initlisted

Bootstraps a complete, project-specific Claude Code harness — a lean CLAUDE.md plus the full .claude/ tree (hooks, agents, commands, skills) — from a one-paragraph description of a project idea, for any stack (web, CLI, backend, mobile, library, data/ML, etc). Use whenever the user substantively describes a new project or app they want to build — even with no mention of Claude Code, a harness, or setup at all, e.g. "I want to build a command-line tool that..." or "We're building a web app with..." — not only when they explicitly ask for Claude Code to be set up. Also use when they explicitly ask for a harness/starter kit/project scaffold, or explicitly invoke this skill. A one- or two-word description with no real substance ("I want to build an app") doesn't have enough to profile yet — ask what kind of project it is first rather than invoking this.
sagaga144/claude-harness-bootstrap · ★ 1 · Data & Documents · score 66

Install: claude install-skill sagaga144/claude-harness-bootstrap

# Bootstrap a project harness Read `BOOTSTRAP.md` in this skill's folder and follow it exactly, in order — it is the actual instruction set, not a summary to paraphrase. It walks through: 1. Profiling the project from the user's description (and the repo, if one exists). 2. Writing a lean, project-specific `CLAUDE.md` — every section derived, no placeholders. 3. Handing off to `HARNESS_REFERENCE.md` (same folder) to build the rest of `.claude/` (hooks, agents, commands, skills), verify it, and — once it's run for a real session — publish an Operator's Manual artifact documenting it. Don't ask the user to approve each step. Do the work, then report back what got built and why, per `BOOTSTRAP.md`'s closing instruction.