Spawn diagnosis agent via tmux when deploy/rollback fails

bin/dispatch-deploy

@@ -151,6 +151,88 @@ write_failure_log() {
   echo "failure details written to $FAILURE_LOG" >&2
 }
 
+spawn_diagnosis_agent() {
+  local step="$1"
+  local tag="$2"
+  local previous_tag="${3:-unknown}"
+
+  # Generate agent ID and session name
+  local agent_id="agt_$(openssl rand -hex 6)"
+  local session_name="dispatch_${agent_id}"
+  local media_dir="/tmp/dispatch-media/${agent_id}"
+  mkdir -p "$media_dir"
+
+  local port
+  port=$(resolve_port)
+
+  # Read DISPATCH_BIN_DIR from server .env
+  local bin_dir=""
+  local env_file="$SERVER_DIR/.env"
+  if [[ -f "$env_file" ]]; then
+    bin_dir=$(grep -E '^DISPATCH_BIN_DIR=' "$env_file" | tail -1 | cut -d= -f2)
+  fi
+  if [[ -z "$bin_dir" ]]; then
+    bin_dir="$SERVER_DIR/bin"
+  fi
+
+  # Pre-register agent in PostgreSQL so reconcileAgents() picks it up
+  local db_url=""
+  if [[ -f "$env_file" ]]; then
+    db_url=$(grep -E '^DATABASE_URL=' "$env_file" | tail -1 | cut -d= -f2)
+  fi
+  if [[ -n "$db_url" ]] && command -v psql &>/dev/null; then
+    psql "$db_url" -c "
+      INSERT INTO agents (id, name, type, status, cwd, created_at, updated_at)
+      VALUES ('$agent_id', 'deploy-diagnosis', 'claude', 'running', '$SERVER_DIR',
+              NOW(), NOW())
+      ON CONFLICT DO NOTHING;
+    " >/dev/null 2>&1 || true
+  fi
+
+  # Build failure context from the failure log
+  local failure_context=""
+  if [[ -f "$FAILURE_LOG" ]]; then
+    failure_context=$(cat "$FAILURE_LOG")
+  fi
+
+  local prompt="The Dispatch deploy pipeline failed and needs diagnosis.
+
+Failed step: $step
+Target tag: $tag
+Previous tag: $previous_tag
+
+Failure details:
+${failure_context:-No failure log available.}
+
+Please diagnose why the deploy/rollback failed:
+1. Check server logs: tail -100 $LOG_FILE
+2. Check if the server process is running
+3. Check launchd service status
+4. Inspect the build output and configuration
+5. Report findings via dispatch-event
+
+If you can fix the issue, do so and redeploy with: bin/dispatch-deploy $tag"
+
+  local user_local_bin="$HOME/.local/bin"
+  local path_prefix="${bin_dir}:${user_local_bin}"
+
+  echo "==> spawning diagnosis agent ($session_name)..."
+  tmux new-session -d -s "$session_name" -c "$SERVER_DIR" \
+    "DISPATCH_AGENT_ID='${agent_id}' \
+     DISPATCH_MEDIA_DIR='${media_dir}' \
+     DISPATCH_PORT='${port}' \
+     PATH='${path_prefix}':\"\$PATH\" \
+     claude --append-system-prompt 'You are a Dispatch deploy diagnosis agent. Call dispatch-event at start (working), when blocked (blocked), and before final response (done/idle).' \
+     -p '${prompt//\'/\'\\\'\'}'" \
+    2>/dev/null || {
+      echo "warning: could not spawn diagnosis agent (tmux unavailable?)" >&2
+      return
+    }
+
+  echo "  agent ID:    $agent_id"
+  echo "  tmux attach: tmux attach -t $session_name"
+}
+
 main() {
   local tag="${1:-}"
   if [[ -z "$tag" ]]; then
@@ -188,6 +270,7 @@ main() {
   echo "==> building $tag"
   if ! build_tag "$tag"; then
     write_failure_log "build" "$tag" "$previous_tag" "" "Build failed for $tag."
+    spawn_diagnosis_agent "build" "$tag" "$previous_tag"
     echo "error: build failed" >&2
     exit 1
   fi
@@ -211,6 +294,7 @@ main() {
 
   if [[ "$previous_tag" == "$tag" || "$previous_tag" == "unknown" ]]; then
     write_failure_log "health check" "$tag" "" "" ""
+    spawn_diagnosis_agent "health check (no rollback available)" "$tag" "unknown"
     echo "error: cannot roll back (no previous tag available)" >&2
     exit 1
   fi
@@ -226,6 +310,7 @@ main() {
 
   write_failure_log "health check" "$tag" "$previous_tag" "failed" \
     "Rollback to $previous_tag also failed. Server may be down."
+  spawn_diagnosis_agent "health check (deploy + rollback failed)" "$tag" "$previous_tag"
   echo "error: deploy failed and rollback failed — server may be down" >&2
   echo "       check logs: tail -f $LOG_FILE" >&2
   exit 1

bin/dispatch-release

@@ -146,8 +146,9 @@ main() {
 
   echo "==> release workflow produced tag: $tag"
 
-  # dispatch-deploy handles rollback and failure logging internally.
-  # We don't spawn a diagnosis agent here since Dispatch may be down.
+  # dispatch-deploy handles rollback, failure logging, and diagnosis agent
+  # spawning internally. It launches diagnosis agents via tmux+claude CLI
+  # directly, so it works even when the server is down.
   if ! "$ROOT_DIR/bin/dispatch-deploy" "$tag"; then
     echo "error: deploy failed — see $HOME/.dispatch/logs/last-release-failure.log" >&2
     exit 1