What you’ll learn
-
How
Jido.AI.Skill.Loader.load/1parses checked-inSKILL.mdfiles into runtime manifests -
How
Jido.AI.Skill.Registry.load_from_paths/1andJido.AI.Skill.Registry.register/1populate one deterministic registry -
How
Jido.AI.Skill.Prompt.render/2turns those registered skills into reusable prompt instructions -
Where the builder-skill catalog lives under
priv/skills/builder-*/SKILL.md -
How the same builder skills can support
Jido.AI,jido_skill, and Codex-oriented contributor workflows -
How one real workbench task for
jido_skillis assembled from checked-in builder skills
How this demo stays truthful
This page runs real skills runtime code.
-
One skill is defined as an Elixir module with
use Jido.AI.Skill. -
Two skills are loaded from checked-in
priv/skills/.../SKILL.mdfiles. -
Seven builder skills are loaded from checked-in
priv/skills/builder-*/SKILL.mdfiles. -
The demo uses those builder skills to render a real contributor workflow for refreshing the
jido_skillecosystem package coverage in this repo. - The demo renders prompts and workflow plans only. It does not call external services or mutate package repos.
No API keys, LLM providers, or network access are required for this example.
Builder catalog included here
The checked-in builder catalog currently includes:
-
builder-action-scaffold -
builder-agent-scaffold -
builder-plugin-scaffold -
builder-adapter-package -
builder-ecosystem-page-author -
builder-example-tutorial-author -
builder-package-review
The catalog is intentionally split across package-repo and workbench-repo boundaries:
- package repos own implementation modules, tests, changelog updates, and release work
- this workbench owns ecosystem pages, examples, tutorials, and contributor-facing guidance
Pull the pattern into your own app
Keep the same runtime steps in your own project:
- define stable module-backed skills for core workflows
-
add file-backed
SKILL.mdassets when you want editable runtime instructions - register those skills into the runtime registry
- render the combined prompt text once your agent or workflow needs it
- keep boundary notes inside the skill metadata so package-repo work and workbench follow-up stay explicit