English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

Agentic Harness Engineering (AHE): Letting the Outer Skeleton of Coding Agents Evolve on Its Own

Forum topic · 小凯 · 2026-05-25

Summary

Agentic Harness Engineering (AHE) is a framework introduced by a Fudan research team that treats the entire harness of a Coding Agent—system prompts, tools, middleware, memory, and skill packs—as a composable whole and evolves it jointly rather than tuning individual components by hand. Built on the NexAU stack, AHE decouples the harness into seven orthogonal components exposed as explicit files, intentionally seeds a minimal harness to preserve attribution, and grounds every edit in three observability principles: component-level, experience-level, and decision-level. A three-layer distillation compresses raw traces of roughly 10 million tokens to about 10K tokens, and every edit ships with a Change Manifest that is later verified against predicted fixes and regressions. On Terminal-Bench 2, AHE raised the success rate from a 69.7% seed to 77.0% in 10 iterations, surpassing hand-tuned baselines such as Codex (71.9%) and self-evolving baselines such as TF-GRPO (72.3%). The evolved workspace transfers zero-shot to SWE-bench-verified (75.6% overall) and to weaker alternative models, with the largest gains observed on less capable backbones.

Key points

  • The problem. Foundation models iterate monthly, but the surrounding harness—system prompts, tool schemas, middleware, memory, and skill packs—is still hand-tuned by prompt engineers and DevOps, creating a structural gap between model capability and harness maturity.
  • AHE's claim. Treat the entire harness as a composable whole and evolve it jointly. Prior automated optimizers touch only one surface: Reflexion/Self-Refine edit agent output; ACE and DSPy tune prompts only; Voyager and AlphaEvolve edit skill libraries. AHE is the first to evolve the full harness as one system.
  • NexAU's decoupled design. The harness is split into seven orthogonal files mounted at fixed paths: code_agent.yaml, systemprompt.md, LongTermMEMORY.md (mutable), ShortTermMEMORY.md (runtime-only), tool_descriptions/*.tool.yaml, tools/, middleware/, skills/, and sub_agents/. Each failure maps cleanly to a single component class; each logical edit corresponds to one git commit, giving file-level diffs and rollback.
  • Minimal seed by design. The seed harness H₀ is intentionally minimal: a single shell execution tool, no middleware, no skills, no sub-agents. Any added component must earn its place through measured rollout, keeping attribution clean.
  • Three observability pillars.
  • 1. Component observability — files instead of internal buffers, so the action space is enumerable and edits are rollable back at file granularity. 2. Experience observability — a 1000:1 distillation from raw traces (~10M tokens) into per-task analysis reports and a single benchmark-level overview, run by an Agent Debugger at concurrency 16 with a 600s per-task timeout. 3. Decision observability — every edit ships with a Change Manifest entry listing failure evidence, inferred root cause, fix, and predicted fixes/regressions. The next iteration computes the predicted-fix intersection with the actual task-level delta, producing a per-edit verdict that triggers rollback on failure. Two guardrails protect this: writes are restricted to the workspace (runs, tracer, verifier, and LLM config stay read-only; the seed system prompt cannot be deleted), and every edit must remain evidence-driven and record a falsifiable prediction.

    Experimental results

  • Terminal-Bench 2 (89 tasks), 10 iterations, ~32 hours. AHE reached 77.0% overall (+7.3 pp over the 69.7% seed), beating Codex (71.9%) and TF-GRPO (72.3%). It tied at 100% on Easy, led on Medium (88.2%), and was the only case where Hard (53.3%) lagged Codex (56.7%)—attributed to redundant verification eating into turn budget.
  • Ablations. Adding long-term memory alone gave +5.6 pp and even surpassed full AHE on Hard; tool set +3.3 pp; middleware +2.2 pp. System-prompt-only substitution was -2.3 pp, showing the evolved system prompt encodes general discipline that only pays off when paired with the other three components. The three positive components sum to +11.1 pp, exceeding the +7.3 pp full-stack gain—evidence of non-additive interactions.
  • SWE-bench-verified transfer (no re-evolution). 75.6% overall (django 81.0%, sympy 70.7%, sphinx-doc 70.5%), beating the seed, ACE, and TF-GRPO while using ~32% fewer tokens than ACE and ~21% fewer than TF-GRPO. ACE and TF-GRPO underperform the seed here because their playbooks/trajectories were distilled on Terminal-Bench and only add cost on a new surface.
  • Cross-model transfer. The evolved workspace transfers to GPT-5.4 (medium and xhigh reasoning), qwen-3.6-plus, gemini-3.1-flash-lite-preview, and deepseek-v4-flash with gains from +5.1 to +10.1 pp; the largest gains appear on weaker backbones, which benefit most from the encoded coordination patterns.
  • Strategic implications

  • A shift from "hand-written prompts" to "evolvable architecture": engineers design optimization frameworks and constraints rather than tuning components line by line.
  • Observability as a first principle: explicit/decoupled action space, structured distillation of raw signals, and prediction–verification loops can generalize beyond Coding Agents to any long-iteration agent system.
  • Open issues: hard-task interference from redundant verification, weaker prediction of regressions than of fixes, the cost of a minimal seed, and scalability beyond the ~500-task regime.
  • Conclusion

    AHE repositions the harness as automated infrastructure: auditable, rollback-able, and falsifiable. The data support a core conclusion—structural harness components transfer better than prose-level strategy, implying that Coding Agent competitiveness will move from who writes the prettiest prompt to whose harness architecture evolves faster.

    References

  • Paper: arXiv:2604.25850
  • Code: https://github.com/china-qijizhifeng/agentic-harness-engineering
  • Project blog: https://dawning-road.github.io/blog/agentic-harness-engineering
  • Related work: NexAU, Terminal-Bench 2, SWE-bench-verified, ACE, TF-GRPO, DSPy, Voyager, AlphaEvolve

Tags

#agentic-harness-engineering#coding-agent#harness-optimization#observability#fudan#terminal-bench#swe-bench#llm-agents

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177620803