Multi-Task Harness Abstraction via Hyper-Agent
Multi-task harness abstraction is a two-tiered optimization workflow designed to synthesize durable procedural runbooks across broad task families:
- Task-Level Execution & Proposal: Individual task agents execute training instances and propose localized procedural changes directly from their observed failure traces.
- Hyper-Agent Evaluation: A more capable supervisory hyper-agent evaluates candidate proposals across multiple instances, evaluating their generality, safety, and cross-task compatibility.
- Reconciliation & Freezing: The hyper-agent reconciles validated lessons into a shared family-level runbook and removes idiosyncratic task workarounds before freezing the profile for held-out evaluation.
This division of labor prevents singular instance failures from triggering over-specialized family rules while automating the synthesis of persistent procedural memory.
0
1
Tags
Prep Sessions
Long-Horizon Agent Reliability: Stateful Scaffolding and Runtime Verification @ University of Michigan - Ann Arbor
Ch.3 State Persistence and Continuous Optimization - Long-Horizon Agent Reliability: Stateful Scaffolding and Runtime Verification @ University of Michigan - Ann Arbor
Failure-Driven Harness Optimization and Procedural Memory - Long-Horizon Agent Reliability: Stateful Scaffolding and Runtime Verification @ University of Michigan - Ann Arbor
Related
Failure-Driven Harness Optimization
Golden Rules for Harness Profile Development
Control Drift in Multi-Task Harness Optimization
Sparse Routing in Shared Agent Runbooks
Multi-Task Harness Abstraction via Hyper-Agent
Procedural Practices in Agent Harnesses
Single-Task Harness Adaptation Loop
Selective Filtering in Harness Learning
Procedural-Memory Gap in Autonomous Agents
Which entities are responsible for proposing targeted modifications to runtime controls after an execution failure?
True or False: Proposed harness modifications take effect immediately in production without requiring review or regression testing against existing tasks.
In failure-driven harness optimization, how are postmortem findings applied to prevent future agent execution failures without retraining or modifying underlying model weights?
Golden Rules for Harness Profile Development
Control Drift in Multi-Task Harness Optimization
Multi-Task Harness Abstraction via Hyper-Agent
Learn After
In the multi-task harness abstraction workflow, what is the primary responsibility of the supervisory hyper-agent during the evaluation phase?
In multi-task harness abstraction, individual task agents freeze the shared family-level runbook before submitting their failure traces to the hyper-agent.
Why does multi-task harness abstraction employ a two-tiered division of labor between task agents and a supervisory hyper-agent rather than allowing task agents to update the family runbook directly?