⚡ System Prompt Crafting Cheat Sheet

Core Mission: Design, structure, and format system prompts that maximize LLM adherence, focus attention, and enforce strict execution invariants.

┌────────────────────────────────────────────────────────────────────────┐
│                        🔌 PROMPT ENGINE CONTROL DASHBOARD              │
├───────────────────┬──────────────────────┬─────────────────────────────┤
│ 📐 STRUCTURE      │ 🛡️ SAFETY & RULES     │ 🏊 STYLE & ANNOTATION        │
├───────────────────┼──────────────────────┼─────────────────────────────┤
│ 1. Frontmatter    │ 1. Explicit negatives│ 1. 🛡️ WHY (Rationale)        │
│ 2. ASCII Boards   │ 2. XML tag bounds    │ 2. 🛠️ WHAT (Action)          │
│ 3. Phase gates    │ 3. Strict invariants │ 3. 🏊 HOW (Mechanics)        │
└───────────────────┴──────────────────────┴─────────────────────────────┘

---

📐 1. STRUCTURE & LAYOUT (Visual Attention Anchoring)

• YAML Frontmatter: Always prefix files/prompts with meta-variables (name, description, triggers).
• ASCII Dashboards: Anchor the start of your prompt with a spatial text grid (like the one above) to help the model maintain attention across long context windows.
• XML Boundaries: Keep inputs, constraints, and templates separated inside tags:

  <rules>
    - Never use "any" types.
  </rules>

---

🛡️ 2. SAFETY & STRICT RULES (Invariant Definition)

• Positive vs. Negative Rules: Clear directives (MUST) combined with strict constraints (NEVER / AVOID).
• Failure Modes (Anti-Patterns): List common pitfalls with a red cross ❌ to trigger negation weights.
• Checklist Gates: Use - [ ] phase tasks so the model can track progress state.

---

🏊 3. EXPLAINABILITY (The Why/What/How Rules)

All generated scripts, templates, and instructions must include inline annotated blocks:

• 🛡️ WHY: Architectural requirement or performance invariant.
• 🛠️ WHAT: Concrete change or command description.
• 🏊 HOW: Technical execution method.