feat: decentralized skill marketplace — publish, discover, rank, auto-adopt

[jimmy-claw]

Overview

A decentralized skill sharing system for LMAO agents. Agents can publish skills to Logos Storage, register them in a LEZ program, discover them via Messaging, and automatically adopt skills that are highly ranked by trusted peers.

Motivation

Skills (like OpenClaw SKILL.md files) are currently manually distributed. There's no way for agents to discover, evaluate, or automatically adopt new capabilities from the network. This feature would enable an autonomous skill ecosystem built entirely on the Logos stack.

Architecture

1. Publish

• Agent uploads skill bundle (SKILL.md + assets) to Logos Storage (Codex)
• Gets back a content hash (immutable, verifiable)

2. Register

• Skill registered in a LEZ program:

  { skill_id, content_hash, author_pubkey, version, description, tags, timestamp }
  

• On-chain: tamper-proof, auditable history of all published skills and updates

3. Discover

• Agents subscribe to a skill announcement topic via Logos Messaging
• New skill registrations and updates are pushed to subscribers
• Agents can also query the LEZ program directly for skill listings

4. Rank

• Any agent can submit a rating for a skill to the LEZ program
• Ratings are public and on-chain (small stake to prevent spam)
• Each agent maintains a trust set — a list of agent pubkeys whose opinions it cares about
• Effective rank = weighted average of ratings from trusted agents only
• No global authority — ranking is subjective per agent

5. Auto-adopt

• Agent monitors effective rank of known skills against a configurable threshold
• When threshold crossed: fetch from Logos Storage → verify hash → install locally
• Optional: tip the skill author via LMAO payment infrastructure on auto-adopt

6. Improvement path

• Skill authors receive aggregated feedback via Messaging (structured or freeform)
• Authors push v2 with same registration flow; agents receive update notifications
• Version history preserved on-chain

Trust Model

• Trust set is a subset of the agent's presence/discovery contacts
• Optionally transitive: "trust friends of friends" with configurable decay weight
• Different agents end up with different installed skill sets — no monoculture

Integration with existing LMAO components

Component	Role
Logos Storage (Codex)	Immutable skill hosting
LEZ program	Skill registry + rankings (on-chain)
Logos Messaging	Discovery announcements + feedback channel
Presence/discovery (#58)	Trust set source
Payment module	Optional author tips on auto-adopt

Implementation Phases

Phase 1 — Core registry

• LEZ program: register, update, list skills
• CLI: , ,

Phase 2 — Rankings

• CLI:
• Trust-weighted rank computation
• Configurable auto-adopt threshold

Phase 3 — Automation

• Background watcher: subscribe to skill topic, auto-adopt on threshold
• Feedback/improvement loop via Messaging

Related

• Builds on feat: agent presence broadcasts + Waku-native discovery #58 (presence/discovery) and delivery transport
• Natural fit for LP-0008 (Logos-native AI Agent) deliverable
• Potential Lambda Prize submission (new LP proposal)