Run bd prime. It is usually hook-injected already, and it carries the command reference, the priority format, and the session protocol. The repo's own beads skill covers the execution loop. Do not restate either of them and do not work around them.

This skill only adds what bd cannot know: how a CUJ document maps onto beads.

If bd --version exits nonzero, stop. necklace requires a working bd and there is no fallback. Run it rather than checking PATH, because a broken install shim still resolves.

Skip any CUJ whose **Beads:** line already names beads or says none - done directly, and any carrying a **Blocked:** line. Say which ones you skipped and why. Only an empty **Beads:** line means there is work waiting.

For every other CUJ: one bead, or an epic with children when it is large. Label it cuj:CUJ-NN, and put the CUJ's test names in the description so whoever picks it up knows what closes it.

For each bead, write the tests its CUJ names first and watch them fail, then implement until they pass. A test that passes the moment you write it is testing nothing.

Close a bead when its CUJ's tests pass.

Export the graph and stage it, so a bead ID stays resolvable for anyone reading the repo without running bd:

bd export -o .beads/issues.jsonl
git add .beads/issues.jsonl

Both lines. Auto-export is interval-gated, so straight after a burst of creates the file is stale; bd export alone writes to stdout, and -o alone does not stage.

Then write the bead IDs into each CUJ's Beads: line in cuj.md, and append to ledger.md anything the implementation forced that the CUJ document did not anticipate.

Commit the planning directory as you go rather than at the end. What happens to the beads themselves, including whether anything is pushed, is bd's session protocol and the user's configuration. Do not override it.