The workflows repo is organized around three reusable building blocks. That separation is what makes the content composable instead of turning into one-off prompts.
A skill is a reusable AI task for one concrete engineering problem. It keeps the task behavior and output shape close to the work itself.
Example: rtl-cdc-linter reviews RTL crossings, classifies synchronizers, and emits a
structured CDC report before implementation.
A rule holds reusable grounding, classification, or best-practice logic so the same policy does not get copied into multiple skills and flows.
Example: rules/common/evidence-grounding.md requires explicit RTL evidence for claims,
and rules/timing/register-evidence.md defines when a timing object is trusted vs unresolved.
A flow combines several skills and rules into one larger engineering artifact with a stable handoff structure.
Example: block-level-rtl-plan turns intent into a traceable RTL handoff package, and
block-dv-plan turns that package or raw RTL into one final DV plan.