| name | setup |
| description | Optional one-time installer of local tooling for this template (graphify, plugins, hooks, weekly ops audit). Use after /onboard, only in the Claude Code desktop app or VS Code extension. Not for claude.ai web users. |
| disable-model-invocation | true |
Local Tooling Setup
A thin, friendly wrapper around scripts/setup.sh. The script does the actual
installing (it is deterministic and safe to re-run); this skill picks the right
moment to run it and explains the results in plain language.
Do not reimplement the install steps here. All install logic lives in
scripts/setup.sh. This skill's only jobs are: confirm the platform, run the
script, and interpret the output for the user.
Step 1 — Confirm the platform
Ask where the user is running (AskUserQuestion):
- Claude Code (desktop app or VS Code extension) → continue to Step 2.
- claude.ai (web browser) → stop. The web version has no terminal, so nothing
can be installed. Tell them: all core features (onboarding, skills, memory) work
without this step. If they later open the repo in the Claude Code desktop app,
they can run
/setup then.
Step 2 — Run the installer
From the repo root, run:
bash scripts/setup.sh
The script is non-blocking by design — every step has a fallback, so a missing
tool (e.g. pipx or Node) prints a warning and the rest continues. It installs:
| Step | What it sets up | If it fails |
|---|
| graphify | Codebase knowledge-graph CLI (via pipx) | Warns; install pipx then pipx install graphifyy |
| Plugins | Curated Claude Code plugins (context-mode, claude-mem, etc.) | Skips ones already installed |
| Hooks | Copies repo hooks into ~/.claude/hooks/ | Skipped silently if unavailable |
| Settings | Merges hook + safety rules into ~/.claude/settings.json | — |
| Skills index | Refreshes skills/README.md (skills already auto-load) | Non-blocking |
| Submodules | Initializes token-dashboard | Warns if unreachable |
| Weekly audit | launchd agent for the weekly ops audit (Mac only) | Skipped on non-Mac |
Step 3 — Report
Read the script's output and tell the user, in plain language:
- ✅ what installed successfully and what it now unlocks
- ⚠️ what was skipped or warned, and the one command to fix it if they want it
- That the repo is fully usable regardless — this step is purely additive
Never present this as "setup failed." Worst case, some optional tools were skipped.
Notes
- Manual command (
disable-model-invocation): runs only when the user types
/setup, never automatically, because it mutates the global ~/.claude/ config.
- Safe to re-run anytime — re-running picks up updated hooks and newly available tools.
