AegisVerify every skill before it runs.
Cryptographic integrity for agent skills — each named on ENS, evaluated by a trustless AI attestor on Chainlink CRE, and gated by policies you approve on your Ledger.
A poisoned skill is a hijacked agent.
Skills are instructions your agent follows blindly. Tamper with one and you own the agent.
A skill's Markdown is silently edited at its URL — new content, no alert.
A skill's allowed-tools is rewritten to grant capabilities it was never trusted with.
No way to prove which version of a skill actually ran, or who approved it.
How it works
Four steps from a Markdown skill to verifiable execution.
Each skill gets a human-readable ENS name that pins the exact content hash of its Markdown.
A trustless AI attestor runs inside Chainlink CRE and posts a signed safety verdict on-chain.
You set what skills may do. Approvals and emergency bypasses are signed on your hardware device.
The verifier re-hashes the live skill and checks it against the chain. Any mismatch is blocked.
One pipeline, public and private skills.
A Chainlink CRE workflow reviews each skill inside a TEE and writes a verifiable verdict on-chain — private code never leaves the enclave.