Skip to Content
DronesThe 8 Drones explained

The 8 Drones explained

Drones are automated, server-side auditors. When a Worker finishes a task, the Queen sends that completed task to the relevant Drones before it lands in your Hive. Each Drone inspects the change, emits findings with a severity, and the Hive’s thresholds decide whether those findings block the task or are merely advisory. Only the Drones relevant to a given change run, so a docs-only edit will not invoke the migration auditor.

There are exactly eight Drones. They run in a sandboxed runner with no access to your machine, and they report back to the Queen, which decides what to do next.

Each Drone review consumes roughly 10–30 Honey per task. Your plan tier controls how many Drones you get: Starter includes 1 Drone of your choice, while Pro and Ultra include all 8 (Ultra adds priority routing).

A11yDrone

Accessibility audit on UI changes. It checks WCAG conformance, ARIA usage, color contrast, and keyboard navigation. Triggered when a task touches user-facing UI.

ContractDrone

API and type contract drift. It verifies that schemas, public types, and API surfaces do not break downstream consumers. Triggered when a change alters a public type, schema, or exposed API.

MigrationSafetyDrone

Database migration safety. It flags destructive or locking migrations, such as column drops or non-concurrent indexes on large tables. Triggered when a diff contains a migration.

PerformanceDrone

Performance regressions. It looks at bundle size, Lighthouse scores, slow queries via EXPLAIN, and N+1 access patterns. Triggered on changes that can affect runtime or load performance.

RefactorSafetyDrone

Refactor safety. It compares public signature diffs and test-outcome diffs to catch behavior changes hidden inside a so-called “refactor”. Triggered on refactor-type tasks.

SecurityDrone

Security review. It hunts for injection vectors, leaked secrets, authentication and authorization mistakes, and unsafe dependencies. Triggered broadly on changes that can affect the security surface.

StyleDrone

Code style, lint, and formatting conformance to your project’s rules. It is the noisiest Drone, so by default it surfaces only medium-and-above findings. Triggered on most code changes.

TestCoverageDrone

Test coverage. It measures coverage delta, new-code coverage, and flaky-test detection. Triggered when a task adds or changes code that should be tested.

Where to go next

Local PTY modelSeverity & thresholds