Provenance record

Agent Harness Awareness: Turning Verified Telemetry into Operational Control for Tool-Using Agents

arXiv cs.AI (tier 1, academic) 2026-11-05T05:00:00.000Z Original ↗

Source note: Preprints. Not peer reviewed at time of posting.

Discourse valence

+50

Favourable

confidence 80% · 1 item

-100

-50

+50

+100

+50

Adverse readingFavourable reading

Single-model reading. This item did not meet the threshold for a multi-model panel.

Existential and catastrophic risk Loss of control and alignment Scaling and architecture

Excerpt as ingested

arXiv:2611.05391v1 Announce Type: new Abstract: Tool descriptions tell an agent what an action can do, but they do not necessarily reveal whether that action is healthy or suitable at the moment of use. Execution harnesses often possess this information through monitors, traces, and recent failures while the model choosing the next action does not. We call this gap operational blindness. We introduce Agent Harness Awareness, a trust-separated design in which the host supplies a small, fresh, verified state snapshot before a consequential action while retaining measurement, enforcement, evaluation, and final dispatch authority. We test its tools-and-network instance with ToolRoute, a controlled isolation benchmark that holds the task and candidate semantics constant and varies the operational information available to the model. In a 216-decision ablation across GPT-5.6 Sol, Claude Sonnet 5, and Gemini 3.7 Flash, verified tool-health state reduced mean observable policy regret by 96.9% against a same-shape neutral control (4,252 ms to 131 ms) and raised single-dispatch completion from 69.4% to 95.8%. In a separate 27-decision end-to-end replication with live local TCP/HTTP execution, OpenTelemetry client spans, and Toxiproxy-injected faults, verified state reduced mean regret from 8,956 ms to 0 ms (zero excess observable operational cost) and completed the live action in 9/9 model-by-fault cases versus 5/9 for the neutral control. These studies establish the tools-and-network instance of Agent Harness Awareness: selected host telemetry materially improves pre-dispatch operati

Every model that read this

Model	Provider	Stage	Score	Conf.	Latency	Prompt	When
Llama 3.3 70B	Meta	analysis	+50	80%	4048ms	v1.0.0 / m1.0.1	2026-11-05 05:43

Llama 3.3 70B · reading

Improves operational choice, reduces regret, but has potential risks and open extensions.

evidence: primary horizon: n/a

Evidence extracted

Verified tool-health state reduces mean observable policy regret by 96.9%
Verified state reduces mean regret from 8,956 ms to 0 ms

The chain

SOURCE     arXiv cs.AI (tier 1)
   ↓
DOCUMENT   197c429c-f8f0-4884-bd30-cda0eb9d4250
           https://arxiv.org/abs/2611.05391
   ↓
EVIDENCE   2 extracted excerpts
   ↓
MODEL RUN  1 run, methodology 1.0.1
   ↓
SCORE      +50  (Favourable)
   ↓
CONFIDENCE 80%