Oh-My-OpenAgent just dropped v4.6.0, and it's all about making OpenCode rock solid. The team tackled some nagging issues that caused duplicate responses, silent failures, and weird edge cases with Anthropic's assistant-tail. If you've been using OpenCode for AI agent workflows, this update should noticeably reduce noise.
The marquee fix is in the internal prompt dispatch. The team hardened it to prevent duplicate parent wakes, continuation prompts, and overlapping internal replies. That means fewer instances where your agent spits out the same response twice or starts talking over itself. Also, they improved recovery for Anthropic assistant-tail and prefill edge cases — those odd moments where the model chokes on the first token. Plus, there's a silent bug fix: zero-message completions in run mode are no longer silent. You'll see a proper output now instead of a blank wall. Session idle and status handling got a refresh to stop duplicate OpenCode responses. And they finally axed the legacy context-window monitors.
For anyone running AI agents in production, duplicate responses and silent failures are the worst. They waste tokens, confuse downstream systems, and erode trust. This release directly addresses those pain points. The Anthropic assistant-tail fix is especially interesting — it's a known rough spot in Claude integrations. By cleaning that up, Oh-My-OpenAgent becomes a stronger option for teams relying on Anthropic models.
But here's the thing: they also removed the legacy context-window monitors. That's a breaking change if you were depending on them. The team must believe the new system is sufficient, but it's worth checking your monitoring pipeline. Overall, this is a quality-of-life release that prioritizes stability over new features. And honestly, that's exactly what OpenCode needed.
Official Source: https://github.com/code-yeongyu/oh-my-openagent/releases/tag/v4.6.0