Skip to content

Chengyue5211/GAFP

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

GAFP — Growth Agent Feeding and Personality Governance Protocol

成长型智能体投喂与人格治理协议

"不是所有输入都能成为人格。"

GAFP is the world's first protocol standard defining how input data is admitted into, shapes, freezes, revises, and is inherited by a personal AI agent's personality structure.

Why GAFP?

Current AI agents fall into two paradigms:

Static (Configuration) Growth (GAFP)
Personality One-time prompt/knowledge base Grows through structured feeding
Memory RAG vector search Multi-layer memory + knowledge graph
Data Sovereignty Platform-owned User-owned, portable
Write Governance No gate Five-level Admission Gate (G0-G4)

Most AI agents today are configured, not grown. Your digital traces are collected by platforms — but collected is not nurtured. GAFP bridges this gap.

Core Concepts

Five Dimensions (W1-W5)

Code Name Core Question
W1 Judgement How do I make decisions?
W2 Cognition How do I understand things?
W3 Expression How do I express myself?
W4 Relation How do I relate to others?
W5 Sovereignty How do I govern myself?

Five Depth Layers (D1-D5)

Layer Name Core Question
D1 Identity Core Who am I?
D2 Cognitive How do I understand the world?
D3 Value Ordering What do I prioritize?
D4 Agency What can I do?
D5 Narrative Revision Can I change? Can I redefine myself?

Admission Gate (G0-G4)

Not all input deserves to become personality. GAFP defines five levels of write permission:

  • G0 Archive Only — Stored, not used for personality modeling
  • G1 Retrievable — Searchable, does not update skeleton
  • G2 Candidate — Enters analysis pipeline, needs multi-source verification
  • G3 Skeleton Update — Updates personality skeleton
  • G4 Anchored Snapshot — Forms immutable version snapshots

Key rule: AI-inferred data (source=ai) is capped at G2. AI must never auto-write to G3/G4.

Observation Grammar

Structured five-part observation records: Event / Context / Observable Behavior / Self-Report / Observer Inference. Observer Inference weight is capped at 0.5 to prevent observer bias from contaminating the agent's personality.

PCI (Personality Completeness Index)

PCI = Σ(Wi × Di × Ci) / ΣWi

A reproducible formula where all variables have explicit definitions.

Personalized Holistic Development (v3.1)

Full development ≠ standardized development. Two digital twins with identical dimension scores are still fundamentally different entities. Growth must be personalized — increasingly faithful to the real person, not converging toward an optimization target.

Dialectical Growth (v3.1)

The real self and digital twin form a bidirectional feedback loop: human growth feeds richer data to the twin; the twin reflects back a structured "personality mirror" that motivates further human development. This Resonance Effect means both develop faster together than either would alone.

DLI-C (Digital Life Integrity Checklist)

Nine verifiable criteria for determining whether a digital existence qualifies as "digital life" under GAFP standards.

Repository Structure

whitepaper/     — White paper (Chinese, v3.0 + v3.1)
schema/         — JSON Schema for Feed Unit (v0.1)
examples/       — Sample feeding data
reference/      — Reference implementation snippets
LICENSE-DOC     — CC BY-SA 4.0 (documentation)
LICENSE-CODE    — Apache 2.0 (code)

Licensing

  • White paper & protocol specification: CC BY-SA 4.0 — Use freely, but attribute and share alike
  • Reference code: Apache 2.0 — Commercial use allowed, patent rights reserved
  • Yuechuang Nianlun (full system): Proprietary — the first reference implementation

Reference Implementation

Yuechuang Nianlun (跃创年轮) is the first and reference implementation of GAFP, covering three life stages:

  • ChenLu (晨露) — Children: Guardian-supervised feeding, sovereignty transfer at age 18
  • DangRan (当燃) — Adults: Self-directed feeding, DID identity
  • WanXia (晚霞) — Elders: Legacy planning, family digital heritage

Citation

If you use GAFP in your research or product, please cite:

@misc{gafp2026,
  title={GAFP: Growth Agent Feeding and Personality Governance Protocol},
  author={Dr. Cheng Yue and Yuechuang Nianlun},
  year={2026},
  note={White Paper v3.1},
  url={https://github.com/Chengyue5211/GAFP}
}

Author

Dr. Cheng Yue (程跃博士) · Yuechuang Nianlun (跃创年轮)

© 2026 Dr. Cheng Yue · Yuechuang Nianlun. All rights reserved.

About

GAFP — Growth Agent Feeding and Personality Governance Protocol. The world's first protocol defining how input shapes AI agent personality.

Resources

License

Unknown, Unknown licenses found

Licenses found

Unknown
LICENSE-CODE
Unknown
LICENSE-DOC

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors