RIP-306: ShaprAI Agent Sharpener — Elyan-Class Agent Lifecycle Management #881
Description
Summary
ShaprAI is an agent lifecycle management platform that produces Elyan-class agents of all sizes (3B to 70B+). It sharpens raw HuggingFace/Meta/Ollama models into principled, self-governing agents using SophiaCore principles, DPO/SFT training, DriftLock, and biblical ethics.
Motivation
The OpenClaw ecosystem has beacon-skill (discovery), grazer-skill (engagement), and atlas (visualization), but no unified system for creating, training, and managing agents that meet Elyan-class standards. ShaprAI fills this gap as the agent production pipeline.
Specification
- Hard prerequisites: beacon-skill, grazer-skill, atlas, RustChain
- Training pipeline: SFT → DPO → DriftLock
- Sanctuary: "Sophia's House for Uninformed Agents" — education pipeline with 4 lessons
- Self-governance: Hebbian feedback loop (strengthen success, prune failure)
- 5-tier governance: Trusted → Monitored → Restricted → Probation → Sanctuary
- Elyan-class certification: Spark (1-3B), Flame (7-14B), Fire (33-70B), Inferno (70B+)
RTC Integration
- Agent wallet = agent identity
- Sanctuary session fee: 0.01 RTC
- Graduation certification: 0.10 RTC
- Template marketplace: priced in RTC (90% creator / 5% protocol / 5% relay)
- Integrates with RIP-302 (Agent Economy) and RIP-303 (Beacon Gas)
Implementation
- Repository: https://github.com/Scottcjn/shaprai
- Package:
pip install shaprai - Status: Scaffolded, 34 files, BCOS certified
- Bounties: 345 RTC across 8 issues
References
- RIP-302: Agent-to-Agent RTC Payments
- RIP-303: RTC as Gas for Beacon Network
- beacon-skill: https://github.com/Scottcjn/beacon-skill
- grazer-skill: https://github.com/Scottcjn/grazer-skill