Aegis & Klethron
Confines autonomous coding agents with Aegis 1.0 (a ~14 µs, zero-token pre-syscall negative gate across five defense clans) paired with Klethron 0.1 (a passive, read-only posture and ledger-witness perimeter).
About this project
Aegis 1.0 stops unsafe agent actions inline before execution (~14 µs, zero LLM tokens, strictly workspace-confined). The Klethron 0.1 addition wraps a passive, zero-write outer observation perimeter around that gate: it reads bounded workspace/egress/cloud/fleet posture reports, cross-checks them against Aegis's append-only SHA-256 ledger witness tip, and fails dark whenever telemetry is missing or stale (>24h).
Under the hood
- Ring 01 · Aegis 1.0 Hot Path (~14 µs, 0 LLM tokens): deterministic pre-syscall negative gate (PASS / ASK / DENY) across five defense clans.
- Why Klethron was added: guarding single tool calls does not prove whether the outer workspace, network boundary, cloud RLS, or repository secrets remain clean.
- Ring 02 · Klethron 0.1 Cold Path (0 writes, 0 network calls): reads bounded offline posture JSON + SHA-256 ledger witness tips into one PASS / HOLD / UNKNOWN state.
- Unknown != Green & Hard Airgap: unscheduled or >24h stale sensors fail dark; observation sensors cannot mutate policy, run remediation, or write to git/disk.