Module skill 

The Agent Skills this crate ships, compiled in: what jev add skill writes into a project, and what dx writes into one when it turns jev on.

There are two, because there are two ways an agent is given jev and a skill that teaches the wrong one is worse than none. Flavour::Command teaches the command — jev -q questions.json -f …, the shell, jq — and Flavour::Tool teaches the same substance for an agent whose jev is a tool it calls with JSON, which has no shell to run a command line in. Keep the two in step when either changes.

They live here rather than in cli/ so that a crate with only the command feature — dx, which offers both shapes — can write them without the CLI’s dependencies.

Enums

Flavour

Which jev a skill teaches.

Constants

FOLDER

The folder a skill is written to, under .agents/skills or .claude/skills. Both flavours share it, since the frontmatter names both jev.

Functions

refuse_symlinks

Refuses a destination reached through a symbolic link. Walks what exists of file below root, since it is the parts already on disk that could redirect a write; std has no way to open a path without following links, so they are found and refused rather than avoided.