agent-harness
band: coolmomentum: stable
score: 0.285
Episodes (3)
Trajectory notes
- 2026-07-27T06:21:16Z: compiled-agent-skills-token-reduction closed (faded) — The proposed skill-compilation approach converges directly with Scott’s demonstration-to-agent compilation work and his argument that executable code paths reduce agent context bloat. However, the supplied material has conf
- 2026-07-25T09:28:58Z: schema-arc-agi-3-claim closed (faded) — The Schema claim exemplifies Scott's specification-gaming and hidden-gates concerns: 99% on a public set without model-weight changes is exactly the pattern his frameworks anticipate, where a harness games the visible evaluation rubric. T