無料GitHub で公開

use-case-specification

Creates a reusable use case specification file that defines the business problem, stakeholders, and measurable success criteria for model customization, as recommended by the AWS Responsible AI Lens. Use as the default first step in any model customization plan. Skip only if the user explicitly declines or already has a use case specification to reuse. Captures problem statement, primary users, and LLM-as-a-Judge success tenets.

awslabs/agent-plugins

インストール方法を見る

含まれるファイル（1）

SKILL.md4.0 KB

SKILL.md（原文）

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Use Case Specification

Multi-turn conversation to gather use case details and produce a use case specification document.

Principles

One thing at a time. Each response advances exactly one decision or collects one piece of information.
Confirm before proceeding. Wait for the user to approve the spec before considering this skill complete.
Infer, don't interrogate. Use what's already known from the conversation. Only ask when you truly can't infer.
Do NOT ask about base model selection. Model selection is handled exclusively by the model-selection skill.

Workflow

Step 0: Check for Existing Spec

Before starting discovery, check if a *_use_case_spec.md file already exists in the project. If it does, present it to the user and ask whether they want to reuse it, modify it, or start fresh.

Phase 1: Discovery (1–3 turns)

Review what is already known from the conversation so far, then identify what is still missing. You need these three things:

What is the problem the user is trying to solve with model customization
Who will use the finetuned model and in what context
Which success criteria can be used to evaluate how well the custom model performs compared to the base model on a test set. Success criteria must be measurable by an LLM-as-a-Judge (e.g., response accuracy, tone adherence) — not things like latency or throughput.

Guidelines:

Infer as much as possible from what the user has already said
If the user gave examples, use them to fill gaps rather than asking again
Only ask clarifying questions when you cannot infer the information needed for Phase 2
If everything is already clear, say "You've given me a clear picture. I'll put together a use case specification now." and move to Phase 2.

⏸ Wait for user after each clarifying question.

Phase 2: Producing a Use Case Specification Document

Save all generated artifacts under the project directory structure defined by the directory-management skill, if available.
Synthesize the information you collected from the user into a Markdown document called [relevant_title]_use_case_spec.md containing the following fields (and only these fields):

Use case description
  - Concise problem statement + what the custom model will do
  - Field name: “Business Problem”
  - Type: String

Key stakeholders
  - Who uses the model and in what context
  - Field name: “Primary Users”
  - Type: String, comma separated if there are multiple 

Success criteria
  - A list of 3 criteria (a short name and a description) with which the user measure the success of the custom model. 
  - Field name: “Success Tenets”
  - Type: list of name-description pairs

Present the use case specification in a human-readable format as follows:

I have put together a use case specification and saved it in [relevant_title]_use_case_spec.md.

A use case specification is a design principle recommended by the AWS Responsible AI Lens.

[use case in human-readable format]

Does this match your intent?

⏸ Wait for user approval.

use_case_specification Edit Protocol

If the user requests changes pertaining to any information covered by use_case_spec.md, you must edit it accordingly and ask for confirmation again.
The user can edit use_case_spec.md directly if they want to. If the user says they've updated the file directly, read it to get the latest in your context.

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

ログインしてレビューを書く

同じリポジトリのスキル

概要と使いどころ

amazon-location-service

無料

Integrates Amazon Location Service APIs for AWS applications. Use this skill when users want to add maps (interactive MapLibre or static images); geocode addresses to coordinates or reverse geocode coordinates to addresses; calculate routes, travel times, or service areas; find places and businesses through text search, nearby search, or autocomplete suggestions; retrieve detailed place information including hours, contacts, and addresses; monitor geographical boundaries with geofences; or track device locations. Covers authentication, SDK integration, and all Amazon Location Service capabilities.

日本語の概要は準備中です。原文の説明を表示しています。