Agent Skills
Portable, job-shaped skill packs that teach AI coding agents how to use Panda correctly.
Agent Skills are coming soon and aren't published yet, so treat the shape below as a preview. For AI agent support that works today, see LLMs.txt and MCP Server.
The problem it's meant to solve
AI coding agents are a normal part of how Panda projects get written today, but without something grounding them,
agents commonly reach for the wrong instincts: emitting Tailwind class strings or Emotion-style patterns in a Panda
project, inventing raw hex or spacing values instead of using tokens, or wiring a design system by hand
(include: ['…/panda.buildinfo.json']) instead of the newer dedicated mechanism. Community skill packs for Panda
exist today but are thin and unofficial, while some other libraries already ship official ones.
How this differs from LLMs.txt and MCP
Panda already has two other AI-facing mechanisms, and Agent Skills is meant to be a third, distinct one, not a replacement for either:
- LLMs.txt hands an agent the documentation itself, a static content dump for context.
- MCP Server answers "what exists right now" in your specific project, live token, recipe, and pattern lookups.
- Agent Skills would answer "how do I do this correctly," procedural, decision-oriented guidance (which API to reach for, what not to do), read once per task rather than queried live.
The planned skills
Rather than one large reference document, the plan is six small, job-shaped skills, so an agent loads only the one relevant to the task in front of it:
| Skill | Job |
|---|---|
panda-styling | Write or edit styles in application or component code |
panda-recipes | Build or change recipes, slot recipes, or cva/sva |
panda-tokens | Author or extend theme tokens and conditions |
panda-setup | Install, initialize, and wire up a build tool |
panda-design-system | Publish or consume a design system |
panda-migrate-v2 | Upgrade from v1, or navigate a v2 beta |
Each skill is planned to stay short (roughly 200 to 300 lines), lead with when an agent should reach for it, and include short do/don't examples rather than long prose, along the lines of what skills.sh (opens in a new tab) already hosts for other libraries.
Installing skills (once published)
The install experience is expected to follow the same shape as any other package on the open skills ecosystem:
npx skills add <owner>/<repo>@panda-stylingWhere exactly Panda's skills will be published from (the main repo, or a dedicated skills-only repo) isn't finalized, so treat the exact install path above as illustrative rather than a real command to run today.
See also
- LLMs.txt and MCP Server for the AI-facing tooling that exists today.
- Upgrading to v2 for the broader context this feature is part of.