One door.
The whole toolkit.
superadvisor reads a request, picks the single best specialist — a super* skill, a persona, or a gstack workflow — and loads only that. One front door instead of a catalog you keep in your head.
and verify health
Everything you'd reach for, behind one prompt.
The router doesn't do the work — it dispatches to the tool that does, and pulls in only that one. 41 targets, three ways in.
Skill tool
super* skills
Full gated workflows. Each stays directly invocable too.
Agent tool
personas
One specialist, one job. Spawned fresh for the task.
reference
gstack adapters
Faithful thin adapters of what the super* suite lacks.
Ask in your words. It picks the lane.
One primary target per turn, announced as route: <target>. When one skill obviously owns the request, it routes straight there.
Symlink it once.
Clone the repo, link it into your Claude Code skills. super* skills stay individually invocable; superadvisor just adds the front door.
# clone git clone https://github.com/cskwork/superadvisor.git # link into Claude Code ln -s "$PWD/superadvisor" ~/.claude/skills/superadvisor # then, in any session /superadvisor <request>Read the docs
SKILL.md is the router. reference/ carries the procedure. Load only what you route to.
- ↳Whole multi-step objective → super* (it owns the gates)
- ↳One atomic specialist step → persona
- ↳Deploy · safety · product-reframe · docs → gstack
gstack adapters distilled from
garrytan/gstack (MIT). super* skills by cskwork.