Skill Commands
Skill commands run a skill, a reusable prompt, as a turn of its own, and write new ones from what the agent just did.
Run a skill
A skill is a prompt, a folder holding a SKILL.md: yours in the
agent's skills/, or the repository's in .agents/skills/ or
.claude/skills/ of Home and each folder above it up to the
repository's root. /skill:deploy staging asks a turn of its own with
what the skill's SKILL.md says, then what you said after its name.
/skill:list, or /skills, names each skill and whose it is, and the
skills section of the instructions names the repository's beside
yours. [commands] skills turns both on.
Which runs by a bare name
/deploy runs a skill too. Yours runs by its name even where a command
has it, so a skill of yours named review replaces Link's command of
that name. A repository's runs by its bare name only where no command
has it, so a cloned repository can't take over /stop. Where yours and
the repository's share a name, yours runs. /skill:<name> always runs
the skill, but the skill namespace's own commands, /skill:list and
/skill:write, stay theirs.
Learn a skill
/skill:write asks a turn of its own, with Link's learn skill: the agent
writes a skill from what it just did, skills/<name>/SKILL.md, or adds
to one it has, or fixes one that misled it, and says which. What you
send after it, /skill:write the deploy, says what to keep. Unlike the
others, it's answered by the model.
Where it's offered
It writes with its file tool, so it's offered only to an agent whose
file tool is on ([tool.file] enabled, which link:personal turns
off) and whose sandbox file lets it write skills/ by name (lnk agent allow --write skills). Elsewhere /skill:write is text for the model,
and /help doesn't name it.