# Swarm Init

> Start a new project connected to the SwarmVault — register it, wire platform adapters, offer git init, set vault-as-default. Use when starting a new project, connecting a project to the vault, or when the user asks to set up SwarmVault in a directory.

- Skill: `anmarhani/swarm-init` (Agent Skill)
- Install (CLI): `npx skillmds@latest add anmarhani/swarm-init`
- Raw SKILL.md: https://api.skillmd.com/api/skills/anmarhani/swarm-init/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: AnmarHani (https://skillmd.com/u/anmarhani)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/anmarhani/swarm-init

---


# swarm-init — new project onboarding

Idempotent: safe to re-run on an initialized project (repairs missing pieces, changes
nothing else).

## Steps

1. **Vault exists?** `swarmvault.py doctor`. If no vault: offer `swarmvault.py init`
   first. Mention once: opening the vault folder in Obsidian gives the graph and
   backlinks — optional, never required (`docs/obsidian-guide.md`).
2. **Ask** (only what is genuinely the user's):
   - Project name (default: directory name).
   - Isolation: shared (cross-project queries see it) or isolated?
   - Which agentic platforms to wire — the tools they actually have (`claude-code`,
     `codex`, `gemini`, `opencode`, `droid`, `cursor`, `copilot`, `amp`, or any other agent
     CLI by name), and global or per-project.
   - Git init, if no repo (recommended — the flow's traceability uses commits; optional).
   - Optional orchestration for this project? Ask it as a **pace** — off / budget /
     balanced / fast — plus whether workers may edit code. Never ask for a worker count
     or a model name (swarm-orchestrate explains why) — but if they enable it, DO ask
     their plan and usage limits per platform: you cannot read those, and without them
     every usage number you print is invented.
   - If not already set: make the vault the default for new projects? → save
     `"auto_init_new_projects": true` in `~/.swarmvault.json`; future inits skip the
     ceremony.
3. **Do it in one command** — it registers, wires every named platform's own instruction
   file (plus Claude Code skills and hooks), syncs, and verifies:

   ```bash
   swarmvault.py setup --platforms <list> --scope global|project \
     [--name <name>] [--isolated] [--autonomy off|budget|balanced|fast] [--read-only]
   ```

   Already-registered project, or only need part of it? `register` / `wire --platforms …`
   / `sync --quiet` / `doctor` are the same steps individually. Never leave the user
   manual file edits.
4. **Disclaimer** (show once, verbatim):
   > SwarmVault stores project knowledge as plain files and runs local scripts. Auditing
   > your environment, installed packages, and the data you place in the vault is your
   > responsibility. Isolation flags are cooperative filtering between agents, not a
   > security boundary.
5. Point forward: `/swarm-flow` starts the SDLC (first stop: swarm-spec).

## Edge cases

- Directory inside an already-registered project → warn (nested projects) and require
  explicit confirmation.
- Name collision → `register` refuses; pick another name.
- Existing marker with a different name → repair with `register --repair` or honor the
  marker; never two names for one directory.

---
*Influences: the original Claude Vault setup — see CREDITS.md.*