Guide · Agent skill
How to use the First Ask Task Refiner agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The first-ask skill (named First Ask Task Refiner) changes how your AI agent starts a task. Instead of jumping straight into code, the agent asks you specific questions about scope, deliverables, constraints and success criteria first. This guide explains what the skill does, how to install it, how to trigger it, and what it requires, so you can decide whether it fits your work.
First Ask Task Refinergithub/awesome-copilot · MIT Download the skill Quick start
- 1Install Joyride. Install the Joyride extension, which the skill requires for its input tool.
- 2Download the ZIP. Download the first-ask ZIP from this page.
- 3Unzip the skill. Unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder.
- 4Or upload in the Claude apps. Go to Settings → Capabilities → Skills and upload the skill there instead.
- 5Give a task. Describe a task in plain language, especially one that is large or vague.
- 6Answer the questions. Reply to each question through the input prompts with specific details.
- 7Confirm and let it work. When asked for further input, say you have none; then review the plan and todo list as the agent starts work.
What does the first-ask skill make the agent do?
Agent skills are folders of instructions that an AI agent loads when they are relevant. The first-ask skill tells the agent to refine its understanding of a task before carrying out that task. It uses the joyride_request_human_input tool to ask clarifying questions about the task scope and objectives, expected deliverables and success criteria, and technical and procedural requirements.
The agent can also explore your project with the tools it has, or do web research if something needs it, to understand the task better. It keeps refining until you say there is nothing more to add.
- Asks specific clarifying questions through the Joyride input tool
- Defines deliverables and success criteria
- Explores the project or researches the web when needed
- Shows a plan with minimal repetition, creates a todo list, then starts work
Install the first-ask skill
Download the first-ask ZIP from this page. The skill is a single file, SKILL.md, inside the package.
For a personal install, unzip the package into ~/.claude/skills/. To make it available only in one project, unzip it into that project's .claude/skills/ folder instead.
If you use the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/
- Single project: .claude/skills/ inside your project
- Claude apps: Settings → Capabilities → Skills
Install the Joyride extension first
The skill requires the Joyride extension. The skill description says so, and its core workflow depends on the joyride_request_human_input tool that Joyride provides. Without Joyride, the skill will not function as designed.
How do you trigger the skill with natural requests?
You do not need a special command. The skill is meant to run when you give the agent a task, so describe the work as you normally would. The agent should then start asking questions through the input tool before it plans or edits anything.
A vague request is a good test case, for example: "Refactor the login code." The agent should ask what the refactor is meant to achieve, which files are in scope, and what done looks like.
- Give a task that is large, unfamiliar, or open to interpretation
- Answer the questions in the input prompts with specifics
- When asked if you have further input, say no once you are satisfied
What happens during a session?
The agent asks questions, you answer, and it keeps going until you confirm there is nothing more to add. Only then does it present a plan, create a todo list, and start work. The plan is meant to be concise, without repeating itself.
- Refine: questions about scope, deliverables, constraints and success criteria
- Confirm: the agent asks whether you have any further input
- Plan: a concise plan is shown to you
- Work: a todo list is created and the task begins
Safety notes: scripts and network
The skill ships only SKILL.md. It does not run scripts and does not use the network, according to the skill's listing. Its interaction happens through the Joyride input tool. Web research only occurs if the task requires it, as the instructions describe.
- Runs scripts: no
- Uses network: no (web research happens only when the task needs it)
Tips for getting the most from it
Answer questions with concrete details such as file names, limits, and the expected output. Short answers lead to more rounds of questions.
For quick, well-specified tasks, the extra questions can slow you down. Consider not using the skill for those.
- Name the files, systems, or constraints involved
- State what success looks like in one sentence
- Tell the agent to stop asking once you have nothing to add
What people use it for
- Refining a vague request like "clean up the API" into defined scope and deliverables before edits begin
- Setting requirements for a large or unfamiliar codebase before an agent changes files
- Gathering constraints and success criteria for a feature before a team starts work
- Having the agent research a topic on the web before it plans the task, when the task needs it
Frequently asked questions
Is the first-ask skill free?
The skill is published under the MIT licence on github/awesome-copilot. The Joyride extension is a separate requirement, and this page does not state its pricing, so check its listing before installing.
What do I need to use the first-ask skill?
You need the Joyride extension, because the skill's core workflow depends on the joyride_request_human_input tool. You also need an agent environment that loads skills from ~/.claude/skills/ or .claude/skills/, or the Claude apps' Skills setting.
How do I install the first-ask skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use or a project's .claude/skills/ folder. In the Claude apps, you can upload it under Settings → Capabilities → Skills.
Does the first-ask skill run scripts or use the internet?
According to its listing, it runs no scripts and uses no network. It only does web research if a task needs it, as its instructions describe.
Why does the agent keep asking me questions?
The skill is designed to refine the task until you have no further input, so it pauses repeatedly for answers. If you want the agent to work without interruption, this skill is not the right fit.
What should I use instead if I don't want clarifying questions?
If you have quick, well-specified tasks or want autonomous work without pauses, use the agent without this skill. The source does not compare it with other skills, so judge alternatives on their own listings.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use first-ask if you often give vague requests, work on large or unfamiliar tasks, or want a structured requirements step before an agent edits code, and you can install Joyride. Skip it for quick, clear tasks, for setups without Joyride, or if you want the agent to act without being interrupted.
Download the skill