Performance Reviews
Ask a manager to review someone’s year and they will mostly remember the last six weeks. Not because they are lazy — because the shipped project from March, the incident handled quietly in July, and the peer who mentioned in passing that this person unblocked their whole quarter are all scattered across Jira, Slack, a goals doc, and a one-on-one notebook. Reconstructing that on a Sunday night is not going to happen. So the review comes out vague, safe, and unhelpful: “consistently strong contributor, could take more ownership.”
Skynet does the retrieval part. It pulls the actual record from the tools where the work happened, organizes it against your review framework, and hands the manager a draft with real examples in the right places. The judgment stays entirely with the manager. The archaeology does not.
How it works
Load your framework
Give the agent your competency model, level expectations, and review template. Whatever structure you use — narrative, ratings against dimensions, goal attainment — the drafts follow it, so nothing arrives in the wrong shape for calibration.
Gather the year
The agent pulls what is already recorded: goals and where they landed, shipped work, one-on-one notes, peer feedback, project outcomes. It builds a timeline across the full period rather than the last month, which is where recency bias comes from.
Draft with evidence
For each section, the agent writes an assessment and attaches the specific examples supporting it. Nothing is asserted without a source the manager can check. Where the evidence is thin or contradictory, it says so instead of smoothing it over.
The manager rewrites it
This is a draft, not a review. The manager corrects what is wrong, adds the context the tools never captured, and makes the actual call on rating and direction. Nothing reaches the employee that a manager has not written into their own words.
Build it from a prompt
Point the agent at the cycle and it prepares the raw material for every manager.
Managers show up to calibration with reviews built on the same kind of evidence, covering the same period, in the same structure — which is what makes calibration mean anything. And the employee gets feedback that names what they actually did in March, not a paragraph of adjectives written under deadline.