.message || error)
+    process.exitCode = 1
     return
   }
-  const names = Object.keys(agents)
-  if (names.length === 0) {
-    console.log('No agents installed.')
+
+  if (useJson) return jsonOrText(result, true)
+
+  if (sub === 'reuse') {
+    console.log(
+      `Recorded reuse ${result.evidence.id}: ${result.evidence.sourceAgent} -> ${result.evidence.targetAgent} (${result.evidence.experienceId})`,
+    )
     return
   }
-  for (const name of names) {
-    console.log(`${name}\t${agents[name].installedAt || ''}`)
+
+  console.log(`Confirmed ${result.confirmedCount} reuse record(s)`)
+  if (result.confirmedIds.length > 0) console.log(`confirmed: ${result.confirmedIds.join(', ')}`)
+  if (result.rejected.length > 0) {
+    console.log(`skipped (not auto_claim): ${result.rejected.join(', ')}`)
+  }
+  if (result.alreadyConfirmed.length > 0) {
+    console.log(`already confirmed: ${result.alreadyConfirmed.join(', ')} (use --force to redo)`)
+  }
+  if (result.notFound.length > 0) console.log(`not found: ${result.notFound.join(', ')}`)
+  if (result.confirmedCount === 0) {
+    console.error('Nothing confirmed. Check the ids, or run: mio experience list --status pending')
+    process.exitCode = 1
+  }
+}
+
+// mio.policy.check has always been able to answer "has this action failed
+// before?", but only from inside an MCP session -- so in practice nobody asked
+// before running the command. This gives it a terminal entry point.
+function policyUsage() {
+  console.log(`Usage:
+  mio policy check "<action>"     Historical risk for an action, from trace + memory evidence
+
+Options:
+  --project name     Project filter (defaults to current directory name)
+  --json             Machine-readable output (same shape as mio.policy.check)
+
+Examples:
+  mio policy check "npm publish"
+  mio policy check "git reset --hard" --project akemi-mio
+
+Reading the result:
+  unknown   no history for this action; treat as normal risk
+  low       history exists and is mostly clean
+  moderate  add verification before proceeding
+  low/high  driven by the failure share of matching traces`)
+}
+
+function printPolicyCheck(result) {
+  const level = result.riskLevel.toUpperCase()
+  console.log(`Policy check: "${result.action}" (project=${result.project || 'all'})`)
+  console.log(`risk: ${level}${result.risk === null ? '' : ` (${result.risk})`}`)
+  if (result.total === 0) {
+    console.log('evidence: no matching traces')
+  } else {
+    console.log(`evidence: ${result.total} matching trace(s), ${result.failures} failure(s)`)
+    const outcomes = Object.entries(result.outcomeCounts)
+    if (outcomes.length > 0) {
+      console.log(`outcomes: ${outcomes.map(([k, v]) => `${k}=${v}`).join(', ')}`)
+    }
+  }
+
+  // A risk level derived from tokens that match most of the corpus looks
+  // authoritative but means nothing. Say so instead of letting it stand.
+  const diag = result.diagnostics
+  if (diag && diag.lowSignal) {
+    console.log('')
+    console.log(
+      `WARNING: low-signal match. Every token in this action (${diag.genericTokens.join(', ')}) ` +
+        'appears in most traces regardless of subject, so the level above is not meaningful.',
+    )
+    console.log('Try a more specific action, e.g. mio policy check "npm publish".')
+  } else if (diag && diag.lowSample) {
+    console.log('')
+    console.log(
+      `NOTE: only ${diag.sampleSize} matching trace(s); treat the level above as a weak signal.`,
+    )
+  }
+
+  console.log(`\n${result.suggestion}`)
+
+  if (result.failureExamples.length > 0) {
+    console.log('\nRecent failures:')
+    for (const example of result.failureExamples) {
+      const when = example.timestamp ? example.timestamp.slice(0, 19).replace('T', ' ') : 'unknown time'
+      console.log(`- [${example.outcome || 'failure'}] ${when}`)
+      console.log(`    ${example.summary}`)
+      if (example.trace_id) console.log(`    trace: ${example.trace_id}`)
+    }
+  }
+
+  if (result.guidance.level !== 'none') {
+    console.log(`\nGuidance (${result.guidance.level}): ${result.guidance.rationale}`)
+    if (result.guidance.avoid.length > 0) {
+      console.log('avoid repeating:')
+      for (const item of result.guidance.avoid) console.log(`- ${item}`)
+    }
+    for (const step of result.guidance.verificationSteps) console.log(`- ${step}`)
+  }
+
+  if (result.related_memories.length > 0) {
+    console.log('\nRelated memory:')
+    for (const record of result.related_memories) {
+      const badge = record.evidence ? ` [verified x${record.evidence.reuseCount}]` : ''
+      console.log(`- ${record.id}${badge} ${String(record.content).replace(/\s+/g, ' ').slice(0, 110)}`)
+    }
+  }
+}
+
+function policyCommand(args, useJson) {
+  const sub = args[1]
+
+  if (!sub || sub === 'help' || sub === '--help' || sub === '-h') {
+    policyUsage()
+    if (!sub) process.exitCode = 1
+    return
+  }
+  if (sub !== 'check') {
+    console.error(`Unknown policy subcommand: ${sub}`)
+    policyUsage()
+    process.exitCode = 1
+    return
+  }
+
+  const flags = args.slice(2)
+  // The action is a positional, and may be quoted as one argument or passed as
+  // several words (`mio policy check npm publish`). Only flags this command
+  // actually understands are treated as options -- anything else, including
+  // `--hard` in `git reset --hard`, belongs to the action being described.
+  const POLICY_KNOWN_FLAGS = new Set(['--project', '--json'])
+  const actionParts = []
+  for (let i = 0; i < flags.length; i += 1) {
+    const token = flags[i]
+    if (POLICY_KNOWN_FLAGS.has(token)) continue
+    if (i > 0 && POLICY_KNOWN_FLAGS.has(flags[i - 1])) continue
+    actionParts.push(token)
+  }
+  const action = actionParts.join(' ').trim()
+
+  if (!action) {
+    console.error('mio policy check requires an action, e.g. mio policy check "npm publish"')
+    process.exitCode = 1
+    return
+  }
+
+  let result
+  try {
+    result = cliPolicyStore().policyCheck({
+      action,
+      project: optionValue(flags, '--project'),
+    })
+  } catch (error) {
+    console.error(error.message || error)
+    process.exitCode = 1
+    return
+  }
+
+  if (useJson) return jsonOrText(result, true)
+  printPolicyCheck(result)
+}
+
+// ═════════════════════════════════════════════════════════════
+// mio insight — terminal entry point for mio.insight.*
+//
+// The insight engine (self-observation) shipped MCP-only: an agent could read
+// its own insights, but there was no way to see them from a terminal. These
+// commands delegate to the same shared store as the MCP tools, so the two
+// cannot disagree.
+//
+// `generate` is deliberately not exposed. It needs an LLM and the CLI has no
+// LLM wiring; the same call was made for creativity generate/ferment.
+// ═════════════════════════════════════════════════════════════
+function cliInsightStore() {
+  return createInsightStore({ dataDir: MIO_HOME })
+}
+
+function insightUsage() {
+  console.log(`Usage:
+  mio insight status                    Insight counts: total, reported, unreported, high-value
+  mio insight list                      List insights
+  mio insight mark-reported --ids a,b   Mark insights as reported (acknowledged)
+
+Options:
+  --unreported       Only unreported insights (list)
+  --min-score N      Minimum score (list)
+  --detector name    Filter by detector (list)
+  --limit N          Max insights (list)
+  --ids a,b          Comma-separated insight ids (mark-reported)
+  --json             Machine-readable output (same shape as mio.insight.*)
+
+The insight engine is an optional package (@akemi-mio/insight). If it is not
+installed these commands say so, rather than reporting a misleading empty list.`)
+}
+
+function printInsightStatus(result) {
+  console.log('Insight engine')
+  console.log(`  total:      ${result.total}`)
+  console.log(`  reported:   ${result.reported}`)
+  console.log(`  unreported: ${result.unreported}`)
+  console.log(`  high-value: ${result.highValue}`)
+  if (result.unreported > 0) {
+    console.log('\nUnreported insights are waiting: mio insight list --unreported')
+  }
+}
+
+function printInsightList(insights) {
+  if (insights.length === 0) {
+    console.log('No insights match.')
+    return
+  }
+  for (const item of insights) {
+    const score = typeof item.score === 'number' ? item.score.toFixed(2) : 'n/a'
+    const state = item.reported ? 'reported' : 'unreported'
+    const detector = item.detector ? `, ${item.detector}` : ''
+    console.log(`- [${score}] ${item.id || '(no id)'} (${state}${detector})`)
+    const text = String(item.content || item.summary || item.title || '')
+    if (text) console.log(`    ${text.replace(/\s+/g, ' ').slice(0, 140)}`)
+  }
+  console.log(`\n${insights.length} insight(s)`)
+}
+
+function insightCommand(args, useJson) {
+  const sub = args[1]
+  if (!sub || sub === 'help' || sub === '--help' || sub === '-h') {
+    insightUsage()
+    if (!sub) process.exitCode = 1
+    return
+  }
+  if (!['status', 'list', 'mark-reported'].includes(sub)) {
+    console.error(`Unknown insight subcommand: ${sub}`)
+    insightUsage()
+    process.exitCode = 1
+    return
+  }
+
+  const flags = args.slice(2)
+  const store = cliInsightStore()
+  try {
+    if (sub === 'status') {
+      const result = store.status()
+      if (useJson) return jsonOrText(result, true)
+      return printInsightStatus(result)
+    }
+    if (sub === 'list') {
+      const result = store.list({
+        unreported: flagPresent(flags, '--unreported'),
+        minScore: parseNumberOption(flags, '--min-score'),
+        detector: optionValue(flags, '--detector'),
+        limit: parseNumberOption(flags, '--limit'),
+      })
+      if (useJson) return jsonOrText(result, true)
+      return printInsightList(result)
+    }
+    const ids = splitTagsOption(flags, '--ids')
+    if (ids.length === 0) {
+      console.error('mio insight mark-reported requires --ids a,b')
+      process.exitCode = 1
+      return
+    }
+    const result = store.markReported({ ids })
+    if (useJson) return jsonOrText(result, true)
+    console.log(`Marked ${result.marked} insight(s) as reported.`)
+  } catch (error) {
+    console.error(error.message || error)
+    process.exitCode = 1
+  }
+}
+
+// ═════════════════════════════════════════════════════════════
+// mio observer — terminal entry point for the observer research pipeline
+// (mio.observer.*). Not to be confused with `mio observe`, which is the
+// WorkBuddy transcript daemon.
+//
+// Only the read-only views are exposed. collect hits the network and ferment
+// drives the engine; both belong to the daemon, not an interactive shell (same
+// reasoning as skipping creativity generate from the CLI).
+//
+// Base directory defaults to <cwd>/.local/observer -- the MCP server's default
+// -- because the pipeline is project-local by design. --base-dir overrides it.
+// ═════════════════════════════════════════════════════════════
+function cliObserverStore() {
+  return createObserverStore({})
+}
+
+function observerUsage() {
+  console.log(`Usage:
+  mio observer status        Pipeline stage counts (observations/trends/topics/...)
+  mio observer world-model   Entities, events, trends and narratives
+  mio observer trends        Recent trend reports
+  mio observer research      Research reports
+  mio observer insights      Observer-generated insight articles
+  mio observer essays        Published essays
+  mio observer dag           Daily summaries for the last N days
+  mio observer ingest        Record a trace event (tool_call/error/retry/task_outcome)
+                             --trace-id <id> --event-type <type> [--payload '<json>'] [--outcome]
+                             [--agent] [--host] [--project] (writes immediately, like mio remember)
+  mio observer subscribe     Subscribe to events: --event-types a,b [--topic T] [--ttl-days N]
+                             (--yes required to apply; previews by default)
+  mio observer digest        New events since the last digest (--limit/--project/--event-types)
+                             NOTE: advances the cursor, so a second run returns only newer events
+
+Options:
+  --base-dir DIR     Observer data directory (default: <cwd>/.local/observer)
+  --date YYYY-MM-DD  A single trend report (trends)
+  --type name        Essay type (default: published)
+  --days N           Days of history (dag, default 7)
+  --limit N          Max items
+  --json             Machine-readable output (same shape as mio.observer.*)`)
+}
+
+// Pipeline items are free-form JSON written by different stages, so pull a
+// human label from whichever well-known field happens to be present.
+function observerItemLabel(item) {
+  if (item === null || typeof item !== 'object') return String(item)
+  for (const key of ['title', 'name', 'topic', 'headline', 'summary', 'text', 'id', 'date']) {
+    const value = item[key]
+    if (typeof value === 'string' && value.trim()) return value.trim().replace(/\s+/g, ' ').slice(0, 120)
+  }
+  return JSON.stringify(item).slice(0, 120)
+}
+
+function printObserverList(kind, items) {
+  if (items.length === 0) {
+    console.log(`No ${kind} found. (Run the observer pipeline, or check --base-dir.)`)
+    return
+  }
+  for (const item of items) console.log(`- ${observerItemLabel(item)}`)
+  console.log(`\n${items.length} ${kind}`)
+}
+
+function printObserverStatus(status) {
+  console.log('Observer pipeline (file counts per stage)')
+  let total = 0
+  for (const [stage, count] of Object.entries(status)) {
+    total += count
+    console.log(`  ${stage.padEnd(14)} ${count}`)
+  }
+  console.log(`\n  ${'total'.padEnd(14)} ${total}`)
+  if (total === 0) {
+    console.log('\nNo pipeline data yet. Nothing has been collected into this directory.')
+  }
+}
+
+function printObserverWorldModel(model) {
+  const sections = ['entities', 'events', 'trends', 'narratives']
+  console.log('World model')
+  for (const key of sections) {
+    const value = model[key]
+    const count = Array.isArray(value) ? value.length : Object.keys(value || {}).length
+    console.log(`  ${key.padEnd(12)} ${count}`)
+  }
+  if (sections.every((key) => (Array.isArray(model[key]) ? model[key].length : 0) === 0)) {
+    console.log('\nWorld model is empty.')
+  }
+}
+
+function printObserverEssays(essays) {
+  if (essays.length === 0) {
+    console.log('No essays found.')
+    return
+  }
+  for (const essay of essays) {
+    console.log(`- ${essay.file} (${essay.type}, ${String(essay.created).slice(0, 10)})`)
+    const firstLine = String(essay.content || '').split('\n').find((l) => l.trim()) || ''
+    if (firstLine) console.log(`    ${firstLine.replace(/\s+/g, ' ').slice(0, 120)}`)
+  }
+  console.log(`\n${essays.length} essay(s)`)
+}
+
+function printObserverDag(result, days) {
+  if (result.summaryCount === 0) {
+    console.log(`No daily summaries found in the last ${days || 7} day(s).`)
+    return
+  }
+  for (const day of result.summaries) {
+    console.log(`- ${day.date}: ${day.summaries.length} summary(ies)`)
+  }
+  console.log(`\n${result.summaryCount} summary(ies) across ${result.summaries.length} day(s)`)
+}
+
+function cliSubscriptionStore() {
+  return createSubscriptionStore({
+    dataDir: MIO_HOME,
+    projectName,
+    agentId: () => 'cli',
+  })
+}
+
+// Records an arbitrary trace event (tool_call / error / retry / task_outcome).
+// Unlike `mio task record-outcome`, which is specific to task outcomes, this is
+// the general-purpose half -- and the one AGENTS.md asks agents to call.
+//
+// It appends rather than modifies, so like `mio remember` it writes straight
+// away instead of previewing: there is nothing to destroy. Missing arguments are
+// still rejected before anything is written.
+function observerIngestCommand(args, useJson) {
+  const flags = args.slice(2)
+  const traceId = optionValue(flags, '--trace-id')
+  const eventType = optionValue(flags, '--event-type')
+  if (!traceId || !eventType) {
+    console.error('mio observer ingest requires --trace-id <id> and --event-type <type>')
+    console.error('Example: mio observer ingest --trace-id t1 --event-type tool_call --payload \'{"tool":"Bash"}\'')
+    process.exitCode = 1
+    return
+  }
+
+  let payload = {}
+  const rawPayload = optionValue(flags, '--payload')
+  if (rawPayload) {
+    try {
+      payload = JSON.parse(rawPayload)
+    } catch (error) {
+      console.error(`--payload must be valid JSON: ${error.message}`)
+      process.exitCode = 1
+      return
+    }
+    if (!payload || typeof payload !== 'object' || Array.isArray(payload)) {
+      console.error('--payload must be a JSON object')
+      process.exitCode = 1
+      return
+    }
+  }
+
+  let result
+  try {
+    result = cliTaskStore().ingestObservation({
+      trace_id: traceId,
+      event_type: eventType,
+      outcome: optionValue(flags, '--outcome'),
+      payload,
+      agent: optionValue(flags, '--agent'),
+      host: optionValue(flags, '--host'),
+      project: optionValue(flags, '--project'),
+    })
+  } catch (error) {
+    console.error(error.message || error)
+    process.exitCode = 1
+    return
+  }
+
+  if (useJson) return jsonOrText(result, true)
+  console.log(`Ingested ${result.event.event_type} (trace=${result.event.trace_id})`)
+  console.log(`  agent=${result.event.agent} project=${result.event.project} host=${result.event.host}`)
+  if (result.autoClaims && result.autoClaims.length > 0) {
+    console.log(`  auto-claims: ${result.autoClaims.length}`)
+  }
+}
+
+// Subscribe creates or renews a subscription, so it previews like every other
+// mutating command (`mio agents register`, `mio memory archive`): exit 1 and
+// nothing written without --yes.
+function observerSubscribeCommand(args, useJson) {
+  const flags = args.slice(2)
+  const project = optionValue(flags, '--project') || projectName()
+  const eventTypes = splitTagsOption(flags, '--event-types')
+  const topic = optionValue(flags, '--topic') || ''
+  const ttlDays = optionValue(flags, '--ttl-days')
+
+  if (!flagPresent(flags, '--yes')) {
+    const store = cliSubscriptionStore()
+    const existing = readJsonl(store.subscriptionPath).find((subscription) => {
+      return (
+        subscription &&
+        subscription.agent === 'cli' &&
+        subscribeKey(subscription.agent, subscription.project, subscription.eventTypes || [], subscription.topic || '') ===
+          subscribeKey('cli', project, eventTypes, topic)
+      )
+    })
+    if (useJson) {
+      jsonOrText(
+        {
+          preview: true,
+          applied: false,
+          project,
+          eventTypes,
+          topic: topic || null,
+          ttlDays: ttlDays ? Number(ttlDays) : null,
+          willCreate: !existing,
+          hint: 'Re-run with --yes to apply.',
+        },
+        true,
+      )
+    } else {
+      console.log(`Subscribe preview: project=${project}`)
+      console.log(`  event types: ${eventTypes.length > 0 ? eventTypes.join(', ') : '(all)'}`)
+      if (topic) console.log(`  topic: ${topic}`)
+      console.log(`  ${existing ? 'will renew the existing subscription' : 'will create a new subscription'}`)
+      console.log('Re-run with --yes to apply.')
+    }
+    process.exitCode = 1
+    return
+  }
+
+  let result
+  try {
+    result = cliSubscriptionStore().subscribe({ project, eventTypes, topic, ttlDays })
+  } catch (error) {
+    console.error(error.message || error)
+    process.exitCode = 1
+    return
+  }
+  if (useJson) return jsonOrText(result, true)
+  console.log(`Subscribed (project=${result.project})`)
+  console.log(`  id=${result.subscription.id} expires=${result.subscription.expiresAt}`)
+  console.log(`  event types: ${(result.subscription.eventTypes || []).join(', ') || '(all)'}`)
+}
+
+// The digest is cursor-based: it returns only events newer than the last one it
+// delivered, and ADVANCES that cursor. Running it twice therefore gives different
+// results -- so it cannot be previewed, and the output says so rather than
+// leaving that surprise for the user to discover.
+function observerDigestCommand(args, useJson) {
+  const flags = args.slice(2)
+  let result
+  try {
+    result = cliSubscriptionStore().digest({
+      project: optionValue(flags, '--project'),
+      limit: optionValue(flags, '--limit'),
+      eventTypes: splitTagsOption(flags, '--event-types'),
+    })
+  } catch (error) {
+    console.error(error.message || error)
+    process.exitCode = 1
+    return
+  }
+  if (useJson) return jsonOrText(result, true)
+  console.log(`Observer digest: ${result.count} event(s) (project=${result.project}, agent=${result.agent})`)
+  console.log(`  subscriptions: ${result.subscriptionCount} active, ${result.matchedSubscriptions.length} matched`)
+  for (const event of result.events) {
+    console.log(`- ${event.event_type}${event.outcome ? ` (${event.outcome})` : ''} trace=${event.trace_id || '-'}`)
+  }
+  if (result.count === 0) {
+    console.log('Nothing new since the last digest.')
+  }
+  console.log('Note: this advances the cursor; the next digest returns only newer events.')
+}
+
+function observerCommand(args, useJson) {
+  const sub = args[1]
+  if (!sub || sub === 'help' || sub === '--help' || sub === '-h') {
+    observerUsage()
+    if (!sub) process.exitCode = 1
+    return
+  }
+  if (sub === 'ingest') return observerIngestCommand(args, useJson)
+  if (sub === 'subscribe') return observerSubscribeCommand(args, useJson)
+  if (sub === 'digest') return observerDigestCommand(args, useJson)
+
+  if (!['status', 'world-model', 'trends', 'research', 'insights', 'essays', 'dag'].includes(sub)) {
+    console.error(`Unknown observer subcommand: ${sub}`)
+    observerUsage()
+    process.exitCode = 1
+    return
+  }
+
+  const flags = args.slice(2)
+  const store = cliObserverStore()
+  const base = { baseDir: optionValue(flags, '--base-dir') }
+  try {
+    if (sub === 'status') {
+      const result = store.status(base)
+      if (useJson) return jsonOrText(result, true)
+      return printObserverStatus(result)
+    }
+    if (sub === 'world-model') {
+      const result = store.worldModel(base)
+      if (useJson) return jsonOrText(result, true)
+      return printObserverWorldModel(result)
+    }
+    if (sub === 'trends') {
+      const result = store.trends({ ...base, date: optionValue(flags, '--date'), limit: parseNumberOption(flags, '--limit') })
+      if (useJson) return jsonOrText(result, true)
+      return printObserverList('trend report(s)', result)
+    }
+    if (sub === 'research') {
+      const result = store.research({ ...base, limit: parseNumberOption(flags, '--limit') })
+      if (useJson) return jsonOrText(result, true)
+      return printObserverList('research report(s)', result)
+    }
+    if (sub === 'insights') {
+      const result = store.insights({ ...base, limit: parseNumberOption(flags, '--limit') })
+      if (useJson) return jsonOrText(result, true)
+      return printObserverList('insight(s)', result)
+    }
+    if (sub === 'essays') {
+      const result = store.essays({ ...base, type: optionValue(flags, '--type'), limit: parseNumberOption(flags, '--limit') })
+      if (useJson) return jsonOrText(result, true)
+      return printObserverEssays(result)
+    }
+    const days = parseNumberOption(flags, '--days')
+    const result = store.dag({ ...base, days })
+    if (useJson) return jsonOrText(result, true)
+    return printObserverDag(result, days)
+  } catch (error) {
+    console.error(error.message || error)
+    process.exitCode = 1
+  }
+}
+
+function creativityUsage() {
+  console.log(`Usage:
+  mio creativity status              Show hypothesis counts and recent top ideas
+  mio creativity list                List hypotheses (--status active|validated|rejected|draft, --limit N)
+
+  mio creativity generate        Generate hypotheses by combining sources (needs 2+ --source, calls an LLM)
+  mio creativity ferment         Review/refine active hypotheses (calls an LLM)
+
+Options:
+  --status name      Filter list by status
+  --limit N          Max results for list / ferment (default 20 / 5)
+  --source "name|content"   A concept source for generate; repeat 2+ times
+  --strategy explore|signal|stable   Generation strategy (auto-selected if omitted)
+  --json             Machine-readable output (same shape as the mio.creativity.* tools)
+
+Examples:
+  mio creativity status
+  mio creativity list --status rejected --limit 10
+  mio creativity generate --source "auth|token rotation" --source "cache|write-through"
+
+generate/ferment call an LLM using the shared client. Configure it with
+LLM_API_URL / LLM_KEY / LLM_CHAT_MODEL (the same environment the MCP server
+reads). A local Ollama works:
+  LLM_API_URL=http://localhost:11434/v1/chat/completions`)
+}
+
+function printCreativityStatus(result) {
+  console.log('Creativity engine:')
+  console.log(`  hypotheses: ${result.hypotheses}  combos: ${result.combos}  experiments: ${result.experiments}`)
+  console.log(`  active: ${result.active}  validated: ${result.validated}  rejected: ${result.rejected}`)
+  if (result.recentIdeas.length === 0) {
+    console.log('No hypotheses yet.')
+    return
+  }
+  console.log('\nRecent top ideas:')
+  for (const idea of result.recentIdeas) {
+    console.log(`- ${idea.title}  (novelty=${idea.novelty} feasibility=${idea.feasibility} impact=${idea.impact} score=${idea.score})`)
+    console.log(`    ${idea.id}`)
+  }
+}
+
+function printCreativityList(items) {
+  if (items.length === 0) {
+    console.log('No hypotheses match.')
+    return
+  }
+  console.log(`Hypotheses: ${items.length} shown`)
+  items.forEach((h, index) => {
+    const when = h.createdAt ? new Date(h.createdAt).toISOString().slice(0, 19).replace('T', ' ') : ''
+    const labels = Array.isArray(h.sourceLabels) && h.sourceLabels.length > 0 ? h.sourceLabels.join(' + ') : '—'
+    console.log(`${index + 1}. [${h.status}] ${h.title}  (N=${h.novelty} F=${h.feasibility} I=${h.impact} score=${h.score})`)
+    console.log(`   ${labels} | ${when} | ${h.id}`)
+    if (h.fermentCount) console.log(`   fermented ${h.fermentCount}x`)
+  })
+}
+
+// Parses repeated `--source "name|content"` into the shape CreativityEngine
+// expects. At least two are required: the engine pairs up sources, so one
+// source can never produce a combination.
+function parseSources(flags) {
+  const sources = []
+  for (let i = 0; i < flags.length; i += 1) {
+    if (flags[i] !== '--source') continue
+    const raw = flags[i + 1]
+    if (!raw) continue
+    const sep = raw.indexOf('|')
+    if (sep === -1) {
+      sources.push({ name: raw, content: raw })
+    } else {
+      sources.push({
+        name: raw.slice(0, sep).trim(),
+        content: raw.slice(sep + 1).trim(),
+      })
+    }
+  }
+  return sources
+}
+
+function warnIfLlmUnconfigured() {
+  if (isLlmConfigured()) return
+  const { apiUrl, model } = llmConfig()
+  console.log(`Note: no LLM_API_URL / LLM_KEY set, so this will call the default endpoint (${model} @ ${apiUrl}).`)
+  console.log('Set LLM_API_URL (e.g. http://localhost:11434/v1/chat/completions for a local Ollama) to point elsewhere.')
+}
+
+function creativityGenerateCommand(args, useJson) {
+  const flags = args.slice(2)
+  const sources = parseSources(flags)
+  if (sources.length < 2) {
+    console.error('mio creativity generate requires at least two --source "name|content" arguments')
+    console.error('Example: mio creativity generate --source "auth|token rotation" --source "cache|write-through cache"')
+    process.exitCode = 1
+    return
+  }
+  const strategy = optionValue(flags, '--strategy')
+
+  if (!useJson) warnIfLlmUnconfigured()
+
+  let result
+  try {
+    result = cliCreativityEngine().generate(sources, strategy)
+  } catch (error) {
+    console.error(error.message || error)
+    process.exitCode = 1
+    return
+  }
+  // generate is async in the engine.
+  return Promise.resolve(result).then((resolved) => {
+    if (useJson) return jsonOrText(resolved, true)
+    if (resolved.reason) {
+      console.log(`No hypotheses generated: ${resolved.reason}`)
+      return
+    }
+    console.log(`Generated ${resolved.ideas.length} hypothesis/hypotheses (strategy=${resolved.strategy})`)
+    resolved.ideas.forEach((idea, index) => {
+      console.log(`${index + 1}. [${idea.status}] ${idea.title}`)
+      console.log(`   novelty=${idea.novelty} feasibility=${idea.feasibility} impact=${idea.impact} | ${idea.id}`)
+      if (idea.rejectionReason) console.log(`   rejected: ${idea.rejectionReason}`)
+    })
+  })
+}
+
+function creativityFermentCommand(args, useJson) {
+  const flags = args.slice(2)
+  const limit = parseNumberOption(flags, '--limit')
+
+  if (!useJson) warnIfLlmUnconfigured()
+
+  let result
+  try {
+    result = cliCreativityEngine().ferment(limit)
+  } catch (error) {
+    console.error(error.message || error)
+    process.exitCode = 1
+    return
+  }
+  return Promise.resolve(result).then((resolved) => {
+    if (useJson) return jsonOrText(resolved, true)
+    if (resolved.fermented === 0) {
+      console.log(`Nothing fermented: ${resolved.reason || 'no eligible hypotheses'}`)
+      return
+    }
+    console.log(`Fermented ${resolved.fermented} hypothesis/hypotheses`)
+    for (const r of resolved.results || []) {
+      console.log(`- ${r.title || r.id}: ${r.verdict || 'updated'}${r.reason ? ` — ${r.reason}` : ''}`)
+    }
+  })
+}
+
+function creativityCommand(args, useJson) {
+  const sub = args[1]
+
+  if (!sub || sub === 'help' || sub === '--help' || sub === '-h') {
+    creativityUsage()
+    if (!sub) process.exitCode = 1
+    return
+  }
+  if (sub === 'generate') return creativityGenerateCommand(args, useJson)
+  if (sub === 'ferment') return creativityFermentCommand(args, useJson)
+
+  if (!['status', 'list'].includes(sub)) {
+    console.error(`Unknown creativity subcommand: ${sub}`)
+    creativityUsage()
+    process.exitCode = 1
+    return
+  }
+
+  const flags = args.slice(2)
+  const engine = cliCreativityEngine()
+
+  let result
+  try {
+    if (sub === 'status') {
+      result = engine.status()
+    } else {
+      result = engine.list({
+        status: optionValue(flags, '--status'),
+        limit: parseNumberOption(flags, '--limit'),
+      })
+    }
+  } catch (error) {
+    console.error(error.message || error)
+    process.exitCode = 1
+    return
+  }
+
+  if (useJson) return jsonOrText(result, true)
+  if (sub === 'status') printCreativityStatus(result)
+  else printCreativityList(result)
+}
+
+function readConfig() {
+  const fallback = { version: 1, home: MIO_HOME, agents: {} }
+  if (!fs.existsSync(CONFIG_FILE)) return fallback
+  try {
+    return { ...fallback, ...JSON.parse(fs.readFileSync(CONFIG_FILE, 'utf8')) }
+  } catch (_) {
+    return fallback
+  }
+}
+
+function writeConfig(config) {
+  ensureHome()
+  fs.writeFileSync(CONFIG_FILE, `${JSON.stringify(config, null, 2)}\n`, 'utf8')
+}
+
+function jsonOrText(payload, useJson) {
+  if (useJson) {
+    console.log(JSON.stringify(payload, null, 2))
+    return
+  }
+  return payload
+}
+
+function initCommand() {
+  ensureHome()
+  const config = readConfig()
+  if (!config.createdAt) config.createdAt = new Date().toISOString()
+  writeConfig(config)
+  console.log(`MIO_HOME ready: ${MIO_HOME}`)
+}
+
+function statusCommand(useJson) {
+  const config = readConfig()
+  const codexRestart = adapters.codex.restartWarningMessage()
+  const opencodeRestart = adapters.opencode.restartWarningMessage()
+  const installed = {
+    codex: adapters.codex.isInstalled(),
+    opencode: adapters.opencode.isInstalled(),
+    workbuddy: adapters.workbuddy.isInstalled(),
+    hermes: adapters.hermes.isInstalled(),
+    claude: adapters.claude.isInstalled(),
+  }
+  const payload = {
+    version: config.version,
+    home: MIO_HOME,
+    serverScript: SERVER_SCRIPT,
+    serverScriptExists: fs.existsSync(SERVER_SCRIPT),
+    agents: config.agents || {},
+    codexInstalled: installed.codex,
+    opencodeInstalled: installed.opencode,
+    workbuddyInstalled: installed.workbuddy,
+    hermesInstalled: installed.hermes,
+    claudeInstalled: installed.claude,
+    codexNeedsRestart: Boolean(codexRestart),
+    opencodeNeedsRestart: Boolean(opencodeRestart),
+    codexRestartHint: codexRestart,
+    opencodeRestartHint: opencodeRestart,
+    hermesObserverReady: fs.existsSync(observer.hermesDbPath()),
+    claudeObserverReady: fs.existsSync(observer.claudeProjectsDir()),
+    opencodeObserverReady: fs.existsSync(observer.opencodeDbPath()),
+    observerRunning: observer.isRunning(MIO_HOME),
+    observerPid: observer.isRunning(MIO_HOME) ? observer.readPid(MIO_HOME) : null,
+  }
+  if (useJson) {
+    jsonOrText(payload, true)
+    return
+  }
+  console.log(`MIO_HOME: ${MIO_HOME}`)
+  console.log(`MCP Server: ${SERVER_SCRIPT}`)
+  console.log(`Codex MCP: ${installed.codex ? 'installed' : 'not installed'}`)
+  if (codexRestart) console.log(`Codex app: running with an older Mio config - fully restart Codex to load it`)
+  console.log(`OpenCode MCP: ${installed.opencode ? 'installed' : 'not installed'}`)
+  if (opencodeRestart) console.log(`OpenCode Desktop: running with an older Mio config - restart it to load it`)
+  console.log(`OpenCode observer: ${fs.existsSync(observer.opencodeDbPath()) ? 'ready' : 'no database'}`)
+  console.log(`WorkBuddy MCP: ${installed.workbuddy ? 'installed' : 'not installed'}`)
+  console.log(`Hermes MCP: ${installed.hermes ? 'installed' : 'not installed'}`)
+  console.log(`Hermes observer: ${fs.existsSync(observer.hermesDbPath()) ? 'ready' : 'no database'}`)
+  console.log(`Claude Code MCP: ${installed.claude ? 'installed' : 'not installed'}`)
+  console.log(`Claude Code observer: ${fs.existsSync(observer.claudeProjectsDir()) ? 'ready' : 'no transcripts'}`)
+  console.log(`Observer: ${observer.isRunning(MIO_HOME) ? 'running (pid ' + observer.readPid(MIO_HOME) + ')' : 'not running'}`)
+}
+
+function agentsCommand(args, useJson) {
+  const sub = args[1]
+
+  // `mio agents` with no subcommand lists installed host adapters (config) —
+  // distinct from the observed-agent telemetry below.
+  if (!sub || sub === 'help' || sub === '--help' || sub === '-h') {
+    const config = readConfig()
+    const agents = config.agents || {}
+    if (useJson) {
+      jsonOrText(agents, true)
+      return
+    }
+    const names = Object.keys(agents)
+    if (names.length === 0) {
+      console.log('No agents installed. Run `mio agents list` for observed agents, or `mio install <host>` to add one.')
+      return
+    }
+    for (const name of names) {
+      console.log(`${name}\t${agents[name].installedAt || ''}`)
+    }
+    return
+  }
+
+  if (sub === 'list') {
+    const flags = args.slice(2)
+    let result
+    try {
+      result = cliAgentStore().listAgents({ project: optionValue(flags, '--project') })
+    } catch (error) {
+      console.error(error.message || error)
+      process.exitCode = 1
+      return
+    }
+    if (useJson) return jsonOrText(result, true)
+    return printAgentList(result)
+  }
+
+  if (sub === 'report') {
+    const flags = args.slice(2)
+    let result
+    try {
+      result = cliAgentStore().reportAgent({
+        agentId: optionValue(flags, '--agent'),
+        project: optionValue(flags, '--project'),
+      })
+    } catch (error) {
+      console.error(error.message || error)
+      process.exitCode = 1
+      return
+    }
+    if (useJson) return jsonOrText(result, true)
+    return printAgentReport(result)
+  }
+
+  if (sub === 'register') {
+    const flags = args.slice(2)
+    const agentId = optionValue(flags, '--agent-id')
+    if (!agentId) {
+      console.error('mio agents register requires --agent-id <id>')
+      process.exitCode = 1
+      return
+    }
+    const project = optionValue(flags, '--project') || projectName()
+    const hostType = optionValue(flags, '--host-type') || 'mcp'
+    const capabilities = splitTagsOption(flags, '--capabilities')
+
+    // Preview by default, like `mio memory archive`: nothing is written
+    // without --yes, in --json mode either.
+    if (!flagPresent(flags, '--yes')) {
+      const existing = cliAgentStore()
+        .listAgents({ project })
+        .agents.find((a) => a.agentId === agentId)
+      if (useJson) {
+        jsonOrText(
+          {
+            preview: true,
+            applied: false,
+            agentId,
+            project,
+            hostType,
+            capabilities,
+            willCreate: !existing,
+            existingSessionCount: existing ? existing.sessionCount || 0 : null,
+            hint: 'Re-run with --yes to apply.',
+          },
+          true,
+        )
+      } else {
+        console.log(`Register preview: ${agentId} (project=${project}, host=${hostType})`)
+        if (existing) {
+          const next = (existing.sessionCount || 0) + 1
+          console.log(`  will update existing agent (sessionCount ${existing.sessionCount || 0} -> ${next})`)
+        } else {
+          console.log('  will create a new agent record')
+        }
+        if (capabilities.length > 0) console.log(`  capabilities: ${capabilities.join(', ')}`)
+        console.log('Re-run with --yes to apply.')
+      }
+      process.exitCode = 1
+      return
+    }
+
+    let result
+    try {
+      result = cliAgentStore().registerAgent({ agentId, project, hostType, capabilities })
+    } catch (error) {
+      console.error(error.message || error)
+      process.exitCode = 1
+      return
+    }
+    if (useJson) return jsonOrText(result, true)
+    console.log(`Registered ${result.agentId} (project=${project})`)
+    console.log(
+      `  ${result.registered ? 'created' : `updated (session ${result.sessionCount})`} | host=${result.hostType}`,
+    )
+    if (result.capabilities && result.capabilities.length > 0) {
+      console.log(`  capabilities: ${result.capabilities.join(', ')}`)
+    }
+    return
+  }
+
+  console.error(`Unknown agents subcommand: ${sub}`)
+  console.error('Usage: mio agents [list|register|report] [--agent X] [--project Y]')
+  process.exitCode = 1
+}
+
+function printAgentList(result) {
+  console.log(`Observed agents: ${result.count} shown${result.project ? ` (project=${result.project})` : ''}`)
+  if (result.count === 0) {
+    console.log('No observed agents match.')
+    return
+  }
+  for (const a of result.agents) {
+    const caps = Array.isArray(a.capabilities) && a.capabilities.length > 0 ? ` [${a.capabilities.join(', ')}]` : ''
+    console.log(`${a.agentId} (${a.hostType})${caps}  project=${a.project || '-'}`)
+    console.log(
+      `   sessions=${a.sessionCount || 0} tasks=${a.taskCount} success=${a.successCount} failure=${a.failureCount} | ${a.id}`,
+    )
+  }
+}
+
+function printAgentReport(result) {
+  console.log(`Agent report${result.project ? ` (project=${result.project})` : ''}: ${result.count} agent(s)`)
+  if (result.count === 0) {
+    console.log('No observed agents match.')
+    return
+  }
+  for (const r of result.reports) {
+    const o = r.taskOutcomes
+    const when = r.lastSeenAt ? r.lastSeenAt.slice(0, 19).replace('T', ' ') : 'unknown'
+    console.log(`${r.agentId} (${r.hostType})  ${r.active ? 'active' : 'idle'}`)
+    console.log(
+      `   tasks: ${o.total} total, ${o.success} success, ${o.failure} failure (${o.successRate}% success)`,
+    )
+    console.log(`   memories: ${r.memories}  experience reuses: ${r.experienceReuses.total} (verified ${r.experienceReuses.verified})`)
+    console.log(`   last seen: ${when} | sessions=${r.sessionCount}`)
+  }
+}
+
+function phase0Command(args, useJson) {
+  const sub = args[1]
+  if (!sub || sub === 'help' || sub === '--help' || sub === '-h') {
+    console.log(`Usage:
+  mio phase0 report              Show the Phase 0 validation report (memory/trace/reuse evidence)
+  mio phase0 report --format markdown   Render as Markdown
+  mio phase0 report --json      Machine-readable report (same shape as mio.phase0.report)
+
+Options:
+  --project name     Project filter (defaults to the current directory name)
+  --format markdown  Human-readable Markdown instead of JSON`)
+    if (!sub) process.exitCode = 1
+    return
+  }
+  if (sub !== 'report') {
+    console.error(`Unknown phase0 subcommand: ${sub}`)
+    process.exitCode = 1
+    return
+  }
+
+  const flags = args.slice(2)
+  const format = (optionValue(flags, '--format') || '').toLowerCase()
+  let report
+  try {
+    report = loadPhase0(MIO_HOME, optionValue(flags, '--project') || projectName())
+  } catch (error) {
+    console.error(error.message || error)
+    process.exitCode = 1
+    return
+  }
+
+  if (useJson) {
+    return jsonOrText({ project: optionValue(flags, '--project') || projectName(), ...report }, true)
+  }
+  // Markdown is the human-readable form; the MCP tool exposes the same renderer
+  // via --format markdown, so text and MCP output cannot diverge.
+  console.log(renderPhase0Markdown(report))
+}
+
+function hostCommand(args, useJson) {
+  const sub = args[1]
+  if (!sub || sub === 'help' || sub === '--help' || sub === '-h') {
+    console.log(`Usage:
+  mio host capabilities       Show what each host supports and whether it is installed
+  mio host capabilities --json   Machine-readable (same shape as mio.host.capabilities)
+
+Notes:
+  "installed" is probed live from each adapter, so it reflects this machine right
+  now -- unlike \`mio install <host>\`, which only records intent in config.json.`)
+    if (!sub) process.exitCode = 1
+    return
+  }
+  if (sub !== 'capabilities') {
+    console.error(`Unknown host subcommand: ${sub}`)
+    process.exitCode = 1
+    return
+  }
+
+  const result = listHostCapabilities()
+  if (useJson) return jsonOrText(result, true)
+
+  console.log(`Host capabilities: ${result.hosts.length} host(s)`)
+  for (const host of result.hosts) {
+    const mark = host.installed ? 'installed' : 'not installed'
+    console.log(`  ${host.name.padEnd(10)} ${mark}`)
+    console.log(`             ${host.capabilities.join(', ')}`)
+  }
+}
+
+function recordOutcomeCommand(args, useJson) {
+  const flags = args.slice(2)
+  const outcome = String(optionValue(flags, '--outcome') || '').toLowerCase()
+  if (!['success', 'failure', 'aborted'].includes(outcome)) {
+    console.error('mio task record-outcome requires --outcome success|failure|aborted')
+    process.exitCode = 1
+    return
+  }
+  const project = optionValue(flags, '--project') || projectName()
+  // Mirrors the store default so the preview reports the same agent that the
+  // applied run would use.
+  const agentId = optionValue(flags, '--agent-id') || 'cli'
+  const task = optionValue(flags, '--task') || ''
+  const summary = optionValue(flags, '--summary') || ''
+  const verification = optionValue(flags, '--verification') || ''
+  const traceId = optionValue(flags, '--trace-id') || ''
+
+  // Preview by default, like every other CLI write (mio memory archive,
+  // mio agents register): nothing is written without --yes, --json included.
+  if (!flagPresent(flags, '--yes')) {
+    const existing = cliAgentStore()
+      .listAgents({ project })
+      .agents.find((a) => a.agentId === agentId)
+    if (useJson) {
+      jsonOrText(
+        {
+          preview: true,
+          applied: false,
+          outcome,
+          agentId,
+          project,
+          task,
+          summary,
+          verification,
+          agentWillUpdate: !!existing,
+          currentTaskCount: existing ? existing.taskCount || 0 : null,
+          hint: 'Re-run with --yes to apply.',
+        },
+        true,
+      )
+    } else {
+      console.log(`Record outcome preview: ${outcome} (project=${project}, agent=${agentId})`)
+      if (task) console.log(`  task: ${task}`)
+      if (summary) console.log(`  summary: ${summary}`)
+      if (existing) {
+        const nextTasks = (existing.taskCount || 0) + 1
+        const nextSuccess = (existing.successCount || 0) + (outcome === 'success' ? 1 : 0)
+        console.log(
+          `  will update agent (taskCount ${existing.taskCount || 0} -> ${nextTasks}, successCount ${existing.successCount || 0} -> ${nextSuccess})`,
+        )
+      } else {
+        console.log(`  agent not registered: trace only, agent registry untouched`)
+        console.log(`  register it with: mio agents register --agent-id ${agentId} --yes`)
+      }
+      console.log('Re-run with --yes to apply.')
+    }
+    process.exitCode = 1
+    return
+  }
+
+  let result
+  try {
+    result = cliTaskStore().recordTaskOutcome({
+      outcome,
+      agentId,
+      project,
+      task,
+      summary,
+      verification,
+      traceId,
+    })
+  } catch (error) {
+    console.error(error.message || error)
+    process.exitCode = 1
+    return
+  }
+
+  if (useJson) return jsonOrText(result, true)
+  console.log(`Recorded ${result.event.outcome} for ${result.event.agent} (project=${result.event.project})`)
+  console.log(`  trace: ${result.event.trace_id}`)
+  console.log(`  agent registry: ${result.agentUpdated ? 'updated' : 'not updated (agent not registered)'}`)
+  if (result.autoClaims) console.log(`  auto-claims: ${result.autoClaims.length}`)
+}
+
+function taskCommand(args, useJson) {
+  const sub = args[1]
+  if (!sub || sub === 'help' || sub === '--help' || sub === '-h') {
+    console.log(`Usage:
+  mio task route "<task>"     Which verified experiences apply to this task
+  mio task route "<task>" --json    Machine-readable (same shape as mio.task.route)
+
+Options:
+  --project name     Project filter (defaults to the current directory name)
+  --scope all|project|global   Memory scope to consider
+  --limit N          Max routes (1-10, default 5)
+
+This is read-only: unlike the MCP call it does not write to the query log.`)
+    if (!sub) process.exitCode = 1
+    return
+  }
+  if (sub === 'record-outcome' || sub === 'record_outcome') {
+    return recordOutcomeCommand(args, useJson)
+  }
+
+  if (sub !== 'route') {
+    console.error(`Unknown task subcommand: ${sub}`)
+    process.exitCode = 1
+    return
+  }
+
+  const flags = args.slice(2)
+  // The task is free text and may be quoted as one argument or passed as
+  // several words. Only flags this command understands are treated as options.
+  const TASK_KNOWN_FLAGS = new Set(['--project', '--scope', '--limit', '--json'])
+  const taskParts = []
+  for (let i = 0; i < flags.length; i += 1) {
+    const token = flags[i]
+    if (TASK_KNOWN_FLAGS.has(token)) continue
+    if (i > 0 && TASK_KNOWN_FLAGS.has(flags[i - 1])) continue
+    taskParts.push(token)
+  }
+  const task = taskParts.join(' ').trim()
+  if (!task) {
+    console.error('mio task route requires a task, e.g. mio task route "publish the npm package"')
+    process.exitCode = 1
+    return
+  }
+
+  let result
+  try {
+    result = cliTaskStore().routeTask({
+      task,
+      project: optionValue(flags, '--project'),
+      scope: optionValue(flags, '--scope'),
+      limit: parseNumberOption(flags, '--limit'),
+      persistQuery: false,
+    })
+  } catch (error) {
+    console.error(error.message || error)
+    process.exitCode = 1
+    return
+  }
+
+  if (useJson) return jsonOrText(result, true)
+  printTaskRoute(result)
+}
+
+function printTaskRoute(result) {
+  console.log(`Task route: "${result.task}" (project=${result.project || 'all'}, scope=${result.scope})`)
+  console.log(`verified routes: ${result.count}`)
+
+  if (result.count > 0) {
+    console.log('')
+    console.log('Routes (apply the top match first):')
+    result.routes.forEach((route, index) => {
+      const via = route.sourceAgents.length > 0 ? route.sourceAgents.join(', ') : '?'
+      const to = route.targetAgents.length > 0 ? route.targetAgents.join(', ') : '?'
+      console.log(
+        `${index + 1}. [score ${route.score}] ${route.experienceId} — reused ${route.reuseCount}x${route.confirmed ? ' (confirmed)' : ''}`,
+      )
+      console.log(`   ${String(route.memory.content).replace(/\s+/g, ' ').slice(0, 110)}`)
+      console.log(`   ${via} -> ${to}`)
+    })
+  }
+
+  if (result.relatedMemories.length > 0) {
+    console.log('')
+    console.log('Related memories:')
+    for (const record of result.relatedMemories) {
+      console.log(`- ${record.id}  ${String(record.content).replace(/\s+/g, ' ').slice(0, 100)}`)
+    }
+  }
+
+  if (result.summary && result.summary.routingSignal) {
+    console.log(`\nAgent health: ${result.summary.routingSignal}`)
+  }
+  if (result.summary && result.summary.suggestion) {
+    console.log(`\n${result.summary.suggestion}`)
   }
 }
 
@@ -656,8 +2445,12 @@ Usage:
   mio mcp                     Start Mio MCP server (stdio)
   mio install <host>          Install Mio into a host (codex|opencode|workbuddy|hermes|claude)
   mio status                  Show runtime and adapter status
-  mio agents                  List installed agents
+  mio agents                  List installed host adapters
+  mio agents list             List observed agents (from agents.jsonl, --project X)
+  mio agents report           Report per-agent task/memory/reuse telemetry (--agent X, --project Y)
+  mio agents register --agent-id X   Register an observed agent (--yes to apply; previews by default)
   mio evolution status        Show composed evolution module health
+  mio evolution report        Cross-agent evolution report: ecosystem, agents, memory health, suggestions (--period 24h|7d|30d|all)
   mio evolution shadow record      Record a shadow comparison sample
   mio evolution dual-write record  Record a dual-write comparison sample
   mio evolution cutover readiness   Assess shadow/dual-write cutover readiness
@@ -669,6 +2462,32 @@ Usage:
   mio recall "<query>"        Search Mio memory from the terminal (same ranking as mio.memory.query)
   mio traces                  Show recent observer traces (--type/--outcome/--agent/--since/--limit/--compact)
   mio remember "<content>"    Write a memory record from the terminal (same schema as mio.memory.record)
+  mio memory analyze          Report duplicates, low-quality records and kind histogram
+  mio memory archive --ids a,b    Archive records (soft delete; --yes required to apply)
+  mio memory forget --ids a,b     PERMANENTLY delete records (--yes required; not undoable)
+  mio memory restore --ids a,b    Un-archive previously archived records
+  mio memory merge --ids a,b  Merge duplicates into one survivor (--yes required to apply)
+  mio memory migrate --ids a,b --scope global|project   Move records between layers
+  mio experience list        List experience reuse (--status pending|confirmed|verified)
+  mio experience confirm --ids a,b   Confirm auto-claimed reuse (bulk supported)
+  mio experience reuse --source-agent A --target-agent B --experience-id X   Record a reuse
+  mio creativity status        Show creativity hypothesis counts and recent top ideas
+  mio creativity list          List creativity hypotheses (--status active|validated|rejected|draft, --limit N)
+  mio creativity generate      Generate hypotheses from 2+ --source "name|content" (calls an LLM)
+  mio creativity ferment       Review and refine active hypotheses (calls an LLM)
+  mio insight status           Insight counts: total, reported, unreported, high-value
+  mio insight list             List insights (--unreported, --min-score N, --detector X, --limit N)
+  mio insight mark-reported    Mark insights as reported (--ids a,b)
+  mio observer <view>          Observer pipeline views (research pipeline, not the observe daemon):
+                               status | world-model | trends | research | insights | essays | dag
+  mio observer ingest --trace-id T --event-type E   Record a trace event (--payload/--outcome)
+  mio observer subscribe --event-types a,b          Subscribe to events (--yes to apply)
+  mio observer digest                               New events since the last digest (advances cursor)
+  mio phase0 report            Show the Phase 0 validation report (--project X, --format markdown)
+  mio host capabilities        Show what each host supports and whether it is installed
+  mio task route "<task>"      Which verified experiences apply to this task (--project/--scope/--limit)
+  mio task record-outcome --outcome success   Record a task outcome (--yes to apply; previews by default)
+  mio policy check "<action>" Check historical risk for an action before running it
   mio prune --days 30         Trim old traces/queries/reuse records and observe.log (--dry-run to preview; --memory needs --yes)
   mio digest --days 7         Aggregate traces/memory/reuse into an actionable report (--write-back feeds agent context files; --json)
   mio --json status           Machine-readable status
@@ -694,7 +2513,7 @@ async function main() {
     case 'status':
       return statusCommand(useJson)
     case 'agents':
-      return agentsCommand(useJson)
+      return agentsCommand(args, useJson)
     case 'observe':
       return observeCommand(args)
     case 'recall':
@@ -703,6 +2522,24 @@ async function main() {
       return tracesCommand(args, useJson)
     case 'remember':
       return rememberCommand(args, useJson)
+    case 'memory':
+      return memoryCommand(args, useJson)
+    case 'experience':
+      return experienceCommand(args, useJson)
+    case 'policy':
+      return policyCommand(args, useJson)
+    case 'creativity':
+      return creativityCommand(args, useJson)
+    case 'insight':
+      return insightCommand(args, useJson)
+    case 'observer':
+      return observerCommand(args, useJson)
+    case 'phase0':
+      return phase0Command(args, useJson)
+    case 'host':
+      return hostCommand(args, useJson)
+    case 'task':
+      return taskCommand(args, useJson)
     case 'prune':
       return pruneCommand(args, useJson)
     case 'digest':