Claude Performance Review Skill: Turn a Messy Year Into a Self-Review and Promotion Packet

One Claude skill reads your scattered year of work and writes the self-review — without inventing a single number.

One Claude skill. It reads your scattered year of work and writes the self-review — without inventing a single number.

---

Twice a year, a specific dread arrives. The self-review is due Friday, and your "year of accomplishments" is a graveyard of half-remembered wins scattered across Slack threads, a calendar you barely recognize, and a nagging sense that you did a lot but can't prove any of it.

So you do what everyone does. You open a blank box the night before, and you either write corporate mush that says nothing "consistently drove impact across key initiatives" or, worse, you reach for a number that sounds right. "Reduced load time by 40%." Did you? You think so. It felt like a lot.

That second failure is the expensive one. A performance review with one invented metric is worse than a review with no metrics at all, because the invented one gets caught. It gets caught in calibration, when a peer or a skip-level asks "where's that number from?" and you don't have an answer. And the moment you're caught rounding up once, every other real accomplishment on the page inherits the doubt.

People pay career coaches $150–400 an hour to sit with them and solve exactly this: drag the year out of you, separate what you can prove from what you only feel, and shape it into an argument a promotion committee will actually buy.

This is one Claude skill that does that job. You paste in the chaos; it returns a submission-ready packet. And the entire thing is built around a single non-negotiable rule: it never invents a number.

Let me show you how it works, what's inside it, and — the part I care most about — how we tested it hard enough to trust it.

---

What the skill actually produces

You give it whatever you have. Genuinely whatever — Slack fragments, PR titles, a line from a 1:1, a manager's offhand compliment, three failures you're not sure you should mention. Unordered, incomplete, contradictory. That's the expected input, not the ideal one.