EvoX: Rethinking the Agent Harness as a Runtime Policy
Organizations: Infinite Evolution Lab, EvoMap
Abstract
Large language model (LLM) agents increasingly interact with tools and environments, yet plausible model decisions do not always translate into successful execution due to tool-call mismatches, interrupted workflows, and unsupported completion claims. We present EvoX, a Rust-based, multi-model agent runtime guided by The Second Policy, which views the harness as an Operational Policy complementing the foundation model's Semantic Policy. EvoX combines model-facing guidance with deterministic runtime control through three complementary mechanisms: Action Realization, Execution Continuity, and Outcome Grounding. To evaluate harnesses without suppressing their native behaviors, we introduce Contrast, a controlled and auditable framework integrating explicit fairness constraints, provider-bound request verification, and evidence-based adjudication. Across five benchmarks, six foundation models, and six agent harnesses, EvoX achieves the highest numerical success rate of 56.2% over 691 common task--model pairs, exceeding the average of the other five harnesses by approximately 3.0 percentage points. Its larger advantages on GLM-5.3 (+10.7 points) and Kimi K3 (+7.9 points) highlight the model-dependent effectiveness of harness design and the potential for cross-provider adaptability. Achieved without memory, evolution, or extensions, these results establish EvoX as a competitive multi-model agent runtime and underscore the importance of harness-level operational control in realizing foundation-model capabilities.
Figures & tables
| Benchmark | EvoX | Pi | Claude Code | Hermes | Prime | Codex | (pp) |
| Aider Polyglot | 64.7 | 61.7 | 62.3 | 63.5 | 61.1 | 55.1 | +4.0 |
| GAIA | 86.7 | 85.7 | 82.9 | 85.7 | 85.7 | 78.1 | +3.0 |
| Harbor Index | 27.3 | 25.9 | 31.5 | 24.5 | 23.1 | 27.3 | +0.8 |
| SWE-Live | 48.4 | 49.2 | 49.2 | 43.7 | 35.7 | 37.3 | +5.4 |
| Terminal-Bench 2.1 | 57.3 | 60.7 | 55.3 | 59.3 | 50.0 | 54.0 | +1.5 |
| Normal | Other outcomes | Rate (%) | ||||||||
| Harness | Pass | Fail | Exit | TO-F | TO-P | Cap-P | Cap-F | Other | Normal | Pass Normal |
| EvoX | 373 | 134 | 117 | 49 | 4 | 12 | 2 | 0 | 73.4 | 73.6 |
| Pi | 365 | 187 | 5 | 107 | 6 | 18 | 2 | 1 | 79.9 | 66.1 |
| Claude Code | 344 | 161 | 111 | 30 | 5 | 37 | 3 | 0 | 73.1 | 68.1 |
| Hermes | 361 | 150 | 159 | 2 | 0 | 14 | 5 | 0 | 74.0 | 70.6 |
| Prime Agent | 330 | 152 | 119 | 68 | 4 | 15 | 3 | 0 | 69.8 | 68.5 |
LLM Agent Harnesses
LLM agent harnesses are the software and runtime systems that organize an agent’s tools, context, state, execution, and safeguards.
See all 428 papers