Each one is documented in the OpenClaw issue tracker. Each one is verifiable.
Issue #28622 · OpenClaw
0
messages received, bot shows running
Silent message drop
Telegram stops delivering messages to the bot. No error in OpenClaw logs. No webhook failure. The bot is alive. Messages are gone. Users think you broke something.
Orchesis detects it in 90 seconds — Compares expected message rate against actual. Telegram alert with channel, time, and last successful delivery.
Issue #33013 · OpenClaw
VPN
breaks Telegram polling silently
VPN kills your bot at 3am
VPN rotation disconnects Telegram long-polling. OpenClaw doesn't know. Retries silently fail. Your bot is offline until someone manually restarts it. Nobody's watching at 3am.
Orchesis detects VPN failure in 2 minutes — Monitors polling continuity. Sends Telegram alert with reconnect command before the first user notices.
Issue #43178 · OpenClaw
18
agents sharing one OAuth profile
Watchdog cascade
18 agents sharing one OAuth profile. Agent A refreshes the token. Agent B uses the stale one, gets 401, falls back to Opus at $0.186/turn. Watchdog triggers. All 18 agents restart simultaneously. $636/month from one config decision.
Orchesis identifies the cascade root cause — Cross-agent token tracking shows which agent caused the refresh race. One fix, not 18 restarts.
Issue #9828 · OpenClaw
~100K
extra tokens per message, default config
Config schema cost on every message
Default OpenClaw config injects the full configuration schema into every system prompt. Every Telegram message triggers a call with ~100,000 extra tokens. Breaks Anthropic prompt caching. You pay full price on every call instead of 90% savings.
orchesis verify finds it in 30 seconds — Shows exact overhead, cost per message, and the one-line fix.