Nobody can answer what am I good at, and who says so without a week of archaeology.
The evidence already exists. It is in meeting transcripts, old performance reviews, LinkedIn recommendations, praise emails you never filed, and cover letters you wrote at midnight. It has just never been in one place. This puts it there, interviews you to fill what is missing, and keeps doing both until you tell it to stop.
Plain markdown at ~/Career. Yours, on your machine, readable without any of this tooling.
PROFILE.md the living one-pager: what you are good at, and the proof Roles/ every job and contract, with scope and how it really ended Wins/ one file per achievement, including who benefited Praise/ what people said, quoted exactly and attributed Skills/ each capability linked to the evidence that proves it People/ collaborators, advocates, potential references Voice/ how you write, learned from your own past writing Incoming/ drop zone: old resumes, reviews, anything
You only need to remember the first one. Each command ends by telling you what to run next.
Your time is only worth spending on what nothing else can know: what you are proud of, what you would refuse to do again, why a decision was hard. So it ingests first and interviews second. It should arrive already knowing your last four jobs. Being asked to recite your own CV is what makes tools like this get abandoned in week one.
The record is never truly finished, so the questions never entirely stop. But say "ease off" or "that's enough for now" and it drops down a level and stays there. It will not creep back up. Everyone hits their own limit and wants to get on with actually finding a job, and this is built to expect that rather than fight it.
Every line in your profile traces to a specific win or a quote from a named person on a known date. Anything resting on one weak source gets marked thin, and thin claims become the next questions. A record that flatters you fails at the interview instead of at the application.
Drag anything career-related into ~/Career/Incoming/ and run /career-ingest incoming. Old resumes and CVs. Cover letters you have sent, especially ones that got you an interview. Performance reviews and 360 feedback. Talk abstracts and conference feedback forms. Screenshots of nice things people said. Your own messy notes.
Two things happen to each file. The facts become evidence, so one old resume fills in most of your timeline and one old performance review is denser in attributed praise than anything else you own. The prose teaches it how you write, so the cover letters it drafts later sound like you rather than like a language model.
Praise is quoted with the person's name and the date, because "Sarah said in March that you unblocked the whole team" is evidence and "someone once said something nice" is not.
To keep that safe, Sources/ and Incoming/ are excluded from git, so raw transcripts and private messages never enter a repo you might later push. No skill sends transcript, message, or email content to any external service. If you ever want to share the vault, there is a flag that produces an anonymised copy.
As a plugin, from inside Claude Code. Nothing outside Claude Code runs.
/plugin marketplace add bendechrai/career-brain /plugin install career-brain@career-brain
With npx, in one line. skills is a third-party CLI from Vercel Labs, not from Anthropic.
npx skills add bendechrai/career-brain --all -g -a claude-code -y
Or clone it and run the installer, which needs nothing but git.
git clone https://github.com/bendechrai/career-brain.git cd career-brain ./install.sh
Any of the three, then start Claude Code anywhere and run /career-init. The last two keep the short command names; installed as a plugin they are prefixed, so it becomes /career-brain:career-init. You rarely type either, since describing what you want triggers the right one.
# optional, but each one adds a lot Granola MCP unsolicited praise, and where your time really goes Claude in Chrome LinkedIn profile, recommendations, comments Gmail + Calendar praise threads, reviews, your real collaborators
None are required. /career-init checks what you have and adapts. The interview on its own still produces something useful.