Skip to main content
Agent Skills teach agents like Claude Code and Codex the HeyBee workflow, so you can say “evaluate these two checkpoints” and get back a voting link and, later, a decision. Five Skills are available: heybee for the full lifecycle, plus heybee-create-evaluation, heybee-collect-feedback, heybee-analyze-results, and heybee-run-rlhf for focused tasks.

​
Setup

The Skills ship inside the heybee package, so installing the SDK installs them too.
Then point your harness at the Skills. heybee skills path prints where they live, and heybee skills list shows the five available:
Your agent uses the credentials from heybee auth login through the CLI and never sees them.

​
What agents can and cannot do

Skills drive the structured CLI only; they never scrape the app or improvise HTTP calls. Reads, drafts, uploads, and exports run freely. Anything with real-world consequences stops and asks you first: publication, paid spend, invitations, role changes, key creation, deletion, and account changes. Harnesses without Skill support can run the same workflows through the CLI with --json.