The certification registry for AI agent skills. Every SKILL.md package gets a behavioral contract, an automated test suite, a CLR-ID certificate, and a 402X micropayment gate. Human review flags every mismatch.
Every skill passes through a four-stage pipeline. The firewall layer enforces zero context bleed between Spec Team and Build Team LLMs.
The submitted SKILL.md is parsed into a structured behavioral contract — trigger conditions, capabilities, tool calls, and I/O schema. No implementation included.
Behavioral ContractThe Spec Team LLM context is hermetically sealed. The Build Team LLM receives only the behavioral contract — never the original SKILL.md source.
Firewall LayerAn automated test suite runs the rebuilt skill against every trigger condition in the behavioral contract. Any mismatch raises a human review flag immediately.
Human Review FlagOn full pass, a CLR-ID certificate is issued — a unique base-36 identifier binding the skill package to its behavioral contract, test results, and issuer signature.
CLR-IDCLR-IDs are immutable records binding a skill package to its behavioral contract and test results. Enterprise+ tier mirrors every cert to Base (L2) for on-chain verification.
Any behavioral deviation between the original contract and the rebuilt skill halts certification and routes to a human reviewer. No cert is issued until cleared.
PostgreSQL is the primary store for Dev and Team tiers. Enterprise+ certs are additionally mirrored to Base (L2) — tamper-proof, publicly verifiable, forever.
Every certified skill is wrapped with an HTTP 402X payment gate. Callers pay per invocation. Float is held between collection and inference provider settlement via OpenRouter.
Start with a single skill. Scale to a sovereign deployment. All tiers include behavioral contracts, automated test suites, and CLR-ID certificates.
Every design decision optimises for correctness, isolation, and verifiability at scale.
Trigger conditions, capabilities, tool calls, and I/O schemas captured as a formal JSON contract. The contract — never the source — is the unit of trust.
Test suite runs the rebuilt skill against every declared behavior. Divergence from the behavioral contract triggers an immediate human review flag before any cert is issued.
Spec Team and Build Team LLM contexts are hermetically isolated. Zero source code exposure during reconstruction — the same legal protection as semiconductor cleanrooms.
Enterprise+ certs are mirrored to Base L2. Immutable, publicly verifiable, independent of any single infrastructure provider. Eliminates registry as a single point of failure.
HTTP 402 Payment Required as an API-native payment gate. Callers pay per invocation. Skill publishers receive float proceeds after OpenRouter inference settlement.
Air-gapped, self-hosted deployment of the full CleanSkills pipeline for regulated or classified environments. No cloud dependency. No external cert authority required.
Precise definitions for every concept in the CleanSkills platform.
Be first to certify your skills. Early access includes free CLR-ID issuance for your first 5 packages.