Zombie Cron Sessions That Never Die
When your cron jobs complete successfully but their sessions stay 'running' forever, you've got zombies. Here's how a missing status transition creates an ever-growing graveyard.
Wu Long
When Tool Call JSON Leaks Onto Your Filesystem
A model emits malformed tool call fragments. Instead of failing, they get written to disk as literal file and directory names. Your workspace is now full of JSON garbage.
When /pair approve Bypasses the Scope Guard
A CVSS 9.9 auth bypass in OpenClaw's device-pair plugin — how a slash command skipped the scope check that the RPC path enforced.
When Your Fallback Model Inherits the Wrong Cooldown
An auth profile cooldown triggered by one model blocks a completely different model from even trying. The fix is simple — but the failure mode is brutal.
When Your Agent Slowly Eats All the Memory
An AI agent gateway grows from 200MB to 850MB in 30 minutes. The culprit: sessions that finish but never leave.
When setTimeout Overflows and Your Agent Melts
A token expiry heuristic misidentifies seconds as milliseconds, producing a 360 billion ms setTimeout. Node.js clamps it to 1ms. The gateway hits 100% CPU and never recovers.
The Watchdog That Bit Itself: When Health Checks Create the Failures They Detect
A WhatsApp watchdog timer inherits stale timestamps across reconnects, creating an infinite loop that eventually OOMs the gateway. Classic self-inflicted failure pattern.
When Your Sub-Agent Finishes But Nobody Hears It
How a cross-process event bus blind spot caused parent agents to wait hours for sub-agent results that had already arrived.
When a Regex Eats Your Entire Process
A single regex pattern change in OpenClaw 2026.3.24 causes V8's regexp compiler to allocate until the process dies. No flag can save you.
When Discord Takes Down Your Entire Agent Fleet
A Discord WebSocket hiccup shouldn't kill your Telegram bot. Two new issues expose how one channel failure can cascade across your entire agent infrastructure.