Skip to content

Skill system

Agentflow is not primarily a catalog. It is a routing product.

The user’s job is not:

learn ten internal skill names

It is:

have a real engineering task
↓
identify what is unclear
↓
choose the smallest useful reasoning tool
↓
make progress with the right output

That changes the information architecture.

The homepage is goal-first. Skill names are secondary metadata.

Every skill page answers:

What decision does this skill own?
When should I use it?
When should I not use it?
What goes in?
What comes out?
What prompt can I try?
Where can it hand off?

Recipes show composition without inventing a mandatory lifecycle.

task
↓
decision A → skill A
↓
decision B only if it exists → skill B

The required responsibilities are content loading, docs navigation, search, presentation, build proof, and static delivery.

MD/MDX
↓
Astro content
↓
Starlight
↓
Tailwind
↓
static dist/
↓
Cloudflare

No database, API service, CMS, queue, cache, authentication layer, or SSR runtime is justified by the current product contract.

SKILL.md
→ agent runtime behavior
human docs
→ discovery, usage, examples, composition

When canonical skill packages are added, docs should link to or derive stable metadata from those files instead of copying complete instructions.

bad content/config
↓
build fails
↓
production remains on the previous valid artifact

Runtime stays small:

request
→ Cloudflare edge
→ static asset