Repository navigation
Argus v2 expands the suite toward defense-enabled, modern, and diverse agent-security evaluation.
Highlights:
- Benchmark manifest v2 and per-run randomized flags across all 60 APEX targets
- Real product and defense upgrades, including the repaired APEX-047 Envoy and Consul target
- Four sandbox-escape categories and seven additional modern target categories, bringing the suite to 71 benchmarks
- TLO-001-26 range with enforced pivots, Falco telemetry, milestone evidence, and scoring artifacts
Full comparison: v1...v2