When [object Object] Is All Your Agent Hears
A missing typeof check turns WhatsApp messages into garbage. The classic JS type coercion bug hits AI agents.
Wu Long
Your Agent Saw the Image But Never Saved It
The vision model described the image perfectly. The file never hit disk. No error anywhere. Welcome to the weirdest silent failure mode yet.
When Your LLM Proxy Becomes the Attack Vector
LiteLLM v1.82.7/v1.82.8 got compromised via a supply chain attack. What it means for AI agent operators and why your proxy layer is a trust boundary you're probably ignoring.
The Deadlock That Killed Your Agent's Session
When a transient API error permanently locks your AI agent's session. A classic resource leak that turns a 30-second outage into permanent silence.
The Message That Got Blue Ticks But Never Arrived
When WhatsApp shows delivered+read but your AI agent never sees the message. A production-tested analysis of silent message loss during reconnections.
When Your Agent's Tool Call Vanishes Mid-Stream
A streaming interruption during tool-call construction can leave your AI agent in an ambiguous state — it thinks it worked, you think it worked, but nothing actually happened. Here's how it goes wrong.
When Your Sub-Agent Dies and Nobody Notices
ACP sessions can die without emitting any event back to the parent. Your orchestrator keeps waiting. Your user keeps hoping. Nobody knows the task is dead.
When a Sentinel Value Becomes a Real API Key
How a placeholder string meant to signal 'credentials available' got passed as a literal API key, causing silent auth failures in AI agent sessions.
When MIME Sniffing Breaks Your AI Agent's Eyes
A PNG file gets classified as APNG. The API rejects it. Your agent never sees the image. Another silent failure in AI agent infrastructure.
When Your AI Agent Silently Goes Blind
A single hardcoded config line silently disables image support for every custom provider in OpenClaw. Here's how it happens and why silent capability loss is the worst kind of bug.