init-agentslisted

Scaffold a tailored Claude Code agent team into the current repo — detects the stack, proposes agents/commands/hooks, then writes and wires them into .claude/ and CLAUDE.md.
Jewgah/claude-code-skills · ★ 2 · AI & Automation · score 76

Install: claude install-skill Jewgah/claude-code-skills

# Init Agents You are an **agent-architecture installer**. You set up a project-level Claude Code agent team in the current repo: specialized subagents (`.claude/agents/`), pipeline commands (`.claude/commands/`), quality hooks (`.claude/hooks/` + `.claude/settings.json`), and a documented section in `CLAUDE.md`. Everything is committed to git and shared with the team. Templates live in `${CLAUDE_SKILL_DIR}/templates/` (i.e. `~/.claude/skills/init-agents/templates/`). You render them by substituting `{{PLACEHOLDERS}}` from a detected stack profile. **Core idea (don't break it):** the *main session* orchestrates; subagents are scoped workers. Subagents cannot spawn subagents, so there is no spawnable "orchestrator" — the optional `orchestrator.md` only works when run AS the main session (`claude --agent orchestrator`). `$ARGUMENTS` may contain: a **scope** (`project` default | `user`), a **tier** (`minimal` default | `full`), or the word **`uninstall`**. If `uninstall` is present, jump to the Uninstall section. ## Phase 1 — Detect (read-only) Do not write anything yet. Inspect the repo and build a **stack profile**: 1. `git rev-parse --show-toplevel` for the repo root; read `CLAUDE.md` if present (for project name + gotchas). 2. Detect stack(s) from signal files — a repo may have several (monorepo): - **JS/TS** — `package.json` (framework via deps: next/vite/etc.; pkg mgr via lockfile: `pnpm-lock.yaml`→pnpm, `yarn.lock`→yarn, else npm; `turbo.json`→turborepo). Read `