Skills
A skill is a directory with a SKILL.md file: ~/.e/skills/<name>/SKILL.md.
It follows the open SKILL.md convention, so other tools can read the same
files.
---
name: release
description: how to cut a release of this project
---
Step one …
e reads skill files on each use. Add or edit a skill and it is live immediately.
On this page
How the model uses a skill
e advertises a catalog in the system prompt. Each entry holds the skill's
name, its description, and the path to its SKILL.md. When the task matches,
the model reads the body in with the ordinary read tool. Until then, only
the descriptions stay in context.
A description: may span lines. Use a > or | block scalar, or an indented
continuation. It folds to one line in the catalog and the picker.
Insert a skill by hand
The $ picker inserts a skill by hand.
disable-model-invocation: true keeps a skill out of the catalog. Only the
$ picker reaches it.
Package skills
An installed package contributes its skills/
directory the same way. A global skill shadows a package's skill of the same
name. The $ picker labels package skills Package.
Repo-local skills
A trusted repository can carry its own skills in .e/skills/, as
<repo>/.e/skills/<name>/SKILL.md, in the same format as above.
These skills load only after /trust, like the repo's AGENTS.md. They shadow
a global skill of the same name, because the closer context wins.