Guide · Agent skill
How to use the Deep Interview agent skill (deep-interview)
Updated · written with AI from the project's own documentation and checked against our catalogue data
Deep Interview is an agent skill that makes an AI agent question you before it starts building. If you have a vague idea, it asks targeted Socratic questions that expose hidden assumptions, scores how clear the requirements are, and waits for your explicit approval before execution. This guide explains what the skill does, how to install it, how to trigger it, and when to skip it.
Deep InterviewYeachan-Heo/oh-my-claudecode · MIT Download the skill Quick start
- 1Download the ZIP. Get the deep-interview ZIP from the skill's page on this site.
- 2Unzip into a skills folder. Unzip it into ~/.claude/skills/ for personal use, or into .claude/skills/ in your project for project-only use.
- 3Or upload in the Claude apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the skill.
- 4Confirm the plugin is present. Make sure the oh-my-claudecode plugin is installed, since the skill references its agents.
- 5Describe your vague idea. Use a natural request such as "interview me" or "I have a vague idea" with a short description of the goal.
- 6Answer one question per round. Respond to each question and watch the ambiguity score and the weakest dimension it names.
- 7Approve the scoped spec. Once the score passes the threshold, explicitly approve the scoped execution path. The spec is saved under .omc/specs.
What an agent skill is and what Deep Interview makes the agent do
An agent skill is a folder containing a SKILL.md file that gives an agent runtime such as Claude Code a set of instructions for a specific task. When the request matches the skill, the agent follows those instructions.
Deep Interview replaces a vague idea with a written specification. The agent asks one question at a time, aimed at the weakest clarity dimension, and scores ambiguity after every answer. It will not proceed until the score passes the threshold for the run and you explicitly approve a scoped execution path.
The output is a spec file saved under .omc/specs rather than code changes. The source names the file pattern .omc/specs/deep-interview-{slug}.md.
- Asks Socratic questions that surface assumptions you did not state
- Shows an ambiguity score after each answer
- Writes a spec file under .omc/specs instead of editing code
- Pauses for your approval before any execution starts
Install the deep-interview skill
Download the ZIP from the skill's page on this site. Then choose one of two install locations depending on how you want to use it.
For a personal install available in all projects, unzip the file into ~/.claude/skills/. For a project-only install, unzip it into the .claude/skills/ folder inside your project.
If you use the Claude apps instead, upload the skill under Settings → Capabilities → Skills.
- Personal: unzip into ~/.claude/skills/
- Project only: unzip into .claude/skills/ in the project root
- Claude apps: Settings → Capabilities → Skills, then upload
Check the requirements before you rely on it
The skill is part of the oh-my-claudecode plugin and references its agents, including explore, plan and autoresearch. It is meant to run in Claude Code or a compatible SKILL.md agent runtime. Without the plugin, the skill's references to those agents will not resolve.
The skill's threshold is read from settings. Deep Interview reads omc.deepInterview.ambiguityThreshold from ~/.claude/settings.json and from ./.claude/settings.json, with the project value taking precedence. If neither is set validly, it uses the default 0.2 and announces where the value came from.
- oh-my-claudecode plugin installed
- Claude Code or a compatible SKILL.md runtime
- Optional: omc.deepInterview.ambiguityThreshold in settings.json
Trigger the skill with natural requests
You do not need a command. Describe the situation in plain words. The source lists phrases that should activate the skill, such as "deep interview", "interview me", "ask me everything", "don't assume" and "make sure you understand". It also responds to "I have a vague idea" and "not sure exactly what I want".
The skill accepts an optional argument hint: [--frontier] [--autoresearch] followed by your idea. The source describes --frontier as a mode that asks the whole settled frontier per round, and --autoresearch as a setup lane for the autoresearch skill.
- "Deep interview: I want a way for users to share their dashboards."
- "Interview me before you touch anything. I have a vague idea about caching."
- "Don't assume. Make sure you understand what I need for the export feature."
Answer the questions and approve the spec
Expect one question per round. Each question is worded so that answering yes accepts the recommended answer. The agent names the weakest dimension, states its score and gap, and explains why the next question targets it. Gathering codebase facts happens before it asks you about them, and brownfield questions cite the file, symbol or pattern that prompted them.
When the score is under the threshold, the agent writes the spec. Execution still waits for your explicit approval of a scoped plan. If you say to skip the questions without an execution path, the source says the agent ends the interview and writes a pending approval spec rather than changing files.
- Answer one question per round
- Read the ambiguity score shown after each answer
- Approve the scoped plan explicitly before execution
Know when not to use it
Deep Interview adds delay when the request is already specific. If your request includes file paths, function names or acceptance criteria, the source says to execute directly. Quick single-line fixes and users who want fast output without answering questions are also poor fits, because the skill pauses for approval by design.
If you want to explore options or brainstorm, the source points to the omc-plan skill instead.
- Detailed requests with file paths and acceptance criteria
- Quick single-line fixes
- Brainstorming, where omc-plan is recommended
- Work you want done without answering questions
What people use it for
- Turning a vague feature idea into a written spec before a complex build starts
- Catching hidden assumptions in a request that previously produced results you did not want
- Producing an approved spec that a team can review before autonomous execution begins
- Setting up an autoresearch mission and evaluator through the --autoresearch option
Frequently asked questions
How do I install the deep-interview skill?
Download the ZIP from this site's skill page and unzip it into ~/.claude/skills/ for personal use or into .claude/skills/ in a project. In the Claude apps, upload it under Settings → Capabilities → Skills.
Is the deep-interview skill free?
The source lists the licence as MIT. It does not state pricing, so check the upstream repository for licence terms and any costs of Claude Code itself.
What does deep-interview require?
It requires the oh-my-claudecode plugin, because it references the explore, plan and autoresearch agents, plus Claude Code or a compatible SKILL.md agent runtime. The source says it does not run scripts and does not use the network.
Does deep-interview write code?
No. According to the source, the output is a written spec file saved under .omc/specs, and the skill does not mutate files before execution is explicitly approved.
What is the difference between deep-interview and omc-plan?
Deep Interview is for clarifying a vague idea through questions and an ambiguity score. The source says that if you want to explore options or brainstorm, you should use the omc-plan skill instead.
Why does the skill keep asking questions instead of starting?
It is designed to refuse execution until ambiguity is below the threshold and you approve a scoped plan. If you want output immediately without answering questions, this skill is not the right choice.
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
Deep Interview suits developers starting a complex or fuzzy feature who want the agent to ask before assuming, and teams that want an approved written spec before autonomous work begins. Skip it for detailed requests with file paths and acceptance criteria, quick fixes, or anyone who wants fast output without answering questions.
Download the skill