autark-cli 0.5.1 → 0.5.3

package/autark.mjs

@@ -320,7 +320,7 @@ function settingsUsage() {
 //      via pnpm. Re-exec autark if its own version moved.
 //   1. GET /v1/me — compare runtime_version + agentmail_email against local.
 //   2. If runtime_version differs: curl the canonical files from the repo,
-//      refresh the 6 kstack skills via pnpm dlx, reload launchd, stamp the
+//      refresh the 6 kstack skills via pnpm dlx, reload the local scheduler, stamp the
 //      new version into ~/.autark/runtime/version.txt.
 //   3. If agentmail_email differs: POST /v1/inbox/rotate to re-mint a fresh
 //      inbox-scoped key, rewrite credentials.json.
@@ -413,8 +413,8 @@ async function update(rest) {
       await refreshRuntimeFiles()
       log('refreshing kstack skills…')
       await refreshSkills()
-      log('reloading launchd…')
-      await reloadLaunchd()
+      log('reloading local scheduler…')
+      await reloadLocalScheduler()
       // Only stamp the version AFTER every step succeeded.
       writeLocalVersion(me.runtime_version)
       sub(`stamped ~/.autark/runtime/version.txt → ${me.runtime_version}`)
@@ -512,13 +512,13 @@ async function refreshRuntimeFiles() {
   const agentSh = path.join(AUTARK_HOME, 'agent.sh')
   await curlFile(`${RAW_RUNTIME_BASE}/agent.sh`, agentSh)
   fs.chmodSync(agentSh, 0o755)
+  const agentPs1 = path.join(AUTARK_HOME, 'agent.ps1')
+  await curlFile(`${RAW_RUNTIME_BASE}/agent.ps1`, agentPs1)
 }
 
 async function refreshSkills() {
-  for (const name of SKILL_NAMES) {
-    const res = await spawnSync('pnpm', ['dlx', '--silent', 'skills', 'add', `kiluazen/skills@${name}`, '-y', '-g'])
-    if (res.status !== 0) throw new Error(`skill ${name} install failed`)
-  }
+  const res = await spawnSync('pnpm', ['dlx', '--silent', 'skills', 'add', 'kiluazen/kstack', '-s', SKILL_NAMES.join(','), '-y', '-g'])
+  if (res.status !== 0) throw new Error(`kstack skills install failed`)
 }
 
 async function reloadLaunchd() {
@@ -549,6 +549,29 @@ async function reloadLaunchd() {
   await spawnSync('launchctl', ['load', plist])
 }
 
+async function reloadWindowsTask() {
+  if (process.platform !== 'win32') return
+  const canonicalHome = path.join(os.homedir(), '.autark')
+  if (path.resolve(AUTARK_HOME) !== path.resolve(canonicalHome)) {
+    return
+  }
+  const agentPs1 = path.join(AUTARK_HOME, 'agent.ps1')
+  if (!fs.existsSync(agentPs1)) return
+  const taskName = 'Autark Runner'
+  const ps = [
+    `$action = New-ScheduledTaskAction -Execute 'powershell.exe' -Argument '-NoProfile -ExecutionPolicy Bypass -File "${agentPs1.replaceAll('"', '`"')}"'`,
+    `$morning = New-ScheduledTaskTrigger -Daily -At 9am`,
+    `$evening = New-ScheduledTaskTrigger -Daily -At 7pm`,
+    `Register-ScheduledTask -TaskName '${taskName}' -Action $action -Trigger @($morning, $evening) -Description 'Autark daily runner' -Force | Out-Null`,
+  ].join('; ')
+  await spawnSync('powershell.exe', ['-NoProfile', '-ExecutionPolicy', 'Bypass', '-Command', ps])
+}
+
+async function reloadLocalScheduler() {
+  await reloadLaunchd()
+  await reloadWindowsTask()
+}
+
 // Promise-wrapped spawn — returns { status: number }. stdio inherited so the
 // user sees pnpm's output. Used for self-update + skills refresh.
 function spawnSync(cmd, args) {
@@ -925,6 +948,7 @@ async function mailSend(rest) {
     subject,
     response: result,
   })
+  noteAutoLog(action, opts)
   printJson({ ...result, autark_action_id: action?.id })
 }
 
@@ -942,6 +966,7 @@ async function mailReply(rest, mode) {
     response: result,
     metadata: { message_id: messageId },
   })
+  noteAutoLog(action, opts)
   printJson({ ...result, autark_action_id: action?.id })
 }
 
@@ -959,9 +984,20 @@ async function mailForward(rest) {
     response: result,
     metadata: { message_id: messageId },
   })
+  noteAutoLog(action, opts)
   printJson({ ...result, autark_action_id: action?.id })
 }
 
+// Explicit human-readable cue so agents (and humans) see that the action
+// landed in autark and don't follow up with a redundant `autark log action`
+// for the same send. Stays on stderr so JSON consumers piping stdout aren't
+// affected. Only fires when --run-id was passed and the log succeeded.
+function noteAutoLog(action, opts) {
+  if (!action?.id) return
+  const runId = opts['run-id'] || opts.run_id || opts.run
+  console.error(`autark: auto-logged action ${action.id} (run ${runId}) — do NOT run \`autark log action\` for this`)
+}
+
 async function mailThreads(rest) {
   const opts = parseArgs(rest)
   const creds = requireAgentmailCredentials()
@@ -1126,6 +1162,21 @@ function actionShortId(id) {
   return String(id || '').slice(0, 8)
 }
 
+// Aggregate sends + replies per channel. The cron writes inbound_count onto
+// each action; agents use this to score hypotheses without re-walking threads.
+// Channels are kept separate (email vs github vs reddit vs hn vs substack) —
+// a public comment reply is a different signal class than an email reply.
+function channelBreakdown(actions) {
+  const by = {}
+  for (const a of (actions || [])) {
+    const ch = a.channel || 'unknown'
+    if (!by[ch]) by[ch] = { sent: 0, replied: 0 }
+    by[ch].sent += 1
+    if ((a.inbound_count || 0) > 0) by[ch].replied += 1
+  }
+  return by
+}
+
 function printProductContext(r) {
   // ---- structured key-block ----
   kv('product.id', r.product.id)
@@ -1140,6 +1191,15 @@ function printProductContext(r) {
     kv(`hypothesis.${h.code}.title`, h.title)
     kv(`hypothesis.${h.code}.status`, h.status)
     kv(`hypothesis.${h.code}.run_count`, h.run_count)
+    kv(`hypothesis.${h.code}.action_count`, h.action_count)
+    kv(`hypothesis.${h.code}.reply_count`, h.reply_count)
+    if (h.needs_human_count) kv(`hypothesis.${h.code}.needs_human_count`, h.needs_human_count)
+    const actions = (h.runs || []).flatMap((run) => run.actions || [])
+    const by = channelBreakdown(actions)
+    for (const ch of Object.keys(by).sort()) {
+      kv(`hypothesis.${h.code}.channel.${ch}.sent`, by[ch].sent)
+      kv(`hypothesis.${h.code}.channel.${ch}.replied`, by[ch].replied)
+    }
   }
   console.log('---')
   // ---- narrative ----
@@ -1152,7 +1212,13 @@ function printProductContext(r) {
   } else {
     console.log(`\n## Hypotheses (${r.hypotheses.length})\n`)
     for (const h of r.hypotheses) {
-      console.log(`- [${h.status}] ${h.code} — ${h.title}  (runs: ${h.run_count})`)
+      const actions = (h.runs || []).flatMap((run) => run.actions || [])
+      const by = channelBreakdown(actions)
+      const channelSummary = Object.keys(by).sort()
+        .map((ch) => `${ch} ${by[ch].replied}/${by[ch].sent}`)
+        .join('  ')
+      const tail = channelSummary ? `  — replies/sends by channel: ${channelSummary}` : ''
+      console.log(`- [${h.status}] ${h.code} — ${h.title}  (runs: ${h.run_count}, actions: ${h.action_count || 0}, replies: ${h.reply_count || 0})${tail}`)
     }
   }
 }
@@ -1165,6 +1231,27 @@ function printHypothesisContext(result) {
   kv('hypothesis.code', result.hypothesis.code)
   kv('hypothesis.title', result.hypothesis.title)
   kv('hypothesis.status', result.hypothesis.status)
+  // Feedback — operator nudges left on this hypothesis, oldest first. The
+  // worker returns them; previously the CLI silently dropped them, so agents
+  // never saw the operator's running corrections. Now surface them up top
+  // (before runs/actions) so the agent reads them at the start of every run.
+  for (const f of (result.feedback || [])) {
+    const shortF = actionShortId(f.id)
+    kv(`feedback.${shortF}.created_at`, f.created_at)
+    if (f.action_id) kv(`feedback.${shortF}.action_id`, f.action_id)
+    kv(`feedback.${shortF}.text`, f.text)
+  }
+  // Per-channel rollup across all runs on this hypothesis. Lets the agent
+  // grep `hypothesis.channel.*` to see what's actually landing without
+  // counting actions itself.
+  const allActions = (result.runs || []).flatMap((run) => run.actions || [])
+  kv('hypothesis.action_count', allActions.length)
+  kv('hypothesis.reply_count', allActions.filter((a) => (a.inbound_count || 0) > 0).length)
+  const channelTotals = channelBreakdown(allActions)
+  for (const ch of Object.keys(channelTotals).sort()) {
+    kv(`hypothesis.channel.${ch}.sent`, channelTotals[ch].sent)
+    kv(`hypothesis.channel.${ch}.replied`, channelTotals[ch].replied)
+  }
   for (const run of (result.runs || [])) {
     const shortRun = actionShortId(run.id)
     kv(`run.${shortRun}.id`, run.id)
@@ -1182,6 +1269,11 @@ function printHypothesisContext(result) {
       if (a.agentmail_thread_id) kv(`action.${shortA}.agentmail_thread_id`, a.agentmail_thread_id)
       if (a.agentmail_inbox_id)  kv(`action.${shortA}.agentmail_inbox_id`, a.agentmail_inbox_id)
       if (a.occurred_at)         kv(`action.${shortA}.occurred_at`, a.occurred_at)
+      kv(`action.${shortA}.inbound_count`, a.inbound_count || 0)
+      if (a.last_inbound_at)     kv(`action.${shortA}.last_inbound_at`, a.last_inbound_at)
+      if (a.reply_checked_at)    kv(`action.${shortA}.reply_checked_at`, a.reply_checked_at)
+      if (a.needs_human)         kv(`action.${shortA}.needs_human`, a.needs_human)
+      if (a.escalation_reason)   kv(`action.${shortA}.escalation_reason`, a.escalation_reason)
       // Spread well-known metadata keys to their own lines for greppability.
       const md = a.metadata || {}
       for (const [mk, mv] of Object.entries(md)) {
@@ -1195,6 +1287,14 @@ function printHypothesisContext(result) {
   // ---- narrative ----
   console.log(`# ${result.product.slug}/${result.hypothesis.code} — ${result.hypothesis.title}\n`)
   console.log(result.hypothesis.hypothesis_md)
+  if ((result.feedback || []).length > 0) {
+    console.log(`\n\n## Feedback (operator nudges, oldest first)\n`)
+    for (const f of result.feedback) {
+      const when = f.created_at || ''
+      const tag = f.action_id ? ` (on action ${f.action_id})` : ''
+      console.log(`- [${when}]${tag} ${f.text}`)
+    }
+  }
   for (const run of (result.runs || [])) {
     console.log(`\n\n## Run ${run.run_number}  (${run.id})`)
     console.log(`started: ${run.started_at}${run.finished_at ? `   finished: ${run.finished_at}` : '   (in progress)'}`)
@@ -1202,7 +1302,11 @@ function printHypothesisContext(result) {
       console.log(`\nActions:`)
       for (const a of run.actions) {
         const pointer = a.url || a.agentmail_thread_id || a.recipient || ''
-        console.log(`  [${a.channel}] ${a.title}${pointer ? `  →  ${pointer}` : ''}`)
+        const flags = []
+        if ((a.inbound_count || 0) > 0) flags.push(`replied×${a.inbound_count}`)
+        if (a.needs_human) flags.push('needs_human')
+        const flagStr = flags.length ? `  [${flags.join(', ')}]` : ''
+        console.log(`  [${a.channel}] ${a.title}${pointer ? `  →  ${pointer}` : ''}${flagStr}`)
       }
     }
     if (run.narrative_md) console.log(`\n${run.narrative_md}`)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "autark-cli",
-  "version": "0.5.1",
+  "version": "0.5.3",
   "description": "CLI for autark \u2014 hypothesis-driven product runbooks. Track products, hypotheses, runs, and actions from the terminal.",
   "type": "module",
   "license": "MIT",