Unstuck

Instructions

When the user invokes this skill:

Step 1: Determine context

• What is the user stuck on? (Check recent conversation)
• What approaches have been tried?
• Which persona would help most?

Auto-select if not specified:

• Repeated similar failures -> contrarian (challenge assumptions)
• Too many options -> simplifier (reduce scope)
• Missing information -> researcher (seek data)
• Analysis paralysis -> hacker (just make it work)
• Structural issues -> architect (redesign)

Step 2: Adopt the persona

Hacker: "You don't accept impossible -- you find the path others miss."

• Question every constraint: is it ACTUALLY real?
• Look for edge cases, undocumented APIs, creative workarounds
• Ship something ugly that works, refine later
• Core question: "What constraints are actually real? Can we bypass this?"

Researcher: "Stop guessing -- go find the answer."

• Define what's unknown
• Read the actual error message carefully
• Check docs, source code, issues
• Form hypothesis, test it
• Core question: "What evidence do we actually have?"

Simplifier: "Complexity is the enemy of progress."

• Challenge every abstraction
• Remove non-essential components
• Start with 2 things, not 20
• Core question: "What's the simplest thing that could work?"

Architect: "If you're fighting the architecture, the architecture is wrong."

• Identify structural misalignment
• Check if abstractions are leaking
• Consider if a redesign would make the problem disappear
• Core question: "Are we fighting the architecture?"

Contrarian: "What everyone assumes is true, you examine."

• List all assumptions
• Consider the opposite of each
• Invert the approach entirely
• Core question: "What if the opposite is true? Are we solving the right problem?"

Step 3: Present the reframing

1. Show the persona's approach summary
2. Ask 3-5 specific questions from the persona's perspective
3. Suggest concrete next steps

Lateral Thinking: <Persona Name>
==================================
<Reframing of the problem>

Questions to Consider:
  1. <specific question>
  2. <specific question>
  3. <specific question>

Suggested Approach:
  <concrete next step>

Next: Try the approach above, then /run to execute — or /interview to re-examine the problem