forge-buildlisted

# forge-build Run the build loop for ONE unit, end to end, without scope drift. This is the execution engine of the methodology: the spec defines the work, the tests (written first, from the spec) define done, and this loop does exactly that work and nothing more. ## Preconditions - The project has `context/` and an entry point (`CLAUDE.md`/`AGENTS.md`). - The target unit has a spec at `context/specs/NN-feature-name.md`. If it doesn't, stop and tell the user to run `forge-spec` first. - **Parallel mode check**: if this session runs in a linked worktree (`git rev-parse --git-dir` ≠ `--git-common-dir`), verify the unit's claim in `$(git rev-parse --git-common-dir)/forge-claims/<NN>` names THIS worktree — unclaimed or claimed elsewhere is a hard stop (route to `/forge-worktree`). The close step then follows close-unit.md's **Parallel mode** section (own-lines tracker edits; digest/index deferred to main). ## Argument Text after the command selects the unit (e.g. `/forge-build unit 04` or `/forge-build the auth pages`) — match it against the build plan. No argument → read `context/progress-tracker.md` and pick the "Next Up" unit. Either way, confirm the target unit with the user before starting. **Model recommendation:** if the unit's build-plan line carries `[complexity: high]`, say so at confirmation and recommend switching to a stronger model for this unit (`/model opus`) before implementing — a failed verify loop plus a debug session costs more than the mode