32 shortcut hacks for faster Claude prompts! These commands work without any custom definitions because Claude, ChatGPT, and Gemini naturally recognize them. Why? They’ve appeared hundreds of millions of times in training data. Thanks to instruction tuning and structured prompt design, these patterns just work. — @lucas_flatwhite (original: @rubenhassid)

Writing longer prompts does not guarantee better results. A single well-designed command often outperforms several paragraphs of instructions. These shortcuts work immediately because LLMs have already internalized these patterns through instruction tuning – no prompt engineering required.

Output Control Commands

The most frequently used group in practice controls output format. /ELI5 simplifies complex concepts for non-experts. /TLDR condenses lengthy reports. /EXEC SUMMARY produces executive-level summaries. /CHECKLIST and /STEP-BY-STEP transform the same information into actionable formats. /FORMAT AS forces specific output structures like tables, JSON, or markdown.

The key insight is that changing the output format alone completely transforms the utility of a response. Summarize meeting notes with /TLDR, then convert with /CHECKLIST, and you have an immediately assignable task list.

Metacognitive Commands for Deeper Thinking

The most interesting group controls the AI’s reasoning process itself. /CHAIN OF THOUGHT reveals intermediate reasoning steps. /FIRST PRINCIPLES strips away assumptions and rebuilds from fundamentals. /DELIBERATE THINKING suppresses hasty answers. /NO AUTOPILOT prevents generic pattern repetition.

/REFLECTIVE MODE and /EVAL-SELF make the model critique its own output. Add /SYSTEMATIC BIAS CHECK and you get bias detection on top. To use AI as a genuine thinking partner rather than a simple Q&A tool, combine these commands deliberately.

Role and Perspective Switching

/ACT AS is the most widely known role assignment pattern. /DEV MODE and /PM MODE instantly shift to profession-specific viewpoints. /MULTI-PERSPECTIVE and /PARALLEL LENSES illuminate a single issue from multiple angles simultaneously. /SWOT and /COMPARE apply decision-making frameworks directly.

Combining /AUDIENCE with /TONE lets you produce both a beginner-friendly explanation and a technical specification from the same input. /GUARDRAIL explicitly constrains response boundaries, preventing the model from drifting off topic.

No need to memorize these upfront – try them once in context and they become second nature. This is the most practical approach to cutting prompt authoring time while raising output quality.

Key takeaways

  • Slash command patterns already learned through LLM instruction tuning work instantly without custom definitions
  • Metacognitive commands like /CHAIN OF THOUGHT, /FIRST PRINCIPLES, and /NO AUTOPILOT let you control the quality of AI reasoning itself
  • Chaining output format switches (/TLDR → /CHECKLIST) with role switches (/DEV MODE, /PM MODE) generates diverse deliverables from a single input

Source