start-openclaw.sh

#!/bin/bash
# start-openclaw.sh -- HF Spaces full startup (13 phases + post-gateway)
set -e

export OPENCLAW_STATE_DIR="${OPENCLAW_STATE_DIR:-/root/.openclaw}"
export PORT="${PORT:-7860}"
export OPENCLAW_GATEWAY_PORT="${OPENCLAW_GATEWAY_PORT:-18889}"

echo "================================================"
echo "  OpenClaw AI Gateway -- HF Spaces Startup"
echo "================================================"

# --- Phase 1: directories ---
echo "[1/13] Creating directories..."
mkdir -p "${OPENCLAW_STATE_DIR}"/{agents/main/sessions,credentials,sessions,workspace,extensions}

# --- Phase 2: System tools pre-installed (OpenClaw, Caddy, WeChat CLI if enabled) ---
echo "[2/13] System tools pre-installed (OpenClaw, Caddy, Claude Code)"

# --- Phase 3: restore backup ---
echo "[3/13] Restoring from backup..."
python3 /app/backup-manager.py restore

# --- Phase 4: generate config ---
echo "[4/13] Generating OpenClaw config..."
python3 /app/config-generator.py

# --- Phase 5-8: postpone agents/skills (will run after gateway warmup) ---
echo "[5-8/13] Skip -- post-gateway tasks postponed"

# --- Phase 9: backup scheduler ---
echo "[9/13] Starting backup scheduler..."
python3 /app/backup-manager.py scheduler &
BACKUP_PID=$!
echo "  -> Backup scheduler PID: $BACKUP_PID"

# --- Phase 10: cc-switch-web ---
echo "[10/13] Starting cc-switch-web..."
CCSWITCH_BIN="${OPENCLAW_STATE_DIR}/cc-switch-server"
if [[ ! -f "$CCSWITCH_BIN" ]]; then
    curl -fsSL "https://github.com/Laliet/cc-switch-web/releases/download/v0.11.1/cc-switch-server-linux-x86_64" \
        -o "$CCSWITCH_BIN" 2>/dev/null && chmod +x "$CCSWITCH_BIN" || echo "  WARNING: Download failed"
fi
if [[ -f "$CCSWITCH_BIN" ]]; then
    HOST=0.0.0.0 PORT=3000 "$CCSWITCH_BIN" > /dev/null 2>&1 &
    CCSWITCH_PID=$!
    echo "  -> cc-switch-web started (PID $CCSWITCH_PID)"
else
    echo "  -> cc-switch-web not available"
fi

# --- Phase 11: Caddy reverse proxy ---
echo "[11/13] Starting Caddy reverse proxy..."
caddy run --config /app/Caddyfile > /dev/null 2>&1 &
CADDY_PID=$!
echo "  -> Caddy PID: $CADDY_PID"
trap "kill $CADDY_PID 2>/dev/null" EXIT

# --- Phase 12: Third-party integrations (WeChat, Feishu) ---
echo "[12/13] Configuring third-party integrations (if enabled)..."

if [[ "${WEIXIN_ENABLED:-false}" == "true" ]]; then
    echo "  -> WeChat integration enabled, activating..."
    bash /app/scripts/wechat-activate.sh
else
    echo "  -> WeChat integration disabled (WEIXIN_ENABLED=false)"
fi

if [[ "${FEISHU_ENABLED:-false}" == "true" ]]; then
    echo "  -> Feishu integration enabled, setting up..."
    bash /app/scripts/feishu-setup.sh
else
    echo "  -> Feishu integration disabled (FEISHU_ENABLED=false)"
fi

# --- Phase 13: Start OpenClaw Gateway ---
echo "[13/13] Starting OpenClaw Gateway..."
GATEWAY_LOG="${OPENCLAW_STATE_DIR}/gateway.log"
openclaw gateway run --port "${OPENCLAW_GATEWAY_PORT}" > "$GATEWAY_LOG" 2>&1 &
OPENCLAW_PID=$!
echo "  -> OpenClaw PID: $OPENCLAW_PID (on :${OPENCLAW_GATEWAY_PORT})"
echo "  -> Gateway log: $GATEWAY_LOG"
# Stream gateway logs to stdout for debugging
tail -f "$GATEWAY_LOG" 2>/dev/null &

# Wait for gateway to be ready (via Caddy on $PORT)
echo "  -> Waiting for gateway to be ready (up to 90s) via Caddy on :$PORT..."
for i in $(seq 1 30); do
    if curl -s -f --connect-timeout 3 --max-time 5 "http://localhost:${PORT}/health" >/dev/null 2>&1; then
        echo "  -> Gateway ready!"
        break
    fi
    if [ $i -eq 10 ] || [ $i -eq 20 ]; then
        echo "  -> Still waiting... (${i}x3s elapsed)"
        if ! kill -0 $OPENCLAW_PID 2>/dev/null; then
            echo "  -> WARNING: Gateway process died! Last log lines:"
            tail -5 "$GATEWAY_LOG" 2>/dev/null || echo "  -> (no log file)"
            break
        fi
    fi
    sleep 3
done

# Check if gateway is alive after wait
if ! kill -0 $OPENCLAW_PID 2>/dev/null; then
    echo "  -> ERROR: Gateway process exited. Full log:"
    cat "$GATEWAY_LOG" 2>/dev/null || echo "  -> (no log)"
    exit 1
fi

# Extra wait for provider auth pre-warming
echo "  -> Waiting for provider auth warmup (60s)..."
for i in 1 2 3 4 5 6 7 8 9 10 11 12; do
    if curl -s -f --connect-timeout 3 --max-time 5 "http://localhost:${PORT}/health" >/dev/null 2>&1; then
        :  # still healthy
    fi
    sleep 5
done
echo "  -> Warmup window done"

# --- Post-gateway: create agents, configure skills ---
(
  sleep 2

  echo "[5/13] Creating agents..."
  for i in 1 2 3 4 5; do
    agent_id="agent-${i}"
    if timeout 15 openclaw agents list 2>/dev/null | grep -q "\"${agent_id}\""; then
      echo "  -> Agent ${agent_id} exists, skipping"
    else
      echo "  -> Creating agent ${agent_id}..."
      openclaw agents add "${agent_id}" \
        --workspace "${OPENCLAW_STATE_DIR}/workspace-agent-${i}" \
        --non-interactive 2>/dev/null || true
    fi
  done

  echo "[6/13] Configuring skills..."
  allow_bundled="${SKILLS_ALLOW_BUNDLED:-multi-search-cn,github,summarize}"
  IFS=',' read -ra ALLOW <<< "$allow_bundled"
  for skill in "${ALLOW[@]}"; do
    [[ -n "$skill" ]] && timeout 15 openclaw config set "skills.allowBundled[+]" "$skill" 2>/dev/null || true
  done

  echo "[7/13] Post-gateway setup done"
) &

# Wait for gateway to exit (keep container alive)
wait $OPENCLAW_PID