--- name: setup-macrostack description: Set up or reconcile Macrostack in a fresh or existing project after installing the skills. Establish preferences, record locations, workflow pointers and supporting-tool availability without starting product development or overwriting existing agreements. --- # Set up Macrostack Make this project ready to use the installed workflow. Read [host integration](references/bundle/_shared/host-integration.md), [preferences](references/bundle/_shared/project-preferences.md), [workflow](references/bundle/_shared/development-workflow.md) and [verification policy](references/bundle/_shared/verification-policy.md). Setup is optional when equivalent configuration already exists. 1. Inspect applicable agent instructions, existing preference/conventions files, product records, manifests and lockfiles. Locate installed Macrostack skills and their version when available. Do not assume a host directory or load another project's product brief. 2. Reuse the project's established document locations. Otherwise use `docs/PREFERENCES.md`, `docs/specs/`, `docs/design.md` and `docs/ROADMAP.md`. Do not create empty specs, design, roadmap or phases during setup. 3. Apply inherited preferences to new unconstrained choices: pnpm for JS/TS, shadcn/ui for React web and the applicable supporting UI skills. Preserve existing stacks and lockfiles. Ask only about material conflicts, constraints or choices not already agreed. Record overrides and their rationale using the [preferences template](references/bundle/_shared/assets/project-preferences.md), not a duplicate copy of all defaults. 4. Merge a short Macrostack section into the project's existing host-loaded instruction file. Identify the installed workflow entrypoints and actual reference locations; preserve all unrelated instructions. Never create competing instruction files blindly. Avoid committing machine-specific plugin-cache paths: prefer skill names plus instructions to resolve bundled references from the loaded skill. Explain the cycle, requested-stage boundaries, preferences location and where configured checks live. 5. Check availability of selected supporting skills, required analyzers and browser tooling for the project's languages/frameworks. Record source/version when observable, otherwise unknown. Distinguish available, applicable-but-missing and not applicable. Reuse existing conventions; missing checks become owned setup work when planning delivery. Do not install dependencies, migrate tooling or waive checks merely because setup was invoked; carry out installation only within actual authorization. Missing frontend tools do not block backend work or requirements discovery. 6. Record any supplied model preferences and existing local Git authority. Unknown model availability stays unknown until the host can verify it. Do not probe by launching workers, granting permissions or running product code during setup. 7. Reconcile the records. A repeat invocation should change only newly supplied choices or detected drift; preserve local edits and never append duplicate sections. Return a short result: configured files, inherited defaults/overrides, missing capabilities, and the next skill for the user's requested stage. No application implementation, commits, external publication or cleanup is implied.