Fetching from the wire…
01
02
03
04
05
06
07
08
Agents2026-08-13 · source-backed
AMAP-ML/LongHorizon-Harness (675 stars since August 4, arXiv 2608.01964) executes each round as a bounded step with fresh context, verifies the actual result in the real computer, then checkpoints or feeds failure evidence forward. No new model, no agent replacement. The argument is that the model decides one round and the harness decides the loop, which is exactly the division the runtime-safety papers are converging on.
Each link below shares sources, entities, or timing with this story.
Shared entities / Same source / Shared topic / Earlier coverage
Both cover AMAP, August, Harness, LongHorizon; cite the same source (AMAP-ML/LongHorizon-Harness); overlapping topics (august, context).
Shared entities / Same source / Earlier coverage
Both cover AMAP, Harness, LongHorizon; cite the same source (AMAP-ML/LongHorizon-Harness); earlier AMAP coverage from 2026-08-07.
Shared entity: August / Same source domain / Shared topic / Earlier coverage / Tension
Both cover August; reported by the same outlet (github.com); overlapping topics (august, model).
Shared entity: Harness / Shared topic / Earlier coverage / Tension
Both cover Harness; overlapping topics (agent, bounded, context, each); earlier Harness coverage from 2026-07-30.
Both cover Harness; overlapping topics (agent, context, evidence, model); earlier Harness coverage from 2026-04-01.
Shared entity: August / Same source domain / Shared topic / Earlier coverage
Both cover August; reported by the same outlet (github.com); overlapping topics (agent, august, context).
Both cover August; reported by the same outlet (github.com); overlapping topics (agent, context, model).
Both cover August; reported by the same outlet (github.com); overlapping topics (agent, august, model).