Nothing to hand? Load the — a one-option decision with no recorded drawbacks, decide-later placeholders and no reversibility note, where the correct posture is not-ready and the draft carries honest [TODO] markers — or the , where the correct posture is ready and the useful output is the polish and the confirmation plan.
Paste the decision — the prescan is free
No upload, no AI: as you type, the prescan detects which of the six standard ADR sections your paste already carries (status, context, drivers, options, decision, consequences), counts the options and flags single-option decisions, checks that at least one drawback, cost or risk is recorded anywhere, lints for unfalsifiable words (“obviously”, “best practice”) and decide-later placeholders, and asks whether reversibility is addressed. Each flag explains why it matters. This part costs nothing and happens while you type.
The AI reviews and drafts — this is the metered part
An architect's pass: a ready / needs-work / not-ready posture with the single most important fix named, a section-by-section audit, prioritized gap findings that quote your actual text, and a complete MADR-style ADR — context, decision drivers, considered options with pros and cons, decision outcome, positive and negative consequences, and a confirmation plan. The draft works only from what you pasted: gaps become [TODO] markers and open questions, never invented facts, and this page re-checks the draft client-side before calling it committable. Pricing is honest: a worst-case amount is reserved before the run and only what the run actually uses is charged — the meter next to the button shows both.
File it, answer, re-run
Give the record its number and deciders and the page files it for you: the download becomes docs/adr/0007-use-x-for-y.md with MADR front matter, and a copyable shell block writes that exact file, stages it and commits it with a docs(adr): message. A Copy PR description button emits the summary, the remaining gaps and the open questions as a task list. Any [TODO] left in the draft is listed as a worklist rather than a count, and the header refuses to call the draft committable while gaps remain. Then answer the open questions in the boxes provided — they are appended to your material as a new section — and re-run: the next review is compared against the last one for the same decision, showing the posture change, which findings were resolved, which are still open, which are new, and how the [TODO] count moved. History follows your account, not the browser.
Derived from the @wshobson/architecture-decision-records skill (MIT license).
What is an Architecture Decision Record, and what does this app produce?
An Architecture Decision Record (ADR) is a short document that captures one technical decision, why it was made and what it costs, so a future reader can tell whether it still applies. ADR Studio produces one in MADR style: title, status, context, decision drivers, considered options with pros and cons, decision outcome, positive, negative and neutral consequences, and a confirmation plan — plus a review of the material you pasted, with a ready / needs-work / not-ready posture and prioritized gap findings.
Is any of it free?
Yes. The prescan is free and runs entirely in your browser with no upload: it detects which of the six standard ADR sections your paste carries, counts the options and flags single-option decisions, checks that at least one drawback, cost or risk is recorded, lints for unfalsifiable words and decide-later placeholders, asks whether reversibility is addressed, and flags a generic title. Both bundled examples are saved runs that replay for free without charging anything. Only the AI review is metered, and the meter next to the button shows the worst-case amount reserved before the run — you are charged only what the run actually uses.
Will it invent facts my notes do not contain?
No. The review works only from what you paste. Where the record needs something your material does not supply, the drafted ADR carries an explicit [TODO] marker and the gap is raised as an open question instead of being filled with a plausible-sounding invention. This page then re-checks the draft client-side — core sections present, two or more options, both positive and negative consequences, no [TODO] markers — and refuses to call it committable while any of those fail.
How do I get the ADR into my repository?
Enter the number the record should take in your repo and, optionally, the deciders. The download is then named docs/adr/0007-use-x-for-y.md with MADR front matter carrying the status, date and deciders, and the page shows a shell block you can paste into the repository root that writes that file, stages it and commits it with a docs(adr): message. A Copy PR description button gives you the pull-request body: the summary, the remaining [TODO] gaps and the open questions as a checklist.
What happens when I run it a second time on the same decision?
The second review is compared against the most recent earlier review of the same decision, and the page shows what changed: the posture before and after, which findings were resolved, which are still open, which are new, how many [TODO] markers remain, and which sections of the audit moved between missing, thin and solid. Past reviews are stored on your SkillSafe account so they follow you to another browser or machine, with a copy kept in the browser for instant display, and every entry restores its original material into the form so you can revise and re-run.
What can I paste in?
Whatever the decision currently lives in: meeting notes, a chat thread, a design-doc excerpt, or an existing ADR draft you want reviewed. You can also drag a .md or .txt file onto the box, or pick one with the file chooser. Long pastes are clipped from the middle so both the opening and the closing of the document survive, and the cut is announced in the text that is sent and in the run summary.
Can I drive it from a script?
Yes. The API tutorial documents the base URL, the request envelope, the error codes and every input field, with worked examples in cURL, Python, JavaScript, Go, Java, Ruby, PHP and C#, including the streaming lane. The API tokens page shows the token this browser is using and copies it as a shell export, so you never have to open developer tools.