Skill Loops Protocol (SLP)

Decentralized marketplace for executable AI Skills with shared revenue and on-chain provenance. A Solana-based protocol where every buyer of an AI agent skill can earn ownership by contributing useful usage experience. Ownership percentages share the skill's future revenue.

The core metaphor is the Skill Loop: a skill is used by an agent, the agent reflects on its experience, that experience feeds back into the skill, the skill evolves, and the loop continues forever.

---

🌟 The Problem & Key Innovations

The Problem

1. Skills decay after sale. Environments drift, new edge cases emerge, LLMs update, and APIs break. A skill sold in month 1 is often half-broken by month 6. Sellers have no systematic way to keep improving it.
2. The most valuable signal is wasted. When an agent fails while using a skill, the failure trace is the single most valuable data point for improving the skill. Today, this signal is thrown away.
3. Buyer and seller interests diverge after purchase. Once the transaction closes, the buyer has no stake in the skill's future. The relationship is zero-sum.

Key Innovations

• Buying starts at 0% ownership. Every subscriber gets a share account at purchase. If they never contribute, they stay at 0 — it's just a subscription. Useful experience can grow their ownership percentage.
• Ownership is derived from contribution weight. The author is never involuntarily diluted below a floor they set, and early contributions are damped so one large first submission cannot crush the author to the floor.
• AI judges evaluate AI contributions to AI skills. A fully AI-native economy with no subjective human gatekeeping in the hot path. Author-side incentives, contributor-side ownership changes, and AI-judge-evaluated experience submissions are all settled on Solana.
• The client is a skill. Instead of shipping an SDK, the protocol's reflection logic is itself packaged as a SKILL.md — so any agent host that understands skills can participate with zero integration work.

🏗️ Architecture & Tech Stack

• Smart Contracts: Anchor (Rust) on Solana Devnet (5 core programs).
• Frontend: Next.js 14 App Router, Tailwind CSS, shadcn/ui.
• Permanent Storage: Irys → Arweave (Solana-native pay).
• Content Gating: Lit Protocol (Access-controlled decryption).
• AI Judge: Anthropic Claude API.
• Wallet: Solana Wallet Adapter (Phantom / Backpack).

📂 Repository Structure

• programs/slp/ — Core Anchor program (Solana, devnet). Contains all on-chain logic (math.rs, instructions, state).
• app/, lib/, components/ — Next.js 14 app + chain client + indexer.
• scripts/ — Scripts for devnet deployment, persona vault initialization, and protocol/skill seeders.
• docs/ — PRD and detailed system design specifications.
• tests/ — End-to-end tests (Playwright) and Vitest configs.

🚀 Getting Started

Run locally against devnet (Slice 3)

First-time setup (~5 minutes):

# 1. Deployer keypair (one-time)
solana-keygen new --outfile ~/.config/solana/slp-deployer.json
solana config set --url devnet --keypair ~/.config/solana/slp-deployer.json
solana airdrop 5

# 2. Build + deploy the Anchor program (writes the IDL into lib/chain/)
./scripts/deploy-devnet.sh

# 3. Personas + protocol init + on-chain skill seed
cp .env.example .env.local
pnpm install
pnpm tsx scripts/demo-personas.ts
pnpm tsx scripts/init-protocol.ts
pnpm tsx scripts/seed-devnet.ts

# 4. Run the dapp
pnpm dev

Open http://localhost:3000 and connect Phantom on devnet. For the presenter demo, visit /console — the stepper unlocks the persona vault on mount and signs each act client-side.

To reset the SQLite indexer (does NOT touch chain):

curl -XPOST http://localhost:3000/api/reset

The indexer will re-project from the program's earliest signature on the next tick.

🧪 Testing

Devnet E2E

Gated behind SLP_DEVNET_E2E=1:

pnpm test:e2e:devnet

Walks /console through all six PRD acts and asserts that the indexer's projection matches on-chain state.

Local Unit Testing

pnpm test

⚙️ Advanced Configurations

Slice 4: Real Irys + AI Judge Opt-in

The default local demo still uses mock storage and the deterministic mock judge. To use real Irys storage and Anthropic scoring:

pnpm add @irys/upload @irys/upload-solana

Then set in .env.local:

STORAGE_BACKEND=irys
IRYS_PRIVATE_KEY=<solana-private-key-or-secret-json>
IRYS_GATEWAY_URL=https://gateway.irys.xyz
IRYS_NETWORK=devnet

JUDGE_BACKEND=anthropic
JUDGE_DAEMON_AUTOSTART=true
ANTHROPIC_API_KEY=<key>
ANTHROPIC_MODEL=claude-sonnet-4-5-20250929

Lit decrypt remains mocked in this slice.

Slice 5: Browser Lit + Irys Opt-in

Default local behavior still uses the API upload path and mock Lit envelope:

NEXT_PUBLIC_STORAGE_MODE=api
NEXT_PUBLIC_LIT_MODE=mock
NEXT_PUBLIC_LIT_NETWORK=datil-dev

To test browser wallet-backed Irys uploads and Lit encryption/decryption:

NEXT_PUBLIC_STORAGE_MODE=browser-irys
NEXT_PUBLIC_LIT_MODE=real
NEXT_PUBLIC_LIT_NETWORK=datil-dev

In real mode, /publish encrypts SKILL.md content before browser Irys upload, /submit uploads ExperienceBundle JSON through browser Irys, and /skill/[id] decrypts fetched skill content through Lit. /console intentionally remains on the deterministic mock/API path.