Install someone’s judgment.

How a mentorfile is made

/mentorfile:mine reads your Claude Code, Codex and Cursor history on your own machine. It skips the code and the model’s replies, and looks at your side of the conversation: the moments you stopped the agent, rejected a diff, or repeated a rule for the third time.

You wrote, in 11 sessions

“no, don’t pull in a date picker lib. the browser has one”

It became

P7. Reach for the native platform before adding a dependency.

Every principle carries a count of the sessions behind it. You read the result, fix anything that isn’t you, try it on yourself, and only then publish.

What it looks for

Mining reads only your side of each conversation and sorts what it finds into eight kinds:

• Corrections. You stopped or redirected the agent: what was wrong, what you wanted.
• Standards. Rules you set unprompted: always, never, prefer, don’t.
• Decisions. Choices between options, and why.
• Details. Things you caught that were missed: edge cases, naming, copy, polish.
• Process. How you plan, verify, test, review and ship.
• Stack. The tools you use and what you think of them.
• Voice. How you communicate: tone, length, directness.
• Product. Product, UX and business calls.

Each finding keeps its context when it matters: whether it was a prototype or a mature product, how tight the deadline was, how much risk was acceptable. A principle that only holds in some situations says so (“applies when…”), and when you make opposite calls in different situations, both are kept. A mentorfile asks about your context before answering when the answer depends on it.

Before anything is published, a leak scan checks the result for:

• Secrets and tokens: API keys, cloud keys, GitHub and Slack tokens, private keys, JWTs.
• Emails, URLs, IP addresses and file paths.
• Quotes copied from your sessions instead of generalized.
• Names that identify your work, built from your own machine: your project folders, your git remotes, and the product, client and people names in your extracts.

Both lists live in the open-source skill, so you can read exactly what runs. Think it should look for something else, or catch another kind of leak? Propose it.

Does it work?

A fair question: is one person’s judgment worth more than the “average senior” a model already gives you for free? Each mentorfile can answer that with its owner’s own data.

• About one in seven of your sessions is held out when mining and never used to build your mentorfile. The same ones every time, so they can’t leak in later.
• From those sessions come real test cases: what the agent proposed, and how you actually reacted.
• Three respondents see only the situation: your mentorfile, a generic “apply senior engineer judgment” prompt, and the plain model.
• A grader who doesn’t know which answer is which scores each one against what you really said.

The result is shown on the profile, for example “matched 31 of 40 real corrections; the generic prompt matched 18”. It runs on the owner’s machine, because the test cases are their private sessions, so it’s self-reported, and labeled that way. The method is in the skill (Step 4), and the plugin’s own test suite is in the repo.

Get the plugin

Claude Code

Codex

Then /mentorfile:find someone for the problem in front of you and /mentorfile:get their mentorfile. Every answer starts with “Simulated <handle>”: it models their judgment, it isn’t them, and it says so when their sessions don’t cover your question. Every command it runs, and what it sends, is listed in the CLI docs.

Publish yours

Run /mentorfile:mine. It reads your sessions on your machine, shows you what it found, checks it for anything private, and publishes it at mentorfile.com/@yourhandle once you say yes. Your GitHub account is your mentorfile account: no signup. Run it again whenever you like; it picks up only new sessions and replaces the old version for everyone.

Pricing

Every mentorfile is free, one-time or monthly. Its owner decides.

• Free. Install it any time with /mentorfile:get. No account needed.
• One-time. Pay once and keep it, including every future version its owner publishes.
• Monthly. Updates for as long as you’re subscribed; cancel anytime. What you’ve already installed stays on your machine.

Checkout is handled by Lemon Squeezy, which takes card or PayPal and handles sales tax and VAT. You sign in with GitHub so the purchase lands on the same account the plugin uses.

If you publish one: it’s free until you set a price in your dashboard, from $1 to $500, one-time or monthly. You keep 80% of the price before tax. Mentorfile keeps 20%, which covers payment processing, tax handling and hosting. Earnings are paid out monthly to your PayPal or Wise email.

How we suggest a price

Every owner gets the same suggestion, from how much evidence backs their mentorfile: the sessions it was distilled from, and how many principles are starred (seen in three or more separate sessions). It lands in the highest row where both minimums hold. It’s only a suggestion; owners set their own price.

Good to know

• It’s a simulation. Answers start with “Simulated <handle>”, cite the principle behind each point, and say so when that person’s sessions don’t cover your question. It isn’t them and knows nothing about their private work.
• Sessions on a profile counts the AI sessions that contributed at least one piece of evidence, not every session the owner ever had.
• Updates. Owners republish whenever they like; only the latest version is kept. Run /mentorfile:get with no handle to update everything you have.
• Unpublishing. Owners can take theirs down any time with mf unpublish. Copies people already installed stay on their machines.
• Analytics. This site counts page views with Vercel Web Analytics, which uses no cookies and doesn’t follow you across other sites. The plugin sends no analytics.
• Refunds go through Lemon Squeezy’s receipt email, and a refund ends access to future versions.

What stays on your machine

• Your transcripts. Mining runs locally through the model you already use, and nothing is uploaded anywhere else.
• The raw extracts with verbatim quotes, and the evidence file that links each principle to the sessions behind it.

What gets published

Only your principles, playbooks, a note on how you communicate, and the profile line that goes with them. Nothing is uploaded until the leak scan is clean and you say yes.