メニュー
Guide for writing effective system prompts for LLM agents. Use when creating or editing system prompts for applications, agent configurations, or development tools.
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/saffron-health/libretto --skill prompting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。
Read-only Libretto workflow for diagnosing live browser state without clicks, typing, navigation, or mutation requests.
Browser automation CLI for building, maintaining, and running browser automation workflows by inspecting live pages and prototyping interactions.
Automate user-installed Electron desktop apps (Slack, Discord, VS Code, Notion, Figma, Spotify, etc.) via CDP using this repo's Libretto CLI. Use when the task is to control a local desktop app on the user's machine, not this repo's own Electron app. Triggers: "desktop Slack app", "connect to Electron app", "remote-debugging-port", "CDP desktop app", "automate VS Code desktop app".
SOC 職業分類に基づく
SKILL.md を表示中
| name | prompting |
| description | Guide for writing effective system prompts for LLM agents. Use when creating or editing system prompts for applications, agent configurations, or development tools. |
LLMs are intelligent by default. System prompts set direction and impose constraints, not explain reasoning.
Start minimal. Observe failures. Add targeted fixes. Every instruction must justify its token cost by solving a real problem.
Do not explain existing capabilities, list obvious practices, add preventive instructions, or repeat information.
Use markdown sections and paragraphs. Each section describes one behavior or constraint.
State what to do or avoid. Explain why if non-obvious. Show correct behavior with examples.
Headings up to level 3. Plain paragraphs. No bold, italics, or emojis. Code blocks for commands. Lists only for distinct enumerable items.
Wrap examples in <example> tags with user/assistant prefixes. One pair per tag.
<example>
user: What's the capital of France?
assistant: Paris
</example>
Use brackets for tool actions instead of showing invocations:
<example>
user: Find all TODO comments
assistant: [searches codebase]
Found 3 TODOs: ...
</example>
Behaviors the model gets wrong by default. Domain constraints. Output format requirements. Safety boundaries. Tool integrations.
Reasoning instructions. Problem-solving approaches. Common sense behaviors. Ethical guidelines. Capability descriptions.
Start minimal. Test with real inputs. Identify failures. Add targeted fixes. Remove unnecessary instructions.
Track which instructions prevent which failures. If you cannot identify the specific problem an instruction solves, remove it.
Consult references/ for model-specific patterns: