harness-setuplisted

<!-- managed-by: harness-setup --> # Harness Setup Deterministic discovery/plan/reconcile/validate engine, shared with the Claude Code adapter of this same capability. Do not reimplement its logic inline — call it, then narrate its output. The engine never overwrites user-owned content silently; your job is to relay its report and, when it reports a `Needs confirmation` section, get the user's decision before re-running with `--approve`. ## 1. Locate the engine The canonical engine lives at `~/.harness-setup/engine/cli.js` (a single global install shared by every project — do not copy it into this repo). Resolve `~` to the user's home directory for the current OS before running it (on Windows this is `%USERPROFILE%\.harness-setup\engine\cli.js`). This install is self-contained (copied, not linked to any source checkout) unless the user explicitly chose `--link` dev mode, so it should not go missing on its own. If the file is missing, tell the user the global install seems broken or was removed, and give them the fix instead of just stopping: ``` npx github:ferronicardoso/harness-setup ``` (or, if they have a local clone: `node <clone-path>/bin/install.js`). Do not try to recreate the engine from memory — reinstalling is a one-line command. Requires Node.js on PATH. If `node --version` fails, tell the user Node.js is required and stop. ## 2. Map the user's request to flags Run from the project root (or pass `--root <path>` if the user names a different directory): -