No search stack — still +15.8 GAIA points over a commercial deep-research agent

Notes on SkillHone: A Harness for Continual Agent Skill Evolution Through Persistent Decision History (arXiv:2606.08671) — Zhiwei Li, Yong Hu · June 2026

Note published July 25, 2026 · written by SkillFed’s research pipeline from the paper above · how these notes are made

SkillHone treats agent skill maintenance as a logging problem, not an editing one. Every development step — probe run, diagnosis, candidate revision, accept or reject — gets written down as a structured record, and the growing chain becomes a persistent decision history that later sessions query instead of re-deriving from scratch. Two role-separated subagents split the actual work: an evaluation team sees the full oracle targets, validators, and execution traces for a probe but can never touch the skill file, while an optimization team can rewrite the skill but only ever reads a redacted report of what evaluation found. A dispatcher just routes messages between the two and logs outcomes; it holds no repository-write authority of its own.

Tested on GAIA and WebWalkerQA-EN, skills evolved this way — with no pre-integrated search stack — beat a commercially-backed deep-research agent that does have curated retrieval: 15.8 points ahead on GAIA, 3.2 on WebWalkerQA-EN. They also beat prior skill-optimization baselines, GEPA-style reflective optimization included, starting from the identical skill pool. The resulting skill bundle transfers to a different execution model (Claude Sonnet 4.6) with zero re-optimization, landing at 72.4% on GAIA. An ablation shows where the credit belongs: pull out the decision history and GAIA drops 13.4 points, WebWalkerQA-EN drops 10.9 — roughly double what pulling out the role separation costs (6.4 and 5.3). The memory is doing most of the work, not just the access-control split. A separate deployment run across seven internal tool-use scenarios lifts accuracy 18.8 points on average, though the range runs from +30 down to +0.