PKM Wiki

AI Master Prompt

2 min read

An AI Master Prompt is a carefully structured document that provides relevant context and rules before you start interacting with AI. It is the practical entry point for Context Engineering — the place where your PKM system meets your AI assistant.

What a Master Prompt Contains

A well-structured master prompt includes:

  • General rules — Behavior guidelines, communication style, constraints
  • Personal information — Identity context (values, beliefs, goals, expertise)
  • Work/business context — Projects, products, team, domain knowledge
  • Roles — How the AI should behave in different contexts
  • Tasks — Specific task definitions with description, goals, context, rules, inputs, and outputs
  • Configuration — Tool access, permissions, formatting preferences

Two Structural Approaches

Single Long Document

One file containing everything. Easy to start with but does not scale:

  • Hard to maintain (changes require finding the right section)
  • Forces duplication (same context repeated for different tasks)
  • Hits context limits fast (a 20-page master prompt wastes tokens on irrelevant sections)
  • Cannot be shared or composed

Multiple documents in a linked structure. The master prompt becomes an entrypoint that references other files:

  • No duplication (shared rules in one place, referenced by many tasks)
  • Easier maintenance (change one file, all references update)
  • Scalable via lazy loading (AI fetches what it needs from the graph)
  • Composable (team members inherit shared context, layer personal context)

The entrypoint must include: knowledge base overview, architecture explanation, and instructions for navigating and loading context. This is Context-as-Code in practice — CLAUDE.md → AGENTS.md → skills → agent configs → memory.

The PKM Connection

The master prompt is where your PKM system becomes AI-accessible. The quality of AI output is directly proportional to the quality of context provided, which depends on the quality of the underlying knowledge base.

A master prompt grounded in a well-maintained vault (with identity notes, organized knowledge, consistent metadata) produces dramatically better results than one written from memory. Your Single Source of Truth feeds your master prompt; your master prompt feeds your AI.

This is also implicit intent engineering: a master prompt already encodes roles, tasks, and success criteria — just not yet formalized as a discipline.

Anti-Pattern

The knowledge dump. Sharing a huge knowledge base and hoping AI magically understands it all. Without structure, priority signals, and navigation instructions, more context means more noise. The AI needs a map, not just a territory.

Key Points

  • The master prompt is the practical entry point for context engineering
  • Knowledge graph structure (multiple linked files) beats single long document
  • The PKM vault feeds the master prompt; master prompt feeds the AI
  • Entrypoint + lazy loading + composability = scalable context
  • Anti-pattern: dumping everything without structure

Open Questions

  • How often should a master prompt be reviewed and updated?
  • Can AI agents maintain their own master prompts through accumulated memory?
  • What is the minimum viable master prompt for useful results?

References

  • Vault: AI Master Prompt, How to structure your AI Master Prompt