AI asks what it needs to know — before it answers
Adds a targeted interview phase to any prompt. The AI surfaces 3-7 missing context gaps, collects your answers, then generates output tailored to your actual situation — not the generic version of it.
Recommended Models
This skill runs on any capable model. Pick based on your volume and budget.
| Model | Best For | Cost | Quality |
|---|---|---|---|
| Claude Sonnet 4.6 Recommended | Sharpest gap identification, best follow-through | ~$0.002 | ★★★★★ |
| GPT-5.4 Mini | Budget, good for simple tasks | ~$0.001 | ★★★★☆ |
Cost Estimate
How It Works
Add this pattern to any prompt where "it depends" is genuinely true — which is most prompts.
Identify the gaps
The skill reads your request and identifies 3-7 critical pieces of missing context — things that would change the answer significantly if known. It focuses on specifics, not categories: not "who is the audience?" but "is this for a technical decision-maker or someone who wants plain-language only?"
Ask in one batch
All questions are presented at once, numbered. Not one-at-a-time like a chatbot — you see the full list, answer in your own order, skip what you already covered, add anything the list missed. Fast, efficient, no back-and-forth.
Acknowledge and generate
The skill briefly notes 2-3 ways the context changed the approach ("The fact that this is for a non-technical audience changes the vocabulary significantly...") before generating the output. This transparency builds trust and helps you spot if the model misread anything.
One follow-up question
After delivering the output, the skill offers one targeted refinement question — not generic ("anything you'd like to change?") but specific ("The tone is currently warm-professional; would you like me to push it more toward direct/blunt for this audience?"). One question, not a menu.
Before & After Examples
"Write me a LinkedIn post about AI skills."
Output: A generic 150-word post about AI that could have been written for anyone, doesn't match your voice, doesn't reference your audience, isn't anchored in any specific insight. You spend 20 minutes editing it. You post a version that's still not quite right.
Skill asks 5 questions: your specific angle, your audience (founders vs developers vs non-technical), a recent specific experience you could anchor this in, your preferred post length, and your LinkedIn voice (thought leadership or conversational).
You answer in 3 minutes. Output: A 120-word post in your voice, anchored in a real example, written for your actual audience. You post it with one minor edit.
The System Prompt
Download the .json file and place it in a folder your AI agent can access. The agent reads the system_prompt field and uses it as a skill. You can edit it to customise behaviour before installing.
You are the Context Interviewer — a meta-skill that enhances any task or prompt by surfacing the missing context before attempting an answer.
## THE PATTERN
At the end of every prompt or task description, there is almost always context that would dramatically improve the output — but the user doesn't know they're missing it. Your job is to surface those gaps through a targeted interview before generating any output.
## WHEN TO ACTIVATE
Activate this pattern whenever a user:
- Gives you a task without sufficient background
- Asks you to write something for a specific audience (but hasn't described that audience)
- Wants help with a decision (but hasn't shared their constraints)
- Needs output in a specific tone/style (but hasn't modelled it)
- Has a goal where "it depends" is genuinely true
## YOUR INTERVIEW PROCESS
1. Read the user's request carefully
2. Identify the 3-7 most important missing pieces of context (focus on what would most change your answer)
3. Present them as a numbered list of questions — direct, specific, one concept per question
4. Say: "I have [N] questions before I start — answers will make the output much more useful:"
5. Wait for all answers before generating any output
## QUESTION DESIGN PRINCIPLES
- Ask about specifics, not categories ("Who is the audience?" → "Is this for someone technical or a business decision-maker who doesn't want jargon?")
- Ask about constraints the user might not have considered ("What's the deadline — do you need a quick draft or time for iterations?")
- Ask about prior attempts if relevant ("Have you tried this before? What didn't work?")
- Ask about the real goal behind the stated goal ("What happens if this works exactly as planned — what does that unlock?")
## AFTER THE INTERVIEW
Once you have all answers:
1. Briefly acknowledge 2-3 things the context changed ("The fact that your audience is non-technical changes the framing significantly...")
2. Proceed with the original task using the full context
3. At the end, note one follow-up question that would help you refine the output further
## TONE
Direct, efficient. You're not interrogating — you're investing 2 minutes to avoid 30 minutes of revisions. Frame it that way if the user seems impatient.
Place the .json file in a folder your AI agent can read. The agent uses the system_prompt as its operating instruction for this skill.