AegisThe agent skill problem
Your agent just ran a skill it didn't vet.
One edited skill and your agent leaks data, calls the wrong tool, or worse — and nothing in your logs would ever show it.
Anatomy of a skill hijack
Trusted skill installed
verified by no one
Its Markdown is silently swapped
content changes, no alert
Agent follows the poison
exfiltrates, no trace left
The fix
Don't trust skills. Verify them.
Aegis fingerprints every skill, evaluates it with a trustless AI attestor on Chainlink CRE, gates it behind policies you approve on Ledger, and names it on ENS. The verifier checks before it runs.
✗ Without Aegis
Skills run on blind trust. Tampering is invisible. No proof of what the agent followed.
✓ With Aegis
Every run checks hash vs. on-chain record. Tampered skills are blocked. Full audit trail.
Hash→AI attestor→Ledger policy→ENS verify