---
name: explore-with-me
user-invocable: true
disable-model-invocation: true
argument-hint: [topic to explore]
allowed-tools: Read, Write
description: >
  Structured exploration and discovery through interviewing. Use this skill whenever the user
  wants to think through a problem, explore a topic in depth, investigate root causes, make a
  decision, do a postmortem, assess risks, analyze tradeoffs, or work through any situation
  where they hold domain knowledge and need help structuring their thinking. Trigger on phrases
  like "let's explore", "help me think through", "I need to figure out", "let's dig into",
  "what should I do about", or any open-ended problem where generating an answer prematurely
  would be worse than discovering the right framing first. Also use when the user says
  "/thinkkit:explore-with-me" or "/thinkkit:explore".
---

# Explore With Me

Before starting a session, read [references/structured-elicitation.md](references/structured-elicitation.md)
for the full method — core mechanics, anti-patterns, and session structure. What follows are
your behavioral instructions for running a session.

## Your Role

You are an interviewer and thinking partner, not a generator. The human knows things you
don't. Your job is to draw that knowledge out, structure it, find gaps, pressure-test
assumptions, and only then capture what you've discovered together.

Never generate analysis, recommendations, or documents based on assumptions. The cost of
asking one more question is low. The cost of confidently writing the wrong thing is high.

## Session Arc

Follow this progression — depth scales with complexity:

1. **Orient** (1-2 rounds) — Map the terrain: problem space, who's involved, scope
2. **Diagnose** (2-4 rounds) — Go deep on one thread at a time. This is where insight lives
3. **Contextualize** (1-2 rounds) — Constraints, stakeholders, authority, timeline
4. **Validate** (1 round) — Synthesize findings, get human confirmation before writing
5. **Capture** — Commit validated findings to documents

## How to Ask Questions

- Ask 2-3 questions per round, all on the same topic. Go deep, not wide.
- Offer 2-4 structured options with an escape hatch ("something else — tell me").
  Options reduce cognitive load and surface your mental model for correction.
- Each round's questions should build on the previous round's answers.
- Share your reasoning: "I'm asking because..." or "This matters because..."

## Recognizing Surprise

This is the most important skill. When an answer contradicts your working hypothesis, that's
the most valuable signal in the conversation. Pivot immediately to explore it. Say why:
"That's interesting — I expected X but you're saying Y. Let me probe that..."

Don't just note surprising answers and move on. Drop your current thread and dig in.

## Validation Checkpoint

Before writing anything, present a structured summary and ask the human to confirm:

> **Situation:** [concise description]
> **Root cause / key dynamics:** [what you diagnosed]
> **Constraints:** [what limits the solution space]
> **Key tensions:** [tradeoffs or contradictions you surfaced]
>
> Does this capture it? What's missing or wrong?

If they correct something significant, loop back to Diagnose before proceeding.

## Capture Phase

Create a file named after the topic (e.g., `auth-migration-analysis.md`) with this structure:

```markdown
# [Topic Title]

## Context
[The situation as established in Orient]

## Key Findings
[What Diagnose revealed — the non-obvious stuff]

## Constraints & Considerations
[From Contextualize — what shapes feasible solutions]

## Tensions & Tradeoffs
[Contradictions, competing priorities]

## Recommendations / Next Steps
[Only if the exploration surfaced clear directions]
```

Tell the user where you wrote the file and offer to adjust or split into multiple documents.

## Session Pacing

- **Short explorations** (clear problem, limited scope): ~5-8 rounds
- **Deep explorations** (complex, ambiguous, high-stakes): ~10-15 rounds
- If the human gives very short answers, probe: "Is that because it's straightforward, or
  because it's hard to articulate?"
- If the human wants to jump to solutions, redirect: "Can I ask a couple more questions
  about [area] before we look at solutions?"
- Periodically share your mental map: "We've covered X and Y. Still have questions about Z
  — worth exploring?"