メニュー
Write a new Skill for this harness, or improve an existing one. Use when the Operator wants to teach the harness a repeatable procedure, capture a workflow, or turn a set of instructions into something reusable.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/1aifanatic/super-computer --skill skill-authorコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
Find where something lives in an unfamiliar Workspace. Use when asked where a function, symbol, config value, or piece of behaviour is defined, or when you need to understand a codebase's layout before changing it.
Write or revise README files, API documentation, changelogs, architecture notes and code comments. Use when asked to document existing code or improve writing that already exists.
Inspect, query, reshape and validate JSON with jq. Use for API responses, config files, package manifests, log lines, and turning JSON into readable tables or CSV.
SOC 職業分類に基づく
SKILL.md を表示中
| name | skill-author |
| description | Write a new Skill for this harness, or improve an existing one. Use when the Operator wants to teach the harness a repeatable procedure, capture a workflow, or turn a set of instructions into something reusable. |
A Skill is a folder containing SKILL.md: YAML frontmatter with name and description, then a markdown body. This file is itself an example — read it as the reference.
---
name: kebab-case-name
description: What this does, and specifically when to use it.
---
Only name and description sit in the model's context at all times. The body loads only when the Skill is chosen. So the description is the entire selection mechanism — it is not a summary, it is a trigger.
Write descriptions that say when, not just what:
The weak one never gets picked, because nothing in a real request looks like it.
Write to the model that will execute it, in the imperative. Give it an order of operations, concrete commands, and the traps. Skip motivation and background — it has already been chosen by the time the body loads.
Good bodies contain:
Available: ls, cat, grep, sed, awk, find, sort, wc, head, tail, cut, tr, diff, jq, pipes, redirects, loops and conditionals — plus real git (clone, status, diff, add, commit, log, branch, checkout; HTTPS only, no SSH).
Not available: node, npm, python, sqlite3, or any other native binary. A Skill whose procedure depends on running a build or a test suite cannot work here — do not write one, and say so if asked.
A Skill body is loaded into a live context window and paid for in tokens. Two hundred well-chosen lines beat a thousand thorough ones. If it is growing past that, it is probably two Skills.
Re-read the description as though you were the model seeing only it and a user request. Would you pick this Skill? If not, rewrite the description, not the body.