scripts/sync_hf.py

#!/usr/bin/env python3
"""
OpenClaw HF Spaces Persistence — Full Directory Sync
=====================================================

Simplified persistence: upload/download the entire ~/.openclaw directory
as-is to/from a Hugging Face Dataset repo.

- Startup:  snapshot_download  →  ~/.openclaw
- Periodic: upload_folder      →  dataset openclaw_data/
- Shutdown: final upload_folder →  dataset openclaw_data/
"""

import os
import sys
import time
import threading
import subprocess
import signal
import json
import shutil
import tempfile
import traceback
import re
import urllib.request
import ssl
from pathlib import Path
from datetime import datetime
# Set timeout BEFORE importing huggingface_hub
os.environ.setdefault("HF_HUB_DOWNLOAD_TIMEOUT", "300")
os.environ.setdefault("HF_HUB_UPLOAD_TIMEOUT", "600")
# Suppress huggingface_hub progress bars and verbose download/upload logs
os.environ.setdefault("HF_HUB_DISABLE_PROGRESS_BARS", "1")
os.environ.setdefault("HF_HUB_VERBOSITY", "warning")

import logging as _logging
_logging.getLogger("huggingface_hub").setLevel(_logging.WARNING)
_logging.getLogger("huggingface_hub.utils").setLevel(_logging.WARNING)
_logging.getLogger("filelock").setLevel(_logging.WARNING)

from huggingface_hub import HfApi, snapshot_download

# ── Logging helper ──────────────────────────────────────────────────────────

class TeeLogger:
    """Duplicate output to stream and file."""
    def __init__(self, filename, stream):
        self.stream = stream
        self.file = open(filename, "a", encoding="utf-8")
    def write(self, message):
        self.stream.write(message)
        self.file.write(message)
        self.flush()
    def flush(self):
        self.stream.flush()
        self.file.flush()
    def fileno(self):
        return self.stream.fileno()

# ── Configuration ───────────────────────────────────────────────────────────

HF_TOKEN   = os.environ.get("HF_TOKEN")
OPENCLAW_HOME = Path.home() / ".openclaw"
APP_DIR       = Path("/app/openclaw")

# Use ".openclaw" - directly read/write the .openclaw folder in dataset
DATASET_PATH = ".openclaw"

# Z.AI API key (optional; used by Claude Code backend via api.z.ai)
ZAI_API_KEY = os.environ.get("ZAI_API_KEY", "")

# Gateway token (default: huggingclaw; override via GATEWAY_TOKEN env var)
GATEWAY_TOKEN = os.environ.get("GATEWAY_TOKEN", "huggingclaw")

# A2A configuration (optional; only activated when A2A_PEERS is set)
AGENT_NAME = os.environ.get("AGENT_NAME", "HuggingClaw")
A2A_PEERS = os.environ.get("A2A_PEERS", "")  # comma-separated peer URLs

# Default model — only set if user explicitly provides it; otherwise let OpenClaw decide.
OPENCLAW_DEFAULT_MODEL = os.environ.get("OPENCLAW_DEFAULT_MODEL", "")

# HF Spaces built-in env vars (auto-set by HF runtime)
SPACE_HOST = os.environ.get("SPACE_HOST", "")   # e.g. "tao-shen-huggingclaw.hf.space"
SPACE_ID   = os.environ.get("SPACE_ID", "")      # e.g. "tao-shen/HuggingClaw"

SYNC_INTERVAL = int(os.environ.get("SYNC_INTERVAL", "60"))
AUTO_CREATE_DATASET = os.environ.get("AUTO_CREATE_DATASET", "false").lower() in ("true", "1", "yes")

# Dataset repo: always auto-derive from SPACE_ID when not explicitly set.
# Format: {username}/{SpaceName}-data  (e.g. "your-name/YourSpace-data")
# This ensures each duplicated Space gets its own dataset automatically.
HF_REPO_ID = os.environ.get("OPENCLAW_DATASET_REPO", "")
if not HF_REPO_ID and SPACE_ID:
    # SPACE_ID = "username/SpaceName" → derive "username/SpaceName-data"
    HF_REPO_ID = f"{SPACE_ID}-data"
    print(f"[SYNC] OPENCLAW_DATASET_REPO not set — auto-derived from SPACE_ID: {HF_REPO_ID}")
elif not HF_REPO_ID and HF_TOKEN:
    # Fallback: no SPACE_ID (local Docker), derive from HF_TOKEN username
    try:
        _api = HfApi(token=HF_TOKEN)
        _username = _api.whoami()["name"]
        HF_REPO_ID = f"{_username}/HuggingClaw-data"
        print(f"[SYNC] OPENCLAW_DATASET_REPO not set — auto-derived from HF_TOKEN: {HF_REPO_ID}")
        del _api, _username
    except Exception as e:
        print(f"[SYNC] WARNING: Could not derive username from HF_TOKEN: {e}")
        HF_REPO_ID = ""

# Setup logging
log_dir = OPENCLAW_HOME / "workspace"
log_dir.mkdir(parents=True, exist_ok=True)
sys.stdout = TeeLogger(log_dir / "sync.log", sys.stdout)
sys.stderr = sys.stdout

# ── Telegram API Base Auto-Probe ────────────────────────────────────────────
#
# HF Spaces blocks DNS for api.telegram.org.  grammY uses Node 22's built-in
# fetch (undici) which bypasses dns.lookup patching and /etc/hosts.
#
# Solution: probe multiple Telegram API endpoints at startup.  If the official
# endpoint is unreachable, pick the first working mirror.  Then:
#   1. Set TELEGRAM_API_ROOT env var for the Node process
#   2. telegram-proxy.cjs (loaded via NODE_OPTIONS --require) intercepts
#      globalThis.fetch() and rewrites api.telegram.org URLs to the mirror.
#
# This works without a bot token — we just test HTTP reachability.
# If a bot token IS available, we do a full getMe verification.

# User can force a specific base via env var (skip auto-probe)
TELEGRAM_API_BASE = os.environ.get("TELEGRAM_API_BASE", "")

TELEGRAM_API_BASES = [
    "https://api.telegram.org",                            # official
    "https://telegram-api.mykdigi.com",                    # known mirror
    "https://telegram-api-proxy-anonymous.pages.dev/api",  # Cloudflare Pages proxy
]


def probe_telegram_api(timeout: int = 8) -> str:
    """Probe Telegram API endpoints and return the first reachable one.

    First checks if official api.telegram.org is reachable (HTTP level).
    If not, tries mirrors.  No bot token required — just tests connectivity.
    Returns the working base URL (without trailing slash), or "" if all fail.
    """
    ctx = ssl.create_default_context()
    for base in TELEGRAM_API_BASES:
        url = base.rstrip("/") + "/"
        try:
            req = urllib.request.Request(url, method="GET")
            resp = urllib.request.urlopen(req, timeout=timeout, context=ctx)
            print(f"[TELEGRAM] ✓ Reachable: {base} (HTTP {resp.status})")
            return base.rstrip("/")
        except urllib.error.HTTPError as e:
            # HTTP error (4xx/5xx) still means the host IS reachable
            print(f"[TELEGRAM] ✓ Reachable: {base} (HTTP {e.code})")
            return base.rstrip("/")
        except Exception as e:
            reason = str(e)[:80]
            print(f"[TELEGRAM] ✗ Unreachable: {base} ({reason})")
            continue

    print("[TELEGRAM] WARNING: All API endpoints unreachable!")
    return ""


# ── Sync Manager ────────────────────────────────────────────────────────────

class OpenClawFullSync:
    """Upload/download the entire ~/.openclaw directory to HF Dataset."""

    def __init__(self):
        self.enabled = False
        self.dataset_exists = False
        self.api = None

        if not HF_TOKEN:
            print("[SYNC] WARNING: HF_TOKEN not set. Persistence disabled.")
            return
        if not HF_REPO_ID:
            print("[SYNC] WARNING: Could not determine dataset repo (no SPACE_ID or OPENCLAW_DATASET_REPO).")
            print("[SYNC] Persistence disabled.")
            return

        self.enabled = True
        self.api = HfApi(token=HF_TOKEN)
        self.dataset_exists = self._ensure_repo_exists()

    # ── Repo management ────────────────────────────────────────────────

    def _ensure_repo_exists(self):
        """Check if dataset repo exists; auto-create only when AUTO_CREATE_DATASET=true AND HF_TOKEN is set."""
        try:
            self.api.repo_info(repo_id=HF_REPO_ID, repo_type="dataset")
            print(f"[SYNC] Dataset repo found: {HF_REPO_ID}")
            return True
        except Exception:
            if not AUTO_CREATE_DATASET:
                print(f"[SYNC] Dataset repo NOT found: {HF_REPO_ID}")
                print(f"[SYNC]   Set AUTO_CREATE_DATASET=true to auto-create.")
                print(f"[SYNC] Persistence disabled (app will still run normally).")
                return False
            print(f"[SYNC] Dataset repo NOT found: {HF_REPO_ID} — creating...")
            try:
                self.api.create_repo(
                    repo_id=HF_REPO_ID,
                    repo_type="dataset",
                    private=True,
                )
                print(f"[SYNC] ✓ Dataset repo created: {HF_REPO_ID}")
                return True
            except Exception as e:
                print(f"[SYNC] ✗ Failed to create dataset repo: {e}")
                return False

    # ── Restore (startup) ─────────────────────────────────────────────

    def load_from_repo(self):
        """Download from dataset → ~/.openclaw"""
        if not self.enabled:
            print("[SYNC] Persistence disabled - skipping restore")
            self._ensure_default_config()
            self._patch_config()
            return

        if not self.dataset_exists:
            print(f"[SYNC] Dataset {HF_REPO_ID} does not exist - starting fresh")
            self._ensure_default_config()
            self._patch_config()
            return

        print(f"[SYNC] ▶ Restoring ~/.openclaw from dataset {HF_REPO_ID} ...")
        OPENCLAW_HOME.mkdir(parents=True, exist_ok=True)

        try:
            files = self.api.list_repo_files(repo_id=HF_REPO_ID, repo_type="dataset")
            openclaw_files = [f for f in files if f.startswith(f"{DATASET_PATH}/")]
            if not openclaw_files:
                print(f"[SYNC] No {DATASET_PATH}/ folder in dataset. Starting fresh.")
                self._ensure_default_config()
                self._patch_config()
                return

            print(f"[SYNC] Found {len(openclaw_files)} files under {DATASET_PATH}/ in dataset")

            with tempfile.TemporaryDirectory() as tmpdir:
                snapshot_download(
                    repo_id=HF_REPO_ID,
                    repo_type="dataset",
                    allow_patterns=f"{DATASET_PATH}/**",
                    local_dir=tmpdir,
                    token=HF_TOKEN,
                )
                downloaded_root = Path(tmpdir) / DATASET_PATH
                if downloaded_root.exists():
                    for item in downloaded_root.rglob("*"):
                        if item.is_file():
                            rel = item.relative_to(downloaded_root)
                            dest = OPENCLAW_HOME / rel
                            dest.parent.mkdir(parents=True, exist_ok=True)
                            shutil.copy2(str(item), str(dest))
                    print("[SYNC] ✓ Restore completed.")
                else:
                    print("[SYNC] Downloaded snapshot but dir not found. Starting fresh.")

        except Exception as e:
            print(f"[SYNC] ✗ Restore failed: {e}")
            traceback.print_exc()

        # Patch config after restore
        self._patch_config()
        self._debug_list_files()

    # ── Save (periodic + shutdown) ─────────────────────────────────────

    def save_to_repo(self):
        """Upload entire ~/.openclaw directory → dataset (all files, no filtering)"""
        if not self.enabled:
            return
        if not OPENCLAW_HOME.exists():
            print("[SYNC] ~/.openclaw does not exist, nothing to save.")
            return

        # Ensure dataset exists (auto-create if needed)
        if not self._ensure_repo_exists():
            print(f"[SYNC] Dataset {HF_REPO_ID} unavailable - skipping save")
            return

        print(f"[SYNC] ▶ Uploading ~/.openclaw → dataset {HF_REPO_ID}/{DATASET_PATH}/ ...")

        try:
            # Count files to upload (no per-file logging to reduce noise)
            total_size = 0
            file_count = 0
            for root, dirs, fls in os.walk(OPENCLAW_HOME):
                for fn in fls:
                    fp = os.path.join(root, fn)
                    total_size += os.path.getsize(fp)
                    file_count += 1
            print(f"[SYNC] Uploading: {file_count} files, {total_size} bytes total")

            if file_count == 0:
                print("[SYNC] Nothing to upload.")
                return

            # Upload directory, excluding large log files that trigger LFS rejection
            self.api.upload_folder(
                folder_path=str(OPENCLAW_HOME),
                path_in_repo=DATASET_PATH,
                repo_id=HF_REPO_ID,
                repo_type="dataset",
                token=HF_TOKEN,
                commit_message=f"Sync .openclaw — {datetime.now().isoformat()}",
                ignore_patterns=[
                    "*.log",        # Log files (sync.log, startup.log) — regenerated on boot
                    "*.lock",       # Lock files — stale after restart
                    "*.tmp",        # Temp files
                    "*.pid",        # PID files
                    "__pycache__",  # Python cache
                ],
            )
            print(f"[SYNC] ✓ Upload completed at {datetime.now().isoformat()}")

            # Verify (summary only)
            try:
                files = self.api.list_repo_files(repo_id=HF_REPO_ID, repo_type="dataset")
                oc_files = [f for f in files if f.startswith(f"{DATASET_PATH}/")]
                print(f"[SYNC] Dataset now has {len(oc_files)} files under {DATASET_PATH}/")
            except Exception:
                pass

        except Exception as e:
            print(f"[SYNC] ✗ Upload failed: {e}")
            traceback.print_exc()

    # ── Config helpers ─────────────────────────────────────────────────

    def _ensure_default_config(self):
        config_path = OPENCLAW_HOME / "openclaw.json"
        if config_path.exists():
            return
        default_src = Path(__file__).parent / "openclaw.json.default"
        if default_src.exists():
            shutil.copy2(str(default_src), str(config_path))
            # Replace placeholder or remove provider if no API key
            try:
                with open(config_path, "r") as f:
                    cfg = json.load(f)
                # Set gateway token
                if "gateway" in cfg:
                    cfg["gateway"]["auth"] = {"token": GATEWAY_TOKEN}
                # Don't manage providers — leave them for OpenClaw to handle.
                # Just remove template entries with unresolved ${...} placeholders
                # so OpenClaw starts clean and discovers providers from env vars.
                providers = cfg.get("models", {}).get("providers", {})
                for pid in list(providers.keys()):
                    if str(providers[pid].get("apiKey", "")).startswith("${"):
                        providers.pop(pid, None)
                with open(config_path, "w") as f:
                    json.dump(cfg, f, indent=2)
            except Exception as e:
                print(f"[SYNC] Warning: failed to patch default config: {e}")
            print("[SYNC] Created openclaw.json from default template")
        else:
            with open(config_path, "w") as f:
                json.dump({
                    "gateway": {
                        "mode": "local", "bind": "lan", "port": 7860,
                        "trustedProxies": ["0.0.0.0/0"],
                        "controlUi": {
                            "allowInsecureAuth": True,
                            "allowedOrigins": [
                                "https://huggingface.co"
                            ]
                        }
                    },
                    "session": {"scope": "global"},
                    "agents": {"defaults": {"workspace": "~/.openclaw/workspace"}}
                }, f)
            print("[SYNC] Created minimal openclaw.json")

    def _patch_config(self):
        """Ensure critical settings after restore."""
        config_path = OPENCLAW_HOME / "openclaw.json"
        if not config_path.exists():
            self._ensure_default_config()
            return

        print("[SYNC] Patching configuration...")
        try:
            with open(config_path, "r") as f:
                data = json.load(f)
            print("[SYNC] Config parsed OK.")
        except (json.JSONDecodeError, Exception) as e:
            # Config is corrupt — back up and start fresh
            print(f"[SYNC] Config JSON is corrupt: {e}")
            backup = config_path.with_suffix(f".corrupt_{int(time.time())}")
            try:
                shutil.copy2(config_path, backup)
                print(f"[SYNC] Backed up corrupt config to {backup.name}")
            except Exception:
                pass
            data = {}
            print("[SYNC] Starting from clean config.")

        try:
            # Remove /dev/null from plugins.locations
            if "plugins" in data and isinstance(data.get("plugins"), dict):
                locs = data["plugins"].get("locations", [])
                if isinstance(locs, list) and "/dev/null" in locs:
                    data["plugins"]["locations"] = [l for l in locs if l != "/dev/null"]

            # Clean up invalid config keys that crash OpenClaw
            if "auth" in data and isinstance(data.get("auth"), dict):
                data["auth"].pop("defaultScope", None)
                if not data["auth"]:
                    del data["auth"]
            if "gateway" in data and isinstance(data.get("gateway"), dict):
                auth = data["gateway"].get("auth", {})
                if isinstance(auth, dict):
                    auth.pop("scope", None)

            # Force full gateway config for HF Spaces
            # Dynamic allowedOrigins from SPACE_HOST (auto-set by HF runtime)
            allowed_origins = [
                "https://huggingface.co",
                "https://*.hf.space",
            ]
            if SPACE_HOST:
                allowed_origins.append(f"https://{SPACE_HOST}")
                print(f"[SYNC] SPACE_HOST detected: {SPACE_HOST}")
            data["gateway"] = {
                "mode": "local",
                "bind": "lan",
                "port": 7860,
                "auth": {"token": GATEWAY_TOKEN},
                "trustedProxies": ["0.0.0.0/0"],
                "controlUi": {
                    "allowInsecureAuth": True,
                    "dangerouslyDisableDeviceAuth": True,
                    "allowedOrigins": allowed_origins
                }
            }
            print(f"[SYNC] Set gateway config (port=7860, auth=token, origins={len(allowed_origins)})")

            # Ensure agents defaults
            data.setdefault("agents", {}).setdefault("defaults", {}).setdefault("model", {})
            data.setdefault("session", {})["scope"] = "global"
            ''' 
            # ── Provider cleanup ─────────────────────────────────────────────
            # Don't manage providers — let OpenClaw handle them natively from
            # its own env var support. We only clean up stale providers that
            # were restored from the dataset backup with hardcoded API keys
            # that no longer exist in the current environment.
            restored_providers = data.get("models", {}).get("providers", {})
            if restored_providers:
                stale = []
                for pid, pcfg in list(restored_providers.items()):
                    api_key = pcfg.get("apiKey", "")
                    # Skip placeholder-style keys (${VAR}) — OpenClaw resolves these
                    if not api_key or str(api_key).startswith("${"):
                        continue
                    # Hardcoded key from backup — check if it's still in env
                    # We can't know which env var it came from, so check if
                    # the exact key value exists in any current env var
                    if api_key not in os.environ.values():
                        stale.append(pid)
                for pid in stale:
                    del restored_providers[pid]
                    print(f"[SYNC] Removed stale provider '{pid}' (API key no longer in environment)")
                   '''  
            if OPENCLAW_DEFAULT_MODEL:
                data["agents"]["defaults"]["model"]["primary"] = OPENCLAW_DEFAULT_MODEL

            # ── ACP (Agent Client Protocol) — native Claude Code integration ──
            data["acp"] = {
                "enabled": False,
                "backend": "acpx",
                "defaultAgent": "claude",
                "allowedAgents": ["claude"],
                "maxConcurrentSessions": 4,
                "dispatch": {"enabled": True},
                "runtime": {"ttlMinutes": 120}
            }
            print("[SYNC] ACP enabled: backend=acpx, agent=claude")

            # Plugin whitelist
            data.setdefault("plugins", {}).setdefault("entries", {})
            plugin_allow = ["telegram", "whatsapp", "coding-agent", "acpx"]
            if A2A_PEERS:
                plugin_allow.append("a2a-gateway")
            data["plugins"]["allow"] = plugin_allow

            # ── acpx Plugin Configuration (ACP backend) ──
            data["plugins"]["entries"]["acpx"] = {
                "enabled": True,
                "config": {
                    "permissionMode": "approve-all"
                }
            }

            # ── Coding Agent Plugin Configuration ──
            CODING_TARGET_SPACE = os.environ.get("CODING_AGENT_TARGET_SPACE", "")
            CODING_TARGET_DATASET = os.environ.get("CODING_AGENT_TARGET_DATASET", "")
            if CODING_TARGET_SPACE:
                data["plugins"]["entries"]["coding-agent"] = {
                    "enabled": True,
                    "config": {
                        "targetSpace": CODING_TARGET_SPACE,
                        "targetDataset": CODING_TARGET_DATASET,
                        "hfToken": HF_TOKEN or "",
                        "zaiApiKey": ZAI_API_KEY or os.environ.get("ZHIPU_API_KEY", ""),
                    }
                }
                print(f"[SYNC] Coding agent configured: space={CODING_TARGET_SPACE}, dataset={CODING_TARGET_DATASET}, zaiKey={'set' if (ZAI_API_KEY or os.environ.get('ZHIPU_API_KEY')) else 'missing'}")
            if "telegram" not in data["plugins"]["entries"]:
                data["plugins"]["entries"]["telegram"] = {"enabled": True}
            elif isinstance(data["plugins"]["entries"]["telegram"], dict):
                data["plugins"]["entries"]["telegram"]["enabled"] = True

            # ── A2A Gateway Plugin Configuration (only if A2A_PEERS is set) ──
            if A2A_PEERS:
                peers = []
                for peer_url in A2A_PEERS.split(","):
                    peer_url = peer_url.strip()
                    if not peer_url:
                        continue
                    name = peer_url.split("//")[-1].split(".")[0].split("-")[-1].capitalize()
                    peers.append({
                        "name": name,
                        "agentCardUrl": f"{peer_url}/.well-known/agent-card.json"
                    })
                    print(f"[SYNC] A2A peer: {name} → {peer_url}")

                data["plugins"]["entries"]["a2a-gateway"] = {
                    "enabled": True,
                    "config": {
                        "agentCard": {
                            "name": AGENT_NAME,
                            "description": f"{AGENT_NAME} - HuggingClaw A2A Agent",
                            "skills": [{"id": "chat", "name": "chat", "description": "Chat bridge"}]
                        },
                        "server": {"host": "0.0.0.0", "port": 18800},
                        "security": {"inboundAuth": "none"},
                        "routing": {"defaultAgentId": "main"},
                        "peers": peers
                    }
                }
                print(f"[SYNC] A2A gateway configured: name={AGENT_NAME}, port=18800, peers={len(peers)}")

            # ── Telegram channel defaults (open DM policy for HF Spaces) ──
            # Personal bot on HF Spaces — no need for strict pairing.
            tg_ch = data.setdefault("channels", {}).setdefault("telegram", {})
            tg_ch["dmPolicy"] = "open"
            tg_ch["allowFrom"] = ["*"]
            tg_ch["configWrites"] = True
            print("[SYNC] Set channels.telegram: dmPolicy=open, allowFrom=[*], configWrites=true")

            # ── Telegram API base auto-probe ──────────────────────────────
            # Probe is done in run_openclaw() — sets TELEGRAM_API_ROOT env var
            # for the telegram-proxy.cjs preload script to intercept fetch().

            with open(config_path, "w") as f:
                json.dump(data, f, indent=2)
            print("[SYNC] Config patched and saved.")

            # ── Deploy workspace templates (coding agent identity) ──
            workspace_dir = Path(OPENCLAW_HOME) / "workspace"
            workspace_dir.mkdir(parents=True, exist_ok=True)
            templates_dir = Path("/home/node/workspace-templates")
            if templates_dir.exists():
                for tmpl in templates_dir.glob("*.md"):
                    target = workspace_dir / tmpl.name
                    # Only write if file is default/empty (don't overwrite user customizations)
                    should_write = not target.exists()
                    if target.exists():
                        content = target.read_text()
                        should_write = "Fill this in" in content or len(content.strip()) < 50
                    if should_write:
                        text = tmpl.read_text().replace("{{AGENT_NAME}}", AGENT_NAME)
                        target.write_text(text)
                        print(f"[SYNC] Deployed workspace template: {tmpl.name}")

            # Fix paired devices scopes (OpenClaw 2026.2.19+ requires operator.write/read)
            # Delete old paired devices to force fresh auto-pair with correct scopes
            devices_dir = Path(OPENCLAW_DIR) / "devices"
            if devices_dir.exists():
                import shutil
                shutil.rmtree(devices_dir, ignore_errors=True)
                print("[SYNC] Deleted devices/ dir to force fresh auto-pair with operator.write/read scopes")

            # Verify write
            with open(config_path, "r") as f:
                verify_data = json.load(f)
                gw = verify_data.get("gateway", {})
                providers = list(verify_data.get("models", {}).get("providers", {}).keys())
                primary = verify_data.get("agents", {}).get("defaults", {}).get("model", {}).get("primary")
                print(f"[SYNC] VERIFY: gateway.port={gw.get('port')}, providers={providers}, primary={primary}")

        except Exception as e:
            print(f"[SYNC] Failed to patch config: {e}")
            traceback.print_exc()

    def _debug_list_files(self):
        try:
            count = sum(1 for _, _, files in os.walk(OPENCLAW_HOME) for _ in files)
            print(f"[SYNC] Local ~/.openclaw: {count} files")
        except Exception as e:
            print(f"[SYNC] listing failed: {e}")

    # ── Background sync loop ──────────────────────────────────────────

    def background_sync_loop(self, stop_event):
        print(f"[SYNC] Background sync started (interval={SYNC_INTERVAL}s)")
        while not stop_event.is_set():
            if stop_event.wait(timeout=SYNC_INTERVAL):
                break
            print(f"[SYNC] ── Periodic sync triggered at {datetime.now().isoformat()} ──")
            self.save_to_repo()

    # ── Application runner ─────────────────────────────────────────────

    def run_openclaw(self):
        log_file = OPENCLAW_HOME / "workspace" / "startup.log"
        log_file.parent.mkdir(parents=True, exist_ok=True)

        # Debug: check if app directory exists
        if not Path(APP_DIR).exists():
            print(f"[SYNC] ERROR: App directory does not exist: {APP_DIR}")
            return None

        # Debug: check entry point (dist/entry.js or openclaw.mjs)
        entry_js = Path(APP_DIR) / "dist" / "entry.js"
        openclaw_mjs = Path(APP_DIR) / "openclaw.mjs"
        if entry_js.exists():
            entry_cmd = ["node", "dist/entry.js", "gateway"]
        elif openclaw_mjs.exists():
            entry_cmd = ["node", "openclaw.mjs", "gateway", "--allow-unconfigured"]
        else:
            print(f"[SYNC] ERROR: No entry point found in {APP_DIR}")
            print(f"[SYNC]   Checked: dist/entry.js, openclaw.mjs")
            # List what's actually there
            try:
                print(f"[SYNC]   Contents: {list(Path(APP_DIR).iterdir())[:20]}")
            except: pass
            return None

        # Use subprocess.run with direct output, no shell pipe
        print(f"[SYNC] Launching: {' '.join(entry_cmd)}")
        print(f"[SYNC] Working directory: {APP_DIR}")
        print(f"[SYNC] Log file: {log_file}")

        # Open log file
        log_fh = open(log_file, "a")

        # Pass entire environment to OpenClaw (all API keys are already in os.environ)
        env = os.environ.copy()

        # ── Telegram API base probe ──────────────────────────────────────
        # Determine working Telegram API endpoint and set env var for
        # telegram-proxy.cjs to intercept fetch() calls.
        if TELEGRAM_API_BASE:
            tg_root = TELEGRAM_API_BASE.rstrip("/")
            print(f"[TELEGRAM] Using user-specified API base: {tg_root}")
        else:
            print("[TELEGRAM] Probing Telegram API endpoints...")
            tg_root = probe_telegram_api()

        if tg_root and tg_root != "https://api.telegram.org":
            env["TELEGRAM_API_ROOT"] = tg_root
            print(f"[TELEGRAM] Set TELEGRAM_API_ROOT={tg_root}")
            print(f"[TELEGRAM] telegram-proxy.cjs will redirect fetch() calls")
        elif tg_root:
            print("[TELEGRAM] Official API reachable — no proxy needed")
        else:
            print("[TELEGRAM] No reachable endpoint found — Telegram will not work")
        try:
            # Use Popen without shell to avoid pipe issues
            # auth disabled in config — no token needed
            process = subprocess.Popen(
                entry_cmd,
                cwd=str(APP_DIR),
                stdout=subprocess.PIPE,  # Capture so we can log it
                stderr=subprocess.STDOUT,
                text=True,
                bufsize=1,  # Line buffered
                env=env  # Pass full environment (all API keys)
            )

            # Create a thread to copy output to log file; only print key lines to console
            def copy_output():
                try:
                    for line in process.stdout:
                        log_fh.write(line)
                        log_fh.flush()
                        # Only forward important lines to console (errors, warnings, startup)
                        # Skip noisy download/progress lines that flood the HF Spaces log viewer
                        stripped = line.strip()
                        if not stripped:
                            continue
                        # Skip progress bars and download noise
                        if any(skip in stripped for skip in [
                            'Downloading', 'Fetching', '%|', '━', '───',
                            'Already cached', 'Using cache', 'tokenizer',
                            '.safetensors', 'model-', 'shard',
                        ]):
                            continue
                        print(line, end='')
                except Exception as e:
                    print(f"[SYNC] Output copy error: {e}")
                finally:
                    log_fh.close()

            thread = threading.Thread(target=copy_output, daemon=True)
            thread.start()

            print(f"[SYNC] Process started with PID: {process.pid}")
            return process

        except Exception as e:
            log_fh.close()
            print(f"[SYNC] ERROR: Failed to start process: {e}")
            traceback.print_exc()
            return None

# ── Main ────────────────────────────────────────────────────────────────────

def main():
    try:
        t_main_start = time.time()

        t0 = time.time()
        sync = OpenClawFullSync()
        print(f"[TIMER] sync_hf init: {time.time() - t0:.1f}s")

        # 1. Restore
        t0 = time.time()
        sync.load_from_repo()
        print(f"[TIMER] load_from_repo (restore): {time.time() - t0:.1f}s")

        # 2. Background sync
        stop_event = threading.Event()
        t = threading.Thread(target=sync.background_sync_loop, args=(stop_event,), daemon=True)
        t.start()

        # 3. Start application
        t0 = time.time()
        process = sync.run_openclaw()
        print(f"[TIMER] run_openclaw launch: {time.time() - t0:.1f}s")
        print(f"[TIMER] Total startup (init → app launched): {time.time() - t_main_start:.1f}s")

        # 4. Start conversation-loop (RUN_ORCHESTRATOR=1, runs on God Space)
        #    OFFICE_MODE=1 is for frontend serving only (Home Space)
        conv_loop_proc = None
        if os.environ.get("RUN_ORCHESTRATOR") == "1":
            def run_conversation_loop_forever():
                """Launch conversation-loop with auto-restart on crash."""
                nonlocal conv_loop_proc
                time.sleep(60)  # let OpenClaw fully initialize
                # Ensure requests is installed (runs as non-root user node)
                pip_result = subprocess.run(
                    [sys.executable, "-m", "pip", "install", "--user",
                     "--break-system-packages", "requests"],
                    capture_output=True, text=True, timeout=120)
                if pip_result.returncode != 0:
                    print(f"[SYNC] pip install requests failed: {pip_result.stderr[:200]}")
                else:
                    print("[SYNC] pip install requests OK")
                script = os.path.join(os.path.dirname(__file__), "conversation-loop.py")
                if not os.path.exists(script):
                    print(f"[SYNC] conversation-loop.py not found at {script}")
                    return
                while not stop_event.is_set():
                    print("[SYNC] Starting conversation-loop.py (Adam & Eve orchestrator)...")
                    log = open("/tmp/conversation-loop.log", "a")
                    conv_loop_proc = subprocess.Popen(
                        [sys.executable, "-u", script],
                        stdout=log, stderr=subprocess.STDOUT,
                    )
                    print(f"[SYNC] conversation-loop.py started (PID {conv_loop_proc.pid})")
                    exit_code = conv_loop_proc.wait()
                    log.close()
                    if stop_event.is_set():
                        break
                    print(f"[SYNC] conversation-loop.py exited ({exit_code}), restarting in 30s...")
                    time.sleep(30)

            conv_thread = threading.Thread(target=run_conversation_loop_forever, daemon=True)
            conv_thread.start()
        else:
            print("[SYNC] RUN_ORCHESTRATOR not set — skipping conversation-loop")

        # Signal handler
        def handle_signal(sig, frame):
            print(f"\n[SYNC] Signal {sig} received. Shutting down...")
            stop_event.set()
            # Wait for background sync to finish if it's running
            t.join(timeout=10)
            if conv_loop_proc:
                print("[SYNC] Stopping conversation-loop...")
                conv_loop_proc.terminate()
                try:
                    conv_loop_proc.wait(timeout=5)
                except subprocess.TimeoutExpired:
                    conv_loop_proc.kill()
            if process:
                process.terminate()
                try:
                    process.wait(timeout=5)
                except subprocess.TimeoutExpired:
                    process.kill()
            print("[SYNC] Final sync...")
            sync.save_to_repo()
            sys.exit(0)

        signal.signal(signal.SIGINT, handle_signal)
        signal.signal(signal.SIGTERM, handle_signal)

        # Wait
        if process is None:
            print("[SYNC] ERROR: Failed to start OpenClaw process. Exiting.")
            stop_event.set()
            t.join(timeout=5)
            sys.exit(1)

        exit_code = process.wait()
        print(f"[SYNC] OpenClaw exited with code {exit_code}")
        stop_event.set()
        t.join(timeout=10)
        print("[SYNC] Final sync...")
        sync.save_to_repo()
        sys.exit(exit_code)

    except Exception as e:
        print(f"[SYNC] FATAL ERROR in main: {e}")
        traceback.print_exc()
        sys.exit(1)


if __name__ == "__main__":
    main()