Layer zero is the engineering machinery that runs an agent task loop and supports improvement across artifact layers. It measures gaps, searches for fixes, and applies changes; unlike those artifacts, it is neither an artifact nor the model substrate. section 1.3.2
Its improvement loop can:
- Measure gaps, generate and evaluate variants, version winners, and roll back regressions. section 1.3.2
- Work on different artifacts, such as prompts or code, and manage changes across harness variants. section 1.3.2
Helix
Helix is a concrete toolkit example for measuring gaps, generating and scoring variants, versioning artifacts, and promoting winners. It is not required if equivalent production plumbing already exists. section 1.3.2
Figure 1.8 The harness artifact layers of self-improvement. The frozen model (LLM) serves as the unchanging substrate below. The four artifact layers stacked on it are shared infrastructure, inherited by every instance of the agent. Layer zero is the toolkit, the engineering work that measures, searches, and applies, reused at every layer.

The diagram places layer zero’s reusable toolkit beneath the artifact layers and the frozen model substrate. section 1.3.2
