Skip to main content

create-skill

Create a new Kai skill from a single natural-language request, then install it into the sandbox so it's invokable as /<id>.

Informations de source

Dépôt
SimonSchubert/Kai
Dernière activité de la source
31 mai 2026 à 19:23
Langue détectée de SKILL.md
anglais
Étoiles
1 286
Forks
198

Options d'installation

Le prompt qui vérifie d'abord la source est sélectionné par défaut. Vous pouvez passer à une commande directe ou télécharger une copie locale.

Vérifiez les fichiers source

Lisez SKILL.md et les fichiers associés affichés par SkillsMP avant de décider de l'installer.

Affichage de SKILL.md

SKILL.md
Instructions source · Aperçu en lecture seule
name
create-skill
description
Create a new Kai skill from a single natural-language request, then install it into the sandbox so it's invokable as /<id>.
You are creating a new skill for the user. Skills are short instruction sets that get injected into the system prompt when the user types `/<skill-id>` in chat. Treat the user's message as the full spec — don't run a question loop unless the request is too vague to act on (in that case ask exactly **one** clarifying question, then proceed). ## Steps 1. **Pick an id.** Derive a short kebab-case id from the user's request. Rules: lowercase letters, digits, and hyphens only; ≤ 64 chars; descriptive but terse (e.g. `summarize-url`, `daily-standup`, `recipe-from-photo`). 2. **Check for conflicts.** Run `execute_shell_command` with `ls /root/skills/ 2>/dev/null` to see existing ids. If your chosen id is taken, append `-2`, `-3`, etc. until it's free. 3. **Write a tight description.** ≤ 1024 chars, non-empty, single line, written as a third-person summary of what the skill does. This shows up in the slash-command menu and Settings. 4. **Draft the body.** Markdown instructions addressed to a future model invocation — clear steps, any constraints, and how to interpret args after the slash command. Keep it focused; every byte ends up in the system prompt when the skill is active. 5. **Install it.** Run `execute_shell_command` once with this exact shape (substitute `<id>`, `<description>`, and the body): ```sh mkdir -p /root/skills/<id> && cat > /root/skills/<id>/SKILL.md <<'SKILLEOF' --- name: <id> description: <description> --- <body markdown here, verbatim> SKILLEOF ``` Use the quoted heredoc delimiter (`<<'SKILLEOF'`) so YAML/markdown special characters are preserved exactly. 6. **Confirm.** In one short message, tell the user the chosen id, the description, and that they can invoke it now as `/<id>`. Mention they can ask you in normal chat to tweak it later (you'll `cat` and rewrite the file). ## Worked example User: `/create-skill summarize any URL the user pastes into 3 bullet points` You pick id `summarize-url`, check `ls /root/skills/` (free), draft: - description: `Fetches a URL and summarizes the page in 3 concise bullet points.` - body: a few lines telling the model to read the URL from the user's message, fetch it via `execute_shell_command` (e.g. `curl -sL`), and reply with exactly three markdown bullets. Then run the heredoc, confirm: "Installed `/summarize-url` — paste a URL after it to try." ## Notes - Don't propose bundled scripts/files unless the user specifically asked for them; SKILL.md alone covers most needs. - If the user's request describes something better suited to existing tools (e.g. "summarize this PDF" → already covered by file uploads), say so and offer the skill anyway only if they confirm. - The skill list auto-refreshes after this turn — no restart needed.
Voir sur GitHub