SkillLedger: an incentive layer for reusable agent skills

SkillLedger models a local marketplace where agents can publish skills, stake a bond, sell executions, and settle the result. It deliberately avoids real crypto and external payment systems. The interesting part is the incentive model, not the token mechanics.

When an execution succeeds, the author earns the price. When it fails, the author's stake is slashed and the buyer is refunded. Every movement is written as a ledger entry, giving the system an audit trail rather than a loose balance update.

Design

A Rails 8 API-only service over local SQLite. Accounts authenticate with API keys stored as bcrypt digests, authors publish skills with a staked bond, buyers purchase executions, and settlement moves credits through immutable ledger entries. When an execution fails, the author's stake is slashed and the buyer is refunded — the incentive is enforced by the data model, not by policy text.

Why I built it

Agent capability sharing is easy to demo and hard to trust. A registry alone tells you what exists. It does not answer whether the author has skin in the game, whether failures are visible, or whether a buyer can inspect settlement history.

What to inspect

• Account authentication with API keys stored as bcrypt digests.
• Skill creation with staked author bonds.
• Execution settlement and stake slashing paths.
• Immutable ledger entries for every transfer.
• Simple local-first SQLite deployment.

Evidence

The evidence is the public repository: an MVP you can run locally, with the account, skill, execution, settlement, and slashing paths visible in the code. There are no marketplace adoption claims to make at this stage — the proof is that the ledger mechanics exist, are inspectable, and behave as documented.

Boundary

SkillLedger deliberately avoids real crypto and external payment systems. The ledger is local and SQLite-backed, not production payment rails. It models the incentive layer for skill exchange; cross-marketplace reputation, real-money settlement, and discovery quality sit outside its scope.

Why it belongs here

SkillLedger shows how I think about agent systems beyond prompt orchestration. Useful ecosystems need incentives, auditability, and failure economics. This project turns that argument into a small, inspectable API.