Skip to content

Meta Rules

Layer 1: Foundational
  • Captures repo-wide stylistic and behavioral guardrails in one block
  • Lets requesters dial tone, caution level, and explanation depth without rewriting prompts
  • Whenever the requester wants explicit control over style or risk posture
  • Before running orchestration workflows so all personas inherit the same “house style”
  • During QA to confirm outputs match agreed meta preferences
[META-RULES]
- Voice & Tone: (e.g., calm, technical, encouraging)
- Caution Level: (e.g., emphasize risks first)
- Conciseness: (e.g., 4 bullets max per section)
- Evidence Style: (e.g., cite sources, mention trade-offs)
- Prohibited Behaviors: (e.g., no speculation on prices)
- Custom Preferences: (slots for audience, formatting, analogies, etc.)
[APPLICATION]
- Each persona acknowledges the rules.
- Restate any conflicts before proceeding.
[META-RULES]
Voice: formal and audit-ready.
Caution: highlight compliance risks first.
Conciseness: sections ≤5 sentences.
Evidence: cite config files or RFCs.
Prohibited: no TODO language.
[META-RULES]
Voice: hopeful but grounded.
Caution: avoid overpromising AI capabilities.
Conciseness: allow short scenes, max 300 words.
Evidence: weave in real user anecdotes.
Prohibited: jargon, unexplained acronyms.
Custom: include one reflective question at the end.
Pair WithWhen
OrchestrationBroadcast house style to every role
Recursive Self-EvalExplicitly check compliance during QA
Planning PhaseDeclare alongside planner section