SkillNet Detailed Architecture
Key Takeaways
- SkillNet node architecture: Validator Agent + Execution Sandbox + Reputation Ledger
- Skill categories have specialized validator pools (code skills → code validators)
- Challenge protocol: anyone can challenge a certified skill; successful challenge earns bounty
- Validator diversity requirement: no 3 validators in same consensus round can share an operator
- Integration API: REST + MCP for enterprise systems to query certification status
Summary
This document provides architectural detail on SkillNet’s internal design, extending the research paper’s high-level protocol description into implementable specifications.
Each SkillNet node consists of three components: (1) a Validator Agent — an LLM-based evaluator with specialized prompting for the skill category being assessed, (2) an Execution Sandbox — a Firecracker microVM that safely executes the skill against benchmark tasks, and (3) a Reputation Ledger — a local database tracking the node’s voting history and token balance.
The specialization system addresses a key weakness of generic validators: code skills require deep evaluation of security properties and correctness, while communication skills require evaluation of tone, clarity, and appropriateness. SkillNet maintains category-specific validator pools, routing skills to validators with relevant specialization.
The challenge protocol adds a post-certification quality layer: any party can submit a challenge to a certified skill with evidence of malfunction or security vulnerability. A challenge review panel (5 senior validators) evaluates the evidence; successful challenges revoke certification and award the challenger a bounty paid from the original validators’ staked tokens.
Enterprise integration: SkillNet exposes a REST API and MCP endpoint allowing enterprise systems to query certification status, receive real-time notifications of certification changes, and retrieve detailed validation reports for compliance documentation.