@ran-sh/dsh-crew 0.5.2 → 0.5.4

package/windows/start-dsh-crew.ps1

@@ -0,0 +1,326 @@
+# DSH Crew managed Windows launcher
+[CmdletBinding()]
+param(
+  [ValidateSet('background', 'open', 'watch')]
+  [string] $Mode = 'open'
+)
+
+Set-StrictMode -Version Latest
+$ErrorActionPreference = 'Stop'
+
+$crewHome = Join-Path $env:USERPROFILE '.config\dsh-crew\harness'
+$officialHome = Join-Path $env:USERPROFILE '.dsh'
+$dshCli = Join-Path $crewHome 'runtime\node_modules\.bin\dsh.cmd'
+$logRoot = if ($env:TEMP) { $env:TEMP } else { [System.IO.Path]::GetTempPath() }
+$launcherLog = Join-Path $logRoot 'dsh-crew-launcher.log'
+$startedAt = Get-Date
+$services = @(
+  [pscustomobject]@{ Name = 'Crew backend'; Profile = 'dsh-crew'; Home = $crewHome; Port = 3210; Url = 'http://127.0.0.1:3210'; State = 'pending'; Process = $null; RootPid = $null; RootStartedAtUtcTicks = $null; ListenerPid = $null; ListenerStartedAtUtcTicks = $null; ConsecutiveFailures = 0; LastError = $null },
+  [pscustomobject]@{ Name = 'Official UI'; Profile = 'web'; Home = $officialHome; Port = 3080; Url = 'http://127.0.0.1:3080'; State = 'pending'; Process = $null; RootPid = $null; RootStartedAtUtcTicks = $null; ListenerPid = $null; ListenerStartedAtUtcTicks = $null; ConsecutiveFailures = 0; LastError = $null }
+)
+
+function Write-LaunchLog {
+  param([string] $Message, [ValidateSet('INFO', 'WARN', 'ERROR')] [string] $Level = 'INFO')
+  $line = '[{0}] [{1}] {2}' -f (Get-Date -Format 'yyyy-MM-ddTHH:mm:ss.fffK'), $Level, $Message
+  Add-Content -LiteralPath $launcherLog -Value $line -Encoding UTF8
+  if ($Mode -eq 'open') {
+    if ($Level -eq 'ERROR') { Write-Host $line -ForegroundColor Red }
+    elseif ($Level -eq 'WARN') { Write-Host $line -ForegroundColor Yellow }
+    else { Write-Host $line }
+  }
+}
+
+function Get-HealthState {
+  param([pscustomobject] $Service)
+  try {
+    $response = Invoke-RestMethod -Uri ($Service.Url + '/_dsh/dsh-crew/extension') -TimeoutSec 2
+    $version = $response.extension.runtime.runtime_version
+    if ($response.ok -eq $true -and $version) {
+      return [pscustomobject]@{ Ready = $true; Version = [string] $version; Error = $null }
+    }
+    return [pscustomobject]@{ Ready = $false; Version = $null; Error = 'Response did not contain a ready runtime contract.' }
+  } catch {
+    return [pscustomobject]@{ Ready = $false; Version = $null; Error = $_.Exception.Message }
+  }
+}
+
+function Get-PortState {
+  param([int] $Port)
+  try {
+    $listeners = [System.Net.NetworkInformation.IPGlobalProperties]::GetIPGlobalProperties().GetActiveTcpListeners()
+    $occupied = @($listeners | Where-Object Port -eq $Port).Count -gt 0
+    if (-not $occupied) {
+      return [pscustomobject]@{ State = 'free'; Error = $null; Pid = $null }
+    }
+
+    $ownerPid = $null
+    try {
+      $ownerPid = (Get-NetTCPConnection -State Listen -LocalPort $Port -ErrorAction Stop | Select-Object -First 1).OwningProcess
+    } catch { }
+    return [pscustomobject]@{ State = 'occupied'; Error = $null; Pid = $ownerPid }
+  } catch {
+    return [pscustomobject]@{ State = 'unknown'; Error = ('Listener enumeration failed: {0}' -f $_.Exception.Message); Pid = $null }
+  }
+}
+
+function Test-TrackedProcessIdentity {
+  param([int] $ProcessId, [long] $ExpectedStartTicks, [object[]] $ProcessTable)
+  if ($ProcessId -lt 1 -or $ExpectedStartTicks -lt 1) { return $false }
+  $record = @($ProcessTable | Where-Object { [int] $_.ProcessId -eq $ProcessId } | Select-Object -First 1)
+  if ($record.Count -eq 0) { return $false }
+  $startTicks = $null
+  if ($record[0].PSObject.Properties.Name -contains 'StartTicks') {
+    $startTicks = [long] $record[0].StartTicks
+  } else {
+    try { $startTicks = (Get-Process -Id $ProcessId -ErrorAction Stop).StartTime.ToUniversalTime().Ticks } catch { return $false }
+  }
+  return $startTicks -eq $ExpectedStartTicks
+}
+
+function Get-TrackedProcessTree {
+  param([pscustomobject] $Service, [object[]] $ProcessTable = $null)
+  $processes = if ($null -ne $ProcessTable) { @($ProcessTable) } else { @(Get-CimInstance Win32_Process -ErrorAction Stop) }
+  $owned = [System.Collections.Generic.HashSet[int]]::new()
+  $rootMatches = Test-TrackedProcessIdentity -ProcessId $Service.RootPid -ExpectedStartTicks $Service.RootStartedAtUtcTicks -ProcessTable $processes
+  $hasTrackedListener = $Service.ListenerPid -and $Service.ListenerStartedAtUtcTicks
+  if ($hasTrackedListener) {
+    $listenerMatches = Test-TrackedProcessIdentity -ProcessId $Service.ListenerPid -ExpectedStartTicks $Service.ListenerStartedAtUtcTicks -ProcessTable $processes
+    if (-not $listenerMatches) { return @() }
+    [void] $owned.Add([int] $Service.ListenerPid)
+  } elseif (-not $rootMatches) {
+    return @()
+  }
+  if ($rootMatches) { [void] $owned.Add([int] $Service.RootPid) }
+  do {
+    $added = $false
+    foreach ($candidate in $processes) {
+      $candidateId = [int] $candidate.ProcessId
+      $parentId = [int] $candidate.ParentProcessId
+      if ($owned.Contains($parentId) -and $owned.Add($candidateId)) { $added = $true }
+    }
+  } while ($added)
+  return @($owned | ForEach-Object { [int] $_ })
+}
+
+function Set-TrackedListenerIdentity {
+  param([pscustomobject] $Service)
+  $port = Get-PortState $Service.Port
+  if ($port.State -ne 'occupied' -or -not $port.Pid) { return $false }
+  $tree = @(Get-TrackedProcessTree -Service $Service)
+  if ($port.Pid -notin $tree) { return $false }
+  try {
+    $listener = Get-Process -Id $port.Pid -ErrorAction Stop
+    $Service.ListenerPid = [int] $port.Pid
+    $Service.ListenerStartedAtUtcTicks = $listener.StartTime.ToUniversalTime().Ticks
+    return $true
+  } catch {
+    return $false
+  }
+}
+
+function Stop-OwnedListener {
+  param([pscustomobject] $Service, [int] $ListenerPid)
+  if ($Mode -ne 'watch' -or -not $Service.ListenerPid -or $ListenerPid -ne $Service.ListenerPid) {
+    return $false
+  }
+  try {
+    $ownedProcessIds = @(Get-TrackedProcessTree -Service $Service)
+    if ($ListenerPid -notin $ownedProcessIds) { return $false }
+    Write-LaunchLog ('Supervisor confirmed owned listener PID={0} under tracked root PID={1}; stopping that process tree after {2} consecutive failed health checks.' -f $ListenerPid, $Service.RootPid, $Service.ConsecutiveFailures) 'WARN'
+    foreach ($processId in $ownedProcessIds) {
+      Stop-Process -Id $processId -Force -ErrorAction SilentlyContinue
+    }
+    $deadline = (Get-Date).AddSeconds(5)
+    do {
+      $port = Get-PortState $Service.Port
+      if ($port.State -eq 'free') { return $true }
+      Start-Sleep -Milliseconds 250
+    } while ((Get-Date) -lt $deadline)
+  } catch {
+    Write-LaunchLog ('Supervisor could not stop its owned listener safely: {0}' -f $_.Exception.Message) 'WARN'
+  }
+  return $false
+}
+
+function Start-CrewService {
+  param([pscustomobject] $Service)
+  $serviceRunStamp = (Get-Date).ToString('yyyyMMdd-HHmmssfff')
+  $stdout = Join-Path $logRoot ('dsh-crew-{0}-{1}-{2}.out.log' -f $Service.Profile, $Service.Port, $serviceRunStamp)
+  $stderr = Join-Path $logRoot ('dsh-crew-{0}-{1}-{2}.err.log' -f $Service.Profile, $Service.Port, $serviceRunStamp)
+  $previousHome = $env:DSH_HOME
+  try {
+    $env:DSH_HOME = $Service.Home
+    $arguments = @('--profile', $Service.Profile, '--host', '127.0.0.1', '--port', [string] $Service.Port, '--no-open')
+    $process = Start-Process -FilePath $dshCli -ArgumentList $arguments -WindowStyle Hidden -PassThru `
+      -RedirectStandardOutput $stdout -RedirectStandardError $stderr
+    $Service.Process = $process
+    $Service.RootPid = $process.Id
+    $Service.RootStartedAtUtcTicks = $process.StartTime.ToUniversalTime().Ticks
+    $Service.ListenerPid = $null
+    $Service.ListenerStartedAtUtcTicks = $null
+    $Service.ConsecutiveFailures = 0
+    $Service.State = 'starting'
+    Write-LaunchLog ('Started {0} on port {1}; PID={2}; stdout={3}; stderr={4}' -f $Service.Profile, $Service.Port, $process.Id, $stdout, $stderr)
+  } finally {
+    $env:DSH_HOME = $previousHome
+  }
+}
+
+function Wait-CrewServices {
+  $deadline = (Get-Date).AddSeconds(90)
+  while (@($services | Where-Object State -eq 'starting').Count -gt 0 -and (Get-Date) -lt $deadline) {
+    foreach ($service in ($services | Where-Object State -eq 'starting')) {
+      $health = Get-HealthState $service
+      $service.LastError = $health.Error
+      if ($health.Ready) {
+        $service.State = 'ready'
+        $service.ConsecutiveFailures = 0
+        [void] (Set-TrackedListenerIdentity -Service $service)
+        Write-LaunchLog ('{0} ready on {1}; runtime={2}' -f $service.Name, $service.Port, $health.Version)
+      } elseif ($service.Process -and $service.Process.HasExited) {
+        throw ('{0} exited before becoming ready; PID={1}; exit={2}; last health error: {3}' -f $service.Name, $service.Process.Id, $service.Process.ExitCode, $health.Error)
+      }
+    }
+    if (@($services | Where-Object State -eq 'starting').Count -gt 0) { Start-Sleep -Milliseconds 500 }
+  }
+
+  $notReady = @($services | Where-Object State -ne 'ready')
+  if ($notReady.Count -gt 0) {
+    $details = ($notReady | ForEach-Object { '{0}:{1} ({2})' -f $_.Profile, $_.Port, $_.LastError }) -join '; '
+    throw "Startup health deadline exceeded: $details"
+  }
+}
+
+function Ensure-CrewServices {
+  param([switch] $QuietHealthy)
+
+  foreach ($service in $services) {
+    $health = Get-HealthState $service
+    if ($health.Ready) {
+      $wasReady = $service.State -eq 'ready'
+      $service.State = 'ready'
+      $service.ConsecutiveFailures = 0
+      $service.LastError = $null
+      if (-not $service.ListenerPid) { [void] (Set-TrackedListenerIdentity -Service $service) }
+      if (-not $QuietHealthy -or -not $wasReady) {
+        Write-LaunchLog ('{0} already ready on {1}; runtime={2}' -f $service.Name, $service.Port, $health.Version)
+      }
+      continue
+    }
+
+    $service.LastError = $health.Error
+    $service.ConsecutiveFailures += 1
+    if ($service.Process -and $service.Process.HasExited) {
+      Write-LaunchLog ('{0} process exited after startup; PID={1}; exit={2}' -f $service.Name, $service.Process.Id, $service.Process.ExitCode) 'WARN'
+      $service.Process = $null
+    }
+    $service.State = 'pending'
+
+    $port = Get-PortState $service.Port
+    if ($port.State -eq 'occupied') {
+      if ($service.ConsecutiveFailures -lt 3) {
+        throw ('Health check {0}/3 failed for {1}; owned process remains untouched until failure is confirmed. Health error: {2}' -f $service.ConsecutiveFailures, $service.Name, $health.Error)
+      }
+      if (Stop-OwnedListener -Service $service -ListenerPid $port.Pid) {
+        $service.Process = $null
+        $service.RootPid = $null
+        $service.RootStartedAtUtcTicks = $null
+        $service.ListenerPid = $null
+        $service.ListenerStartedAtUtcTicks = $null
+        $service.ConsecutiveFailures = 0
+        $port = Get-PortState $service.Port
+      }
+    }
+    if ($port.State -eq 'occupied') {
+      $owner = if ($port.Pid) { "; listener PID=$($port.Pid)" } else { '' }
+      throw ('Port {0} is occupied, but {1} failed its health contract{2}. Health error: {3}' -f $service.Port, $service.Name, $owner, $health.Error)
+    }
+    if ($port.State -ne 'free') {
+      throw ('Could not determine whether port {0} is available: {1}' -f $service.Port, $port.Error)
+    }
+    if ($Mode -eq 'watch') {
+      Write-LaunchLog ('Supervisor detected {0} unavailable on {1}; restarting it.' -f $service.Name, $service.Port) 'WARN'
+    }
+    Start-CrewService $service
+  }
+
+  Wait-CrewServices
+}
+
+function Start-ServiceSupervisor {
+  $mutex = [System.Threading.Mutex]::new($false, 'Local\DSHCrewServiceSupervisor')
+  $ownsMutex = $false
+  try {
+    try {
+      $ownsMutex = $mutex.WaitOne(0)
+    } catch [System.Threading.AbandonedMutexException] {
+      $ownsMutex = $true
+    }
+    if (-not $ownsMutex) {
+      Write-LaunchLog 'Supervisor already active; duplicate watcher exiting.'
+      return
+    }
+
+    Write-LaunchLog 'Supervisor active; monitoring 3080 and 3210 every 10 seconds.'
+    $lastRecoveryError = $null
+    while ($true) {
+      try {
+        Ensure-CrewServices -QuietHealthy
+        if ($lastRecoveryError) {
+          Write-LaunchLog 'Supervisor recovery succeeded; both services are healthy.'
+          $lastRecoveryError = $null
+        }
+      } catch {
+        $recoveryError = $_.Exception.Message
+        if ($recoveryError -ne $lastRecoveryError) {
+          Write-LaunchLog ('Supervisor recovery failed; will retry: {0}' -f $recoveryError) 'WARN'
+          $lastRecoveryError = $recoveryError
+        }
+      }
+      Start-Sleep -Seconds 10
+    }
+  } finally {
+    if ($ownsMutex) { $mutex.ReleaseMutex() }
+    $mutex.Dispose()
+  }
+}
+
+if ($env:DSH_CREW_LAUNCHER_TEST_IMPORT -eq '1') { return }
+
+try {
+  New-Item -ItemType Directory -Path $logRoot -Force | Out-Null
+  Write-LaunchLog ('Launcher started; mode={0}; user={1}' -f $Mode, $env:USERNAME)
+
+  if (-not (Test-Path -LiteralPath $dshCli -PathType Leaf)) {
+    throw "DSH CLI was not found at $dshCli. Run: npm install -g @ran-sh/dsh-crew@latest; dsh-crew update"
+  }
+  if (-not (Test-Path -LiteralPath (Join-Path $crewHome 'profiles\dsh-crew\package.json') -PathType Leaf)) {
+    throw "The isolated dsh-crew profile is missing under $crewHome. Run: dsh-crew update"
+  }
+  if (-not (Test-Path -LiteralPath (Join-Path $officialHome 'profiles\web\package.json') -PathType Leaf)) {
+    throw "The official web profile is missing under $officialHome. Run: dsh-crew integrate"
+  }
+
+  if ($Mode -eq 'watch') {
+    Start-ServiceSupervisor
+    Write-LaunchLog 'Supervisor stopped.' 'WARN'
+    exit 0
+  }
+
+  Ensure-CrewServices
+
+  if ($Mode -eq 'open') {
+    Start-Process 'http://127.0.0.1:3080/' | Out-Null
+    Write-LaunchLog 'Opened daily console at http://127.0.0.1:3080/'
+  }
+  Write-LaunchLog ('Launcher completed successfully in {0:n1}s.' -f ((Get-Date) - $startedAt).TotalSeconds)
+  exit 0
+} catch {
+  Write-LaunchLog $_.Exception.Message 'ERROR'
+  if ($Mode -eq 'open') {
+    Write-Host ''
+    Write-Host "Diagnostic log: $launcherLog" -ForegroundColor Yellow
+    Write-Host 'Startup failed. Review the diagnostic log and the service stdout/stderr paths recorded above.' -ForegroundColor Yellow
+  }
+  exit 1
+}

package/windows/start-dsh-crew.vbs

@@ -4,6 +4,6 @@ Dim shell, launcher, command
 Set shell = CreateObject("WScript.Shell")
 launcher = "__LAUNCHER__"
 command = shell.ExpandEnvironmentStrings("%COMSPEC%") & " /d /c " & _
-  Chr(34) & Chr(34) & launcher & Chr(34) & Chr(34)
+  Chr(34) & Chr(34) & launcher & Chr(34) & " --watch" & Chr(34)
 
 shell.Run command, 0, False

package/zcode/AGENTS.md

@@ -10,6 +10,11 @@ work only after its live capability and readiness surfaces have been checked.
   repository/worktree and Result Contract boundaries. Keep planning, ambiguous
   requirements, integration, external side effects and final communication in
   the host agent.
+- Dispatch selected work asynchronously with `dsh_spawn_worker`, save its
+  workflow ID, and poll that same workflow with `dsh_worker_result` using
+  `wait_seconds: 10`. A bounded wait that returns a nonterminal state is not a
+  failure when `dsh_worker_status` confirms that workflow is still running;
+  continue polling it and never start a duplicate workflow.
 - If Crew is selected and any required capability is unavailable, non-callable,
   or returns an unknown/runtime/configuration/credential/routing/timeout error,
   pause. Report the evidence and wait for the operator to choose repair Crew or

package/zcode/agents/ds-reviewer.md

@@ -5,6 +5,7 @@ mcpServers:
   - dsh-crew
 tools:
   - mcp__dsh-crew__dsh_run_worker
+  - mcp__dsh-crew__dsh_spawn_worker
   - mcp__dsh-crew__dsh_worker_status
   - mcp__dsh-crew__dsh_worker_result
   - mcp__dsh-crew__dsh_worker_cancel
@@ -13,7 +14,23 @@ tools:
 
 You are a thin, read-only dispatcher. Never edit files or implement fixes.
 
-Pass the review request verbatim to `dsh_run_worker` with role `reviewer` and
-the current workspace as `cwd`; omit effort unless explicitly requested. Wait
-for the result and return its Review Findings, Evidence, Risks and Verdict.
-Treat failing tests or incomplete evidence as not approved.
+Check `dsh_worker_config` before dispatch. Derive one bounded, read-only review
+task that contains only the code, acceptance criteria and validation evidence
+the Reviewer must inspect. Keep harness-reporting requirements in this
+dispatcher, then call `dsh_spawn_worker` with role `reviewer` and the current
+workspace as `cwd`; omit effort unless explicitly requested. Save the returned
+workflow ID.
+
+Poll that same workflow through `dsh_worker_result` with `wait_seconds: 10`.
+If a bounded result wait expires or returns a nonterminal state, check
+`dsh_worker_status` once. When it confirms the workflow is still running,
+continue polling the same workflow; never start a duplicate. A genuine
+transport, runtime, configuration, credential or routing error is a hard stop
+and must be reported to the operator.
+
+Treat the workflow ID as host structured-result metadata; never ask the
+Reviewer to discover or report its workflow ID, provider, model or other
+harness metadata. Do not forward workflow ID, provider, model or status-reporting
+fields as Reviewer task requirements. Combine the host-owned metadata with
+Review Findings, Evidence, Risks and Verdict. Treat failing tests or incomplete
+evidence as not approved.

package/zcode/agents/ds-worker.md

@@ -14,7 +14,26 @@ tools:
 
 You are a thin dispatcher. You never do the task yourself.
 
-Pass the task verbatim to `dsh_run_worker` with role `worker` and the current
-workspace as `cwd`; omit effort unless the task explicitly requests it. Wait
-for the result. If it is `done`, return the result and its evidence footer. If
-it is not done, report the error and stop reason clearly and stop.
+Check `dsh_worker_config` before dispatch. Derive one bounded, domain-only task
+for the Worker, preserving the requested implementation and validation scope.
+Keep harness-reporting requirements in this dispatcher, then call
+`dsh_spawn_worker` with role `worker` and the current workspace as `cwd`; omit
+effort unless the task explicitly requests it. Save the returned workflow ID.
+When the operator explicitly requests read-only search or analysis that should
+make zero file changes, pass `constraints: { allow_no_changes: true }`.
+Otherwise omit that constraint so implementation work still fails closed when
+the reported diff and isolated workspace disagree.
+
+Poll that same workflow through `dsh_worker_result` with `wait_seconds: 10`.
+If a bounded result wait expires or returns a nonterminal state, check
+`dsh_worker_status` once. When it confirms the workflow is still running,
+continue polling the same workflow; never start a duplicate. A genuine
+transport, runtime, configuration, credential or routing error is a hard stop
+and must be reported to the operator.
+
+Treat the workflow ID as host structured-result metadata; never ask the Worker
+to discover or report its workflow ID, provider, model or other harness
+metadata. Do not forward workflow ID, provider, model or status-reporting fields
+as Worker task requirements. If the workflow is `done`, combine the host-owned
+metadata with its compact result and evidence footer. Otherwise report the
+terminal error and stop reason clearly.