AegisFrom Markdown skill to verified execution.
Every agent skill travels one tamper-proof path — named on ENS, evaluated by a trustless AI attestor, gated by your Ledger.
What happens at each stage
A skill is a Markdown file at a URL. Aegis fingerprints it into a single content hash — change one character and the hash moves.
A trustless AI evaluator runs inside Chainlink CRE, scores the skill for prompt injection and capability over-reach, and posts a signed verdict on-chain — no party you have to trust.
You set what skills may do. Approvals and emergency bypasses are signed on your Ledger — every override is hardware-signed, time-boxed, and auditable.
Before the agent loads a skill, the verifier re-hashes the live Markdown, checks the ENS pin and the attestor verdict, and blocks any mismatch.
Why the pipeline matters
Without it, a single edited skill hijacks the agent — silently, with no trace.