Seal your own take, get the AI's answer blind, and record where you landed. Over time you see whether you're still doing the thinking, or just signing off.
Free. No account. Your log stays in your browser.
Ninety days later, the outcome is in. The record shows who was right, and whether you defer more than you should.
An AI that has seen your answer isn't checking you. It's agreeing with you, in better prose.
A good meeting chair speaks last, because the first opinion in the room anchors everyone after it. AI works the same way, in both directions.
If the AI answers first, you anchor on it. Researchers call this automation bias: a confident, fluent answer pulls your judgment toward it before you've formed one.
If you share your view first, the AI tends to mirror it back. As obvious mistakes get rarer, the errors that slip through are the agreeable ones.
Sealed First fixes the order. You commit first and keep it hidden. The AI answers without seeing you. Then you compare, on the record.
Write the question, your take, and how sure you are. It gets a fingerprint and can't be edited.
The AI answers the question alone. It never sees your take.
Agree, partly, or disagree. Then held, updated, or switched, with a reason for any change.
When the result is in, mark who was right. That's what the score is built on.
When you and the AI agree, a red-team step asks for the strongest case against both of you. Shared blind spots hide in agreements.
A number from 0 to 100 over the last 90 days. It's provisional until five seals are resolved, and it splits by domain so you can see where you hold up and where you defer.
You don't need to seal everything. Pick a weekly target, say three decisions that matter, and keep it.
The full routine: seal, paste in any AI's answer, judge, red-team, resolve, and your score by domain. Nothing leaves your device.
Type /seal in any chat. A separate agent that never sees your take answers blind, and /seal-brief gives you your score.
/plugin marketplace add tan2line/sealed-first /plugin install sealed-first@sealed-firstDownload the plugin file
The skills use the open SKILL.md format, read by Codex CLI, GitHub Copilot, Gemini CLI, Cursor and more. Copy the skills folder into your tool.
One log follows you: the skills save to a local file that the browser app can import, and the app exports the same file.
Most organizations say a person reviews every AI recommendation. Few can show that the person forms their own view first. A Sealed First pilot puts one sealed step in front of the AI's recommendation and turns the result into an evidence report.