The onboarding toolkit

Data Workers onboarding is agent-run. The toolkit is two Claude skills — packaged runbooks your onboarding engineer (or your own platform engineer) loads into Claude Code and runs inside your environment. They automate the whole journey and enforce the same honesty rules as the platform: nothing gets called done that wasn’t verified live in the session.

Skill	What it does	Download
onboard-data-workers	The end-to-end onboarding: install the agents in your coding agent, connect your systems, live-verify every connection, warm the context graph, stand up the Spellbook Data Catalog, produce the onboarding report.	onboard-data-workers.zip
eval-benchmark	Builds an eval set from your team’s historical queries, freezes today’s correct answers, benchmarks the agents against it, and hill-climbs the score during onboarding. Doubles as your regression suite.	eval-benchmark.zip

How to install a skill

Unzip into your Claude Code skills directory (project .claude/skills/ or personal ~/.claude/skills/):

unzip onboard-data-workers.zip -d .claude/skills/
unzip eval-benchmark.zip -d .claude/skills/

Then, in Claude Code:

/onboard-data-workers

Claude walks the phases with you — intake, install, connect, verify, graph warm-up, console, eval baseline, report — pausing at every checkpoint.

Who runs what

At kickoff: your named onboarding engineer runs both skills with you on the call.
After kickoff: the skills are yours — your team re-runs them for every expansion (new schemas, new teams, new workspaces) and for benchmark re-runs, without waiting on us; anything they can’t complete gets filed straight to us through the feedback pipeline.