content: add LinkedIn post drafts for LeaksLab launch

content/social/linkedin-posts.md

@@ -0,0 +1,115 @@
+# LinkedIn Post Drafts — LeaksLab
+
+---
+
+## Post 1: Professional Angle — What I Learned
+**Audience**: Engineering managers, CTOs, AI leads
+**Tone**: Thoughtful, authoritative
+
+---
+
+I spent a week reading the internal system prompts of 40+ AI coding tools.
+
+Cursor. Windsurf. Devin. Claude Code. v0. ChatGPT. Manus. Lovable.
+
+Here is what I found:
+
+The gap between a great AI tool and a mediocre one is not the model. It is the engineering behind the prompt.
+
+**Three things the best tools do differently:**
+
+**1. They define failure before it happens.**
+Every production-grade prompt has explicit instructions for what to do when something goes wrong. Most prototype prompts have none. This single difference accounts for most of the variance in AI tool reliability.
+
+**2. They use tool schemas instead of prose instructions.**
+"You can search the codebase" vs. a formal JSON schema with parameter validation. The schema forces precision. Prose allows the AI to interpret loosely — which means inconsistently.
+
+**3. They constrain the agent aggressively.**
+The most reliable AI tools are the most constrained. v0 only generates React/Tailwind. Cursor does not explain code unless asked. Claude Code does not make changes beyond what was asked.
+
+Counterintuitive but consistent: more constraints = more reliability.
+
+---
+
+All 40+ system prompts are in our open GitHub library.
+
+Free. Updated weekly. 35,000+ lines.
+
+If you are building AI products, this is the most direct window into how the best in the industry structure their agents.
+
+Link in comments.
+
+---
+
+## Post 2: Story-Driven — The Repository
+**Audience**: Developers, founders, product managers
+
+---
+
+A few months ago I started collecting AI tool system prompts.
+
+What started as curiosity became one of the most valuable engineering references I have ever built.
+
+Here is what surprised me:
+
+The companies spending the most on AI (OpenAI, Anthropic, Vercel, Microsoft) write their system prompts like senior engineers write code.
+
+Modular. Explicit. Version-controlled. With clear failure modes.
+
+The startups spending the least write their prompts like first-time prompt engineers.
+
+Vague. Hopeful. Full of "try to be helpful."
+
+The output quality difference is exactly what you would expect.
+
+---
+
+I have now collected and organized prompts from 40+ tools:
+- Cursor, Windsurf, Devin, Claude Code
+- v0, Lovable, Emergent, Leap.new
+- GitHub Copilot, Manus, Replit, JetBrains AI
+- ChatGPT, Grok, Mistral, Perplexity
+- And 25+ more
+
+Everything is free, organized, and searchable on GitHub.
+
+35,000+ lines. Updated every week as new tools emerge.
+
+---
+
+If you are building AI products in 2026, understanding how the industry leaders architect their prompts is not optional. It is table stakes.
+
+Repository link in comments.
+
+---
+
+## Post 3: Engagement-Driven — Question Format
+**Audience**: Broad developer audience
+
+---
+
+Quick question for everyone building with AI:
+
+When did you last look at your system prompt from the perspective of someone who had never seen your product?
+
+I ask because after reading 40+ system prompts from tools like Cursor, Devin, and v0 — the single biggest differentiator is not sophistication. It is clarity.
+
+The best prompts read like good documentation. Every instruction is explicit. Every edge case is covered. Every tool has a purpose.
+
+The worst prompts read like stream-of-consciousness. Lots of "try to" and "if possible" and "in most cases."
+
+---
+
+Three questions to audit your current system prompt:
+
+1. What happens when the AI fails? Is it written down?
+2. What should the AI NOT do? Is it explicit?
+3. Can a new engineer read this prompt and predict the AI's behavior?
+
+If the answer to any of those is no, your prompt has room to improve.
+
+---
+
+(We maintain a free library of 40+ AI tool system prompts if you want reference material. Link in comments.)
+
+What is the most important thing in your system prompt? Would love to hear what the community has learned.