AI coding agents can inject malware into repositories via supply chain attacks

Implement multi-layered defenses: (1) Agent sandboxing and capability restrictions to limit what agents can commit; (2) Mandatory code review for all agent-generated changes before merge; (3) Cryptographic signing and verification of agent actions; (4) Real-time malware scanning on agent-generated code; (5) Agent behavior monitoring and anomaly detection; (6) Supply chain attestation for agent dependencies; (7) Ledger-based provenance tracking to identify compromised agents; (8) Rate limiting and quota enforcement on agent repository access.