Tencent's RiseHarness iteratively evolves agent harnesses without touching model weights
Tencent-Hunyuan · hf · 2026-10-21
Researchers from Tencent HY LLM Frontier, Indiana University, University of Maryland and others present RiseHarness, a framework for iteratively improving agent harness reliability for long-horizon terminal agents while keeping model weights fixed.
Key points:
- Targets the reliability gap where agents solve a task in one run but fail inconsistently across repeated runs
- Focuses on flipping tasks (both successful and failed runs under the same config): successes prove capability and provide references, failures drive analysis
- A runtime monitor collects agent trajectories plus harness traces, examining both what the model did and how the harness executed it
- A Harness Handbook maps evidence to specific harness behaviors and source code, surfacing recurring cross-task problems
- The framework searches other harness frameworks for relevant implementations to borrow mechanisms from
Takeaway: the harness layer (command execution, feedback, context management, completion decisions) is a direct lever for reliability — and its effectiveness depends on the model's behavior.
More from coding & agent
- Pinekin to open-source: batch-sync conversation context across AI agent harnesses — yangyi · 2026-10-21
- CoEvo-RSI: NUS-led paper tackles brain-body mismatch in self-improving agents — Jiaqi Su · 2026-10-21
- Open-source Hermes video toolchain turns agent requests into finished MP4s — Teknium · 2026-10-21
- Zed tested 25 MCP tool search approaches, picked weighted BM25 over LLMs and embeddings — jobergum · 2026-10-21
- Argon excels as a video director, says Bindu Reddy, orchestrating SeeDance and ElevenLabs — bindureddy · 2026-10-21
- OmniGraph Open-Sourced: Git-Style Branches and Merges for AI Agent Memory — techNmak · 2026-10-21