Daily Vibe - 10-23-26 · Vibe Coders

17h • News

Daily Vibe - 10-23-26

Quick one today, but the lead story is a "stop what you're doing" kind of bug if you run OpenCode on Windows or WSL. Everything else is standards drama and a $65K excuse to vibe code a game.

OpenCode's MCP sign-in bug let a server run commands on your machine

Here's the setup: OpenCode connects to MCP servers to give your agent outside tools. When you sign in to one of those servers, the server gets to hand you a web address for OAuth — the "go here to authorize" link. On Windows and WSL, OpenCode passed that address into a PowerShell command without escaping it. Pillar Security found that a malicious address could carry a PowerShell payload and run it, with your access, the moment you signed in.

The scary part isn't really the bug. Command injection happens. It's how it got fixed. Anomaly (the company behind OpenCode) patched it in version 1.18.33 on September 28, buried inside a pull request labeled as a "refactor" of browser-launch code. No security advisory, no CVE. Pillar reported it privately on August 29 and asked Anomaly to request a CVE; Anomaly's answer, per Pillar, was that a malicious MCP server is "outside our trust boundary." That means your dependency scanner and vulnerability alerts have nothing to match against. The only way to know you're exposed is to go check the version number yourself.

Worth noting: your MCP server doesn't need to start out malicious. One you vetted and added months ago could get compromised later, or its domain could lapse and get bought by someone else, and you'd never know until sign-in runs a command on your box.

The fix: upgrade to 1.18.33 or later if you're on Windows or WSL. If you can't upgrade today, run opencode mcp list, look at what remote servers you've got configured, and rip out anything you haven't fully vetted. macOS and Linux aren't affected — OpenCode doesn't route the browser launch through a shell there. And this is separate from the OpenCode Console breach reported around the same time, which may have exposed stored API keys.

[YOUR TAKE: do you audit your MCP server list regularly, or is it a "set it and forget it" thing until something like this happens?]

Quick hits

Anthropic gave away Agent Skills. Anthropic donated the Agent Skills format — the SKILL.md standard your Claude Code setup might already use — to the Agentic AI Foundation, the same neutral home that houses MCP. Anthropic says (its own numbers) over 70 platforms support the format, with more than a million published skills and close to 280 million installs. Three other companies dropped projects into the same foundation on the same day: OpenAI's Open Responses spec, Hugging Face's OpenEnv for training agents, and Alibaba's OpenSandbox for running them safely. Translation: the skills format you're using probably isn't going anywhere, and it's less likely to get quietly reshaped by one vendor's roadmap. Read more — wait, wrong link, see sources below for the right ones.

OpenAI opened up its decision-model API. OpenAI's Decisions API — basically their answer to TypeSafe's Jev, for apps that need a fast choice or score instead of generated text — is now available to everyone. OpenAI says it shrank response times from 1–3 seconds down to 100–180 milliseconds, and customer Speak reported cutting action latency from 1.4 seconds to 0.13 seconds. On public benchmarks it lands eighth and fourteenth against competitors, so it's not winning on raw accuracy, but if your app just needs "yes/no, how confident" fast and cheap, it's worth a look. Read more

Epic will pay you to vibe code a game. The 2026 AI MegaJam runs November 12–15. Teams of up to five, build in Unreal Engine or UEFN, use whatever AI tools you want — Epic initially said AI was mandatory, then walked that back and called it a mistake. $65,000 in total prizes, with standout projects eligible for development grants up to $50,000. Read more

What this means for you

Check your OpenCode version today if you're on Windows or WSL. Anything below 1.18.33 is exposed, and there's no CVE to nag you about it.
Run opencode mcp list and actually look at it. A server you trusted six months ago isn't automatically still trustworthy.
If your workflow needs classification, not prose, the Decisions API and Jev-style models are worth testing against your actual use case — the benchmarks disagree with each other, so trial it yourself rather than picking by leaderboard.
If you've got four days free next month, the MegaJam is a built-in deadline to prototype something with an agent in the loop.

[YOUR TAKE: anyone actually benchmarked a decision model against a regular LLM call for their own app, or is it still "sounds nice, haven't needed it"?]

What's your OpenCode version right now — did you have to go check? Drop it in the thread.

Sources