A free Claude skill, by unfayr
Ref keeps the review fair. You make the calls.
Ref runs the review and approval cycle for one marketing asset. It gathers the feedback, sorts it, flags where reviewers disagree and remembers every decision. It never writes your copy and never sends a message.
Free. No email, no account. Works in claude.ai, the Claude desktop app and Claude Code.
Ref’s call
Open items
Can we go back to the old headline?reverses an approval · locked by Jordan in R1
Can we get an SMS version too?possible change order
Looks good to me.approval
Three rules
An extra set of hands. Not another writer.
Most AI tools want to do the work for you. Ref does the part nobody wants: keeping track.
Ref never writes or rewrites the creative.
Ask for a new headline and it declines, then gives your writer a precise change list instead.
Ref never sends anything.
Review requests, reminders and conflict messages are drafts. You decide what goes out.
Ref shows its work.
Every flag points to the reviewer’s exact words and the spot in the asset, so you can check it in seconds.
How it works
From first draft to sign-off, in plain words.
No commands to learn. Say what you need and Ref picks up where the review left off.
- 01
Set up.
The asset, the reviewers, who has final say on what, and the deadline.
- 02
Ask for review.
A tailored request per reviewer. Legal gets the three claims to check, not “thoughts?”
- 03
Gather feedback.
Paste it in, or let Ref find it in Gmail, Slack or Drive. It shows you what it found first.
- 04
Sort it.
Must fix, preference, question, out of scope, or reverses something already approved.
- 05
Settle conflicts.
Both quotes, who has authority, a suggestion and a drafted message. You decide.
- 06
Fix-List.
One checklist for whoever makes the edits. Nothing rewritten.
- 07
Check the revision.
Every item confirmed, plus any change nobody asked for.
- 08
Sign off.
Rounds, decisions and who approved which version, when. Forward it as the record.
Why it exists
Reviews are where launches stall.
More marketers name reviews and approvals as their biggest time sink than anything else. The work is not hard. It is scattered across email, Slack and comments, and somebody has to remember what was decided in round one.
- 58% of marketers name admin, reviews and approvals as their biggest time sink.
- 60% of teams need four or more people to approve a single email.
- Approvals cause more missed launches than creative production does.
Sources: Adobe 2025 global marketing study and Knak 2026 survey, via Marq.
Our test
32/36 vs 25/36 for Claude alone
In 7 test scenarios (one run each, graded by Claude), Ref passed 32 of 36 behaviour checks. Claude on its own passed 25.
We wrote the scenarios alongside Ref, so they measure how well each follows Ref’s way of running a review. Both caught the conflicts and reversals; Ref’s edge was quoting exact words, applying who has final say and keeping a usable record. Read the full method and every transcript.
Install
Two minutes, wherever you use Claude.
Ref needs code execution on. It uses two small scripts to keep the review file valid. They make no network calls.
claude.ai.
- Download the Ref zip.
- Settings, then Capabilities. Turn on code execution.
- Under Skills, upload the zip. Don’t unzip it.
- Start a chat: “Start a review for my spring promo email.”
Claude desktop app.
- Download the Ref zip.
- Settings, then Capabilities. Turn on code execution.
- Upload the zip under Skills and switch Ref on.
Claude Code.
Run these in a session. You’ll get updates when we ship them.
/plugin marketplace add timlaneseo-tech/unfayr-ref/plugin install ref@unfayrTry it first
Run a sample review before your real one.
The sample kit has a spring promo email, a review already set up and feedback from four reviewers. It includes a Legal and Brand disagreement, an out-of-scope request and a plain “looks good.”
What it won’t do.
- Write or rewrite headlines, copy, subject lines, CTAs or layouts.
- Send emails, Slack messages or anything else.
- Make your decisions. It recommends. You decide.
- Review more than one asset per file. Five assets is five reviews.
Privacy.
The review file keeps reviewers’ quotes and links to where they came from, not whole threads. Ref reads only what you paste in or point it at.
Questions
The short answers.
Is it really free?
Yes. Use it for your own work and your clients’ work. You can’t resell it, modify it or rebrand it. The license is on GitHub.
Do I need a paid Claude plan?
Skills and code execution are available on paid Claude plans. Ref runs inside your Claude, so there’s nothing else to pay for.
Where does the review history live?
In one review file per asset. In Claude Code it saves to a reviews folder automatically. In claude.ai you download it at the end of a session and bring it to the next chat.
Does it work for agencies?
Yes. Tag each reviewer as client or internal. Client requests outside the brief show up as a possible change order, and drafts to clients are written diplomatically.
Will it write my copy if I ask nicely?
No. That is the point. It will give your writer a precise list of what each reviewer asked for.