Gateway Watchdog

Monitor OpenClaw Gateway health by detecting abnormal error rates in logs. Use when: (1) setting up Gateway error monitoring, (2) diagnosing repeated API fai...

Security Scan

The skill's code and instructions match its stated purpose: it scans local OpenClaw logs (journalctl or ~/.openclaw/logs), detects error spikes, and writes local state/history — it does not use network calls or external credentials.

Purpose & Capability

Name/description match the behavior. The script reads OpenClaw logs (systemd journal or ~/.openclaw/logs) and computes error rates/spikes. Required binaries (systemctl, journalctl, grep, find, awk, etc.) are appropriate for log analysis. No unrelated credentials, network access, or surprising binaries are requested.

Instruction Scope

SKILL.md instructs running the included shell script and integrating it into heartbeats/cron; the runtime instructions stay within monitoring scope. One minor mismatch: SKILL.md emphasizes "read-only" but the script writes a state.json and history.log under ~/.local/state/gateway-watchdog (local state/history). The script also relies on optional environment variables (WATCHDOG_*) that are documented in SKILL.md but not listed as required in the registry metadata.

Install Mechanism

No install spec or remote downloads — it's instruction-only with an included script. Nothing is pulled from external URLs or written to system-wide locations beyond the user's XDG state directory.

Credentials

The skill declares no required credentials (and none are needed). It does read several optional environment variables (WATCHDOG_THRESHOLD, WATCHDOG_WINDOW, WATCHDOG_SPIKE_RATIO, WATCHDOG_EXTRA_PATTERNS, WATCHDOG_DIR/STATE/LOG) which are reasonable for configurability but are not listed as required in the registry metadata — this is a documentation/metadata gap rather than a direct risk. WATCHDOG_EXTRA_PATTERNS is sanitized against some shell metacharacters, but user-supplied regexes could still be complex or expensive to evaluate.

Persistence & Privilege

The skill does not request permanent "always" inclusion and does not modify other skills or system config. It writes its own state/history under the user's XDG state directory (no elevated privileges). Autonomous invocation is allowed by default (normal for skills) but nothing here amplifies that privilege.

Assessment

This skill appears to do what it says: local, read-oriented log analysis with local state/history. Before installing, consider: (1) review and approve the included script if you don't trust the source; (2) the script writes state/history to ~/.local/state/gateway-watchdog — if that data is sensitive, pick a different path or rotate access; (3) it reads OpenClaw logs (journalctl or ~/.openclaw/logs) which may contain sensitive information, so ensure the environment running it is trusted; (4) WATCHDOG_EXTRA_PATTERNS accepts user regexes (the script filters some unsafe shell chars but not all regex constructs), so avoid supplying untrusted patterns; (5) integrate notifications (cron/heartbeat) carefully so reports are only sent to intended recipients. Overall the package is internally consistent and proportionate to its stated purpose.

Detect abnormal error patterns in the OpenClaw Gateway before they cause damage. Works with all channels: Telegram, WhatsApp, Discord, Slack, Signal, iMessage, Feishu, and more.

Born from a real incident: a silent try-catch caused 76,744 failed retries in 8 hours — undetected until the API quota was exhausted.

What It Detects

Category	Patterns
Rate limiting	HTTP 429, rate.limit, too many requests
Server errors	HTTP 5xx status codes
Auth/permission	HTTP 401/403, unauthorized, forbidden, token expired
Network errors	ETIMEDOUT, ECONNREFUSED, ECONNRESET, ENOTFOUND, socket hang up
Delivery failures	sendMessage failed, deliver failed, fetch failed
Custom	User-defined via WATCHDOG_EXTRA_PATTERNS env var

Smart Analysis

• Error rate (errors/min) — more meaningful than raw count
• Spike detection — alerts when errors jump 3x+ vs last check
• Error concentration — flags when 80%+ errors are one type (single fault source)

Quick Start

bash scripts/gateway-watchdog.sh check     # silent unless errors exceed threshold
bash scripts/gateway-watchdog.sh verbose   # always outputs full report
bash scripts/gateway-watchdog.sh history   # show monitoring history
bash scripts/gateway-watchdog.sh trend     # last 24h error trend

Heartbeat integration

Add to HEARTBEAT.md:

## Gateway Error Monitoring (every heartbeat)
- Run `~/.openclaw/workspace/skills/gateway-watchdog/scripts/gateway-watchdog.sh check`
- If output is non-empty, report to user immediately
- No output = healthy, skip reporting

Cron (optional)

openclaw cron add \
  --name "gateway-watchdog" \
  --schedule "*/30 * * * *" \
  --task "Run gateway-watchdog.sh verbose. If errors detected, notify user with the report." \
  --channel last

Configuration

All via environment variables:

Variable	Default	Description
WATCHDOG_THRESHOLD	30	Error count that triggers alert
WATCHDOG_WINDOW	30	Monitoring window in minutes
WATCHDOG_SPIKE_RATIO	3	Alert when errors jump Nx vs last check
WATCHDOG_EXTRA_PATTERNS	(empty)	Custom regex patterns (e.g., 99991400|99991403)
WATCHDOG_STATE	~/.local/state/gateway-watchdog/state.json	State file
WATCHDOG_LOG	~/.local/state/gateway-watchdog/history.log	History log

Adding channel-specific patterns

# Feishu-specific error codes
export WATCHDOG_EXTRA_PATTERNS='99991400|99991403|99991404|99991429'

# Telegram-specific
export WATCHDOG_EXTRA_PATTERNS='Too Many Requests|FLOOD_WAIT|bot was blocked'

# Discord-specific
export WATCHDOG_EXTRA_PATTERNS='DiscordAPIError|Missing Permissions|Unknown Channel'

Interpreting Results

🔴 Alert (Chinese locale)

🔴 Gateway 最近 30 分钟出现 150 条异常错误（阈值: 30，5/min）
📈 错误突增: 12 → 150（3倍阈值触发）

错误分类：
  429/限流: 120
  5xx服务端错误: 5
  认证/权限: 0
  网络错误: 5
  消息投递失败: 20

  ⚠️  单一错误类型「429/限流」占比 80%，可能是单一故障源

🔴 Alert (English equivalent)

🔴 Gateway detected 150 errors in the last 30 min (threshold: 30, 5/min)
📈 Error spike: 12 → 150 (3x threshold triggered)

Error breakdown:
  429/Rate-limit: 120
  5xx Server errors: 5
  Auth/Permission: 0
  Network errors: 5
  Delivery failures: 20

  ⚠️  Single error type "429/Rate-limit" accounts for 80%+ — likely a single fault source

💚 Healthy

No output from check mode.

Limitations

• Requires systemd + journalctl (falls back to ~/.openclaw/logs/ on macOS)
• Reactive, not preventive
• Cannot pinpoint which extension is failing — check error details for clues

Security

• Read-only: Only reads logs
• No credentials: No API keys accessed
• No network: No outbound requests
• User state only: State in ~/.local/state/gateway-watchdog/ (XDG standard, no elevated permissions needed)