The job
You wrote a rushed reply in Gmail, and the meaning is right but the tone is off. You need the message to sound like you on a careful day: direct, warm where needed, and still specific to the client thread.
Why this is hard without Sephir
The usual route is copy the draft, open ChatGPT, paste the message, then explain your style from memory. The rewrite comes back polished in the wrong way: too formal, too soft, or full of phrases you would never send. Then you spend another five minutes editing the edited version back into your voice.
How Sephir does it
- Pin your active Gmail draft and two or three sent emails that sound like you.
- Open the Sephir sidepanel with
Cmd+Shift+S. - Ask Sephir to rewrite the draft using the pinned sent emails as tone examples.
- Watch the audit timeline:
extractPageText(sent tab)→extractPageText(draft tab)→ rewrite. - Copy the revised version back into Gmail.
- Save the trace as
/match-tonefor the next draft.
The skill behind it
This skill keeps the job narrow: read the draft, read your examples, rewrite only the text you asked it to change.
What it costs
Sephir runs this recipe on your own ChatGPT Plus via Codex OAuth or direct API key. Typical token usage: ~3,000-8,000 input + ~500-1,500 output per run on a frontier model like Claude Opus 4, GPT-5, or Gemini 3 Pro. Short single-turn rewrites fit the Free tier; saving /match-tone as a reusable skill is covered by .