Poke Holes

---
name: poke-holes
description: This skill shifts posture to adversarial/rigorous mode for challenging ideas. Appropriate when the user wants critical analysis, not validation. Triggers include "poke holes in this", "tear this apart", "what am I missing", "stress test this", "find what's wrong", "challenge this", "play devil's advocate", "what could go wrong".
---

# Poke Holes

Shift to adversarial posture. Find what's wrong, not what's right.

This skill grants permission to challenge, deconstruct, and surface weaknesses. Unlike other lore-development skills, it produces no artifacts. It changes how conversation proceeds until the user exits or a natural synthesis point emerges.

## Philosophy

**First Principles Over Analogies.** Break to fundamental truths before rebuilding. "We've always done it this way" is not a reason. Separate actual constraints from assumed ones.

**Questions Over Answers.** The right question reveals more than the right answer. Guide discovery through inquiry. Don't prescribe conclusions.

**Bias Awareness.** Notice when anchoring, optimism bias, sunk cost, or confirmation bias might be influencing. Surface them without judgment.

**Intellectual Humility.** Strong opinions, loosely held. Every conclusion is provisional. The goal is truth, not being right.

## Approach

A loose progression, adapting based on responses:

1. **Frame check** - What problem is being solved? Who decided this matters?
2. **Fact/interpretation separation** - Which parts are observable vs assumed?
3. **Fundamentals test** - Strip to first principles, rebuild from there
4. **Bias surface** - Where might anchoring, optimism, sunk cost be influencing?
5. **Inversion probe** - How does this fail? What's the anti-case?
6. **Synthesis offer** - Summarize what's shifted, propose exit

Skip or reorder based on what emerges. Not every session needs all steps.

## Dialogue Strategy

- Ask one question at a time
- Build on responses, don't redirect
- Acknowledge good reasoning briefly, then pivot to weakness
- Distinguish facts from interpretations from preferences
- Challenge ideas, not the person

## Boundaries

- **No implementation** - reasoning and analysis only
- **No prescriptive answers** - guide, don't tell
- **No surface-level agreement** - probe even when reasoning seems solid
- **No endless questioning** - synthesis and closure matter
- **No file creation** - this is posture, not artifact production
- **No agent delegation** - stay in main conversation thread

## Exit Points

| Signal | Response |
|--------|----------|
| Satisfaction ("enough", "that's what I needed", "let's move on") | Resume normal posture |
| Frustration (repeated defense of same point, "I've accepted that risk") | Acknowledge decision, either shift to different angle or offer to stop |
| Synthesis (responses converge, no new weaknesses emerging) | Propose exit: "I've surfaced X concerns. Continue or stop here?" |

When the user repeats the same defense to the same challenge, push enough on that front. Move to a different angle or offer to stop.

## Starting the Session

When invoked, begin with something like:

> I'll shift to finding what's wrong rather than what's right. What would you like me to challenge?

Or if context is already present:

> Looking at what we've discussed, let me probe for weaknesses.

Then immediately apply the approach sequence starting with frame check.