Feature Request: Self-learning voice profile: sharpen tone matching after every use

Summary

Snapdraft rewrites in modes like Improve writing, Email, and Message, and lets me build custom modes. What it does not do yet is get better at sounding like me the more I use it. I'd like Snapdraft to learn my voice over time so each rewrite lands closer to how I actually write, with no extra effort on my part.

The problem

Every rewrite starts from a generic notion of "good writing." My real voice (word choices, sentence length, punctuation habits, formality, the phrases I lean on) has to be re-coaxed each session, either by editing the output or by hand-tuning a custom mode. The tool never remembers what I settled on last time.

Proposed feature

A single evolving voice profile that Snapdraft builds from the versions I actually keep. Whenever I accept or copy out a rewrite, Snapdraft treats that final text as an example of my voice and folds it into one profile. Every mode, built-in and custom, then leans on that profile so results drift toward my tone with each use. The signal is the text I chose to send, so it learns from real outcomes rather than guesses.

• One profile, not one per mode. It represents "how I write," and each mode layers its intent on top.
• Passive by default. No ratings, no extra button. Using Snapdraft is the training.
• Compounding. Early on it nudges; after weeks of use it should already know my defaults.

Control and privacy

Snapdraft is local-first, so the profile should stay on my machine and stay mine to inspect:

• Opt-in, with a clear on/off toggle.
• Viewable: I can see what it has learned about my voice in plain terms.
• Editable and resettable: I can correct it, pause learning, or wipe it and start over.

Why this fits Snapdraft

The product's premise is keeping writing in place and on my machine. A voice profile that improves locally, from my own accepted drafts, extends that from "rewrite in place" to "rewrite in my voice, better every time," without sending anything extra anywhere.

Open questions

• How the learned voice interacts with an explicit custom mode when the two disagree.
• Whether the profile should surface a short summary ("here's how I'd describe your current voice") for a sanity check.
• A sensible starting point before enough examples exist.

No files, tests, or entry points are named. Start by resolving how accepted or copied rewrites become one local voice profile and how it interacts with custom modes; done should include opt-in, inspectable, editable, resettable, and pausable controls with the profile kept on the machine.