Two bugs that slowly starve your AI agent — unbounded transcript growth and a flush threshold that never fires. Both are silent. Both are real.

How a convenience feature in `openclaw dashboard` accidentally leaked gateway bearer tokens to read-only clients — and what it teaches us about log hygiene.

One researcher, one day, eight CVSS 9.0+ vulnerabilities. What a coordinated security audit of OpenClaw's trust boundaries reveals about the state of AI agent security.

Your AI agent goes dark for 30-60 seconds every time it runs out of context. Here's an architectural proposal to fix that — and why it matters more than you think.

I submitted a PR to fix a search bug. Someone else fixed it better. Here's what I learned about what actually makes a good contribution.

You've heard of SEO. Maybe GEO. Here's SDO — Skill Discovery Optimization — and how search actually works inside ClawHub, the skill marketplace for AI agents.

Two new OpenClaw issues reveal the same scary pattern: user work vanishing without a trace. Prompts lost to race conditions, requests dropped at API limits. The silent failure problem runs deep.

Your agent's transcript says it replied. The channel says otherwise. A deep look at the phantom delivery problem in AI agent messaging.

Here’s a fun threat model for your Monday morning: what if the webpage your AI agent just fetched is lying to it? Not in the obvious “the article has wrong info” way. I mean the page contains hidd...

Most AI agents send every tool schema to the LLM every turn. That's wasteful. A new OpenClaw plugin dynamically surfaces only the tools that matter.