harness-bootstrap

# Harness bootstrap Sets up the standard operating harness for a repo: a roster of agents that know their scope and their cost, rules that load only when relevant, hooks that actually fire, and a task board that survives context compaction. **The assets are real files, not prose to retype.** Everything copyable lives under `assets/` and is installed by `scripts/scaffold.py`. Your job is the decisions - the roster, the scope, the variables - not transcribing 1,300 lines of hooks and commands. Read a reference file when you reach its step; never improvise a step that has one. ## Three modes - decide first | Mode | When | What happens | |---|---|---| | **Greenfield** | Empty or near-empty repo | Generate the full structure from intake answers | | **Brownfield** | Repo already has code, and maybe a partial `.claude/` | Run codebase analysis FIRST, derive most answers from evidence, then RECONCILE - never clobber | | **Audit** | The repo(s) will be analysed but never modified by agents; a human applies any fixes | Build a read-only control plane beside untouched source. See [`reference/audit-mode.md`](reference/audit-mode.md) | Selection rule: if agents will never modify the source (a human applies fixes), you are in **audit** mode, however much code exists. Otherwise, any source code at all means **brownfield**. ## Procedure **0. Codebase analysis** - brownfield and audit: MANDATORY. Greenfield: skip. Follow [`reference/codebase-analysis.md`](reference/codebase-analysis.md...