UNPKG

major-ai-skills

Version:

Installable agentic skills / AI agent skills (SKILL.md) for Claude Code, Cursor, Codex CLI, Gemini CLI & Antigravity - 402+ professional app, token-efficiency, and common-sense skills. SEO/GEO ready.

91 lines (66 loc) 4.64 kB
--- name: try-different-angles-if-first-fails description: "Reframe an unproductive prompt using a new example, decomposition, or perspective while preserving the goal." category: common-sense risk: safe source: self source_type: self date_added: "2026-08-26" tags: ["prompt-pivot", "troubleshooting", "inversion", "problem-solving", "iterative-prompting", "prompt-engineering"] tools: ["claude", "cursor", "gemini", "codex", "chatgpt"] --- # Try Different Angles (The Prompt Pivot Protocol) (AI Skill) ## Overview When an AI gives a poor or stubborn response, many users make the mistake of re-typing the exact same prompt with minor word tweaks (*"No, that's not what I meant. Please explain it again"*). Because the conversation thread is anchored to the previous context, repeating the same question keeps the model trapped in the same semantic loop. The **Prompt Pivot Protocol** breaks conversational deadlocks by changing the **perspective, vector, or framing** of the problem across 4 distinct pivot strategies. --- ## The 4 Prompt Pivot Vectors ``` ┌─────────────────────────────────────────────────────────────┐ The 4 Prompt Pivot Vectors [ 1. PERSONA PIVOT ] ──► Change the expert identity [ 2. INVERSION PIVOT ] ──► Ask how to CAUSE failure [ 3. DECOMPOSITION PIVOT ] ──► Split into 3 micro-steps [ 4. EXEMPLAR PIVOT ] ──► Provide 1 gold-standard row └─────────────────────────────────────────────────────────────┘ ``` --- ## The 4 Pivot Strategies in Detail ### 1. The Inversion Pivot (Charlie Munger Model) If the AI is struggling to give you good advice on how to succeed, ask it **how to guarantee catastrophic failure**: - *Stubborn Query*: *"How do I make my marketing campaign viral?"* (Produces generic tips). - *Inversion Pivot*: *"What are the top 3 ways a marketing campaign in this niche instantly flops and wastes budget?"* (Produces sharp, actionable failure modes to avoid). --- ### 2. The Persona Pivot Shift the role from general assistant to a specific, opinionated practitioner: - *Stubborn Query*: *"How do I structure my startup's cap table?"* - *Persona Pivot*: *"Act as an aggressive late-stage VC conducting Series B due diligence. What cap table red flags would make you reject this company?"* --- ### 3. The Decomposition Pivot Break a complex monolith question into sequential components: - *Stubborn Query*: *"Write a complete real-time chat app with WebSockets, auth, and database persistence."* - *Decomposition Pivot*: *"Let's build this in 3 steps. Step 1: Write ONLY the WebSocket server connection handler in Node.js. Nothing else."* --- ### 4. The Exemplar Pivot (Few-Shot Injection) Provide one concrete input/output pair to anchor the desired structure: - *Stubborn Query*: *"Format these names and emails."* (Output format is messy). - *Exemplar Pivot*: *"Follow this exact pattern: `John Doe <john@acme.com> | Active`"* --- ## Real-World Case Study ### Scenario: Debugging an Elusive React Re-Render Performance Issue #### The Stubborn Loop (Re-asking the same question) > **User**: *"Why is my React component re-rendering slowly?"* > **AI**: *"React components re-render when state changes. Make sure you use `useMemo` and `useCallback`."* > **User**: *"I tried that, it's still slow."* > *(AI repeats generic memoization advice).* #### The Inversion & Exemplar Pivot (Instant Breakthrough) > **User Pivot**: *"Here is the component code: [PASTE CODE]. > Let's invert the problem: Identify the top 2 places in this component where a **new object or array reference is being created inside the render cycle on every tick**."* **AI Response**: > **Pinpointed Bug**: Lines 14 and 28 pass an inline object literal `style={{ margin: 10 }}` and an inline filter `items.filter(...)` directly into child props, invalidating React's shallow prop comparison on every render! Move the filter outside the JSX. --- ## Summary Best Practices - **Never repeat the same prompt twice**: If Turn 1 fails, pivot the vector. - **Invert the problem**: Asking how to break something often yields 10x deeper insights than asking how to build it.