project-bootstraplisted

# Project bootstrap Sets up the standard AI-agent operating structure for a project, Claude Code first (other tools via AGENTS.md). The skill is **self-contained**: every asset it generates has a bundled template under `templates/` (settings.json, hooks, agents, rules, commands, root files, docs). A user-supplied reference repository is OPTIONAL enrichment - when one is given, prefer copying its battle-tested files and adapt; when none is given, generate everything from the bundled packs. This SKILL.md is the navigation layer. Detailed procedures live in `reference/`; copyable assets live in `templates/`. Read each file when you reach that step - do not improvise a step that has a reference doc. ## Three modes - decide first | Mode | When | What changes | |------|------|--------------| | **Greenfield** | Empty or near-empty repo | Generate the full structure from intake answers | | **Brownfield (standardize existing source)** | Repo already has code, and possibly a partial `.claude`/docs | Run codebase analysis FIRST (`reference/codebase-analysis.md`), derive most intake answers from evidence, then RECONCILE - never clobber - the existing setup into the standard shape | | **Audit / assessment** | The repo(s) are ANALYSED but never MODIFIED by agents - e.g. a read-only security/performance audit, possibly over multiple product repos under one workspace root | Build a CONTROL PLANE (root `.claude` + `docs/`) beside the untouched source: agents are read-only on product code