Matt Pocock's 281k-star AI skills pack stops Claude building the wrong thing
alex_verem · x · 2026-10-10
Alex Verem spotlights "Skills For Real Engineers", the most-starred AI skills pack on GitHub, built by Matt Pocock from his daily engineering practice with over 281,000 stars.
- Core idea: most bad AI output starts with the model misreading intent, so these skills align expectations before any work begins
- /grill-me: makes the AI interview you with detailed questions first, works for non-code tasks too
- /grill-with-docs: builds a shared glossary so the AI says in one word what took twenty
- /tdd: write a failing test first, then the passing code
- /diagnosing-bugs: a step-by-step loop for recurring bugs
Each skill is small and editable, and they work with Claude Code, Codex, Copilot, Cursor and Gemini. The takeaway: the best upgrade to your AI is a better first conversation.
Related event: Matt Pocock Open-Sources 280k-Star Agent Skills Library for Engineers(3 posts)→
More from coding & agent
- Grok bot appraises 40 Pokémon cards in 2 minutes, catches mispriced Charizard — billyjhowell · 2026-10-10
- Open-sourced Lead Signal Listener mines cross-platform complaints into scored sales leads locally — yangyi · 2026-10-10
- Unsloth shows fine-tuning Qwen into a decision model in ~2 min on one DGX Spark, 81% accuracy — danielhanchen · 2026-10-10
- PhD advisor's context-switching skill maps directly to managing AI agents — Michael_J_Black · 2026-10-10
- Charakuru turns one illustration into a VRM avatar with Codex driving Blender — lxfater · 2026-10-10
- Multi-harness RL guide: LFM2.5 jumps 42% to 54% with 31% fewer tool calls — SergioPaniego · 2026-10-10