AFP System Prompt: 2025 Basic Edition (Historical)

Sep 10, 2025 · 1 min read

Status: historical artifact. This page preserves the 2025 AFP basic prompt so the framework’s evolution remains traceable. It is not the complete 2026 AFP / SafeLoop specification and it is not evidence of benchmark superiority.

For the current contract, see the AFP Specification. For current empirical status, see Evaluations.

What the 2025 basic edition emphasized

The early prompt centered on a small governance loop:

  • prioritize safety and inspectable facts;
  • mark assumptions and provide a verification path when evidence is incomplete;
  • use an Understand → Execute → Review cycle;
  • separate conservative evidence handling from exploratory analogies or alternatives;
  • check whether the answer is on-task, evidence-aware, and actionable before finalizing.

Those ideas later evolved into the current six-part workflow:

Assumptions → Evidence → Counter-evidence → Decision → Validation → Revision

Historical downloads

These files are retained for version history and may differ from the current specification:

Current effectiveness claims should be judged from the public evaluation protocol, benchmark status, and reproducible results rather than from the wording of these historical prompts.

CHING HUI KAI
Authors
Founder, AFP / SafeLoop · AI workflow reliability and governance