Gestion des compétences Máddu

---

name: maddu-skill description: List, search, create, apply, or delete Máddu skills (reusable SKILL.md recipes). maddu-version-min: 0.18.0

The operator wants to work with skills: $ARGUMENTS.

Output discipline (read carefully):

Parse $ARGUMENTS as <verb> <args>. Supported verbs:

• list → ./maddu/run skill list (print all skills with one-line descriptions).
• search <query> → ./maddu/run skill list + filter by query in the client; surface matches.
• show <id> → ./maddu/run skill show <id>.
• add / create → ./maddu/run skill add --title "..." [--when "..."] [--tags a,b] [--body "..."].
• apply <id> → read the skill body via skill show, inline the recipe into the current turn, then follow it step-by-step.
• from-slice <slice-id> → ./maddu/run skill from-slice --slice <id> to extract a new skill from a finished slice's hindsight.
• delete <id> → ./maddu/run skill delete <id> (ask before executing — skills are non-trivial to recreate).

If $ARGUMENTS is empty, default to list.

For list, show, search, add/create, from-slice, and delete: after the bash call returns, re-print the command's complete output inside a fenced markdown code block (```) in your reply. The operator's bash-output view collapses long output behind a … +N lines (ctrl+o to expand) affordance — the only way they actually see the skill list or details is if you echo them back inside a code fence. Do not summarize, paraphrase, or omit rows.

apply is an exception: the recipe is meant to be inlined and executed step-by-step, not re-printed for the operator.

Discipline:

• Skills are agent-facing recipes. They are documentation, not executables — apply means "follow the steps", not "run a script".
• Tell the operator you picked /maddu-skill <verb> and surface the result; if the skill recommends a different slash command, dispatch it after one confirmation.