SkillHone is a harness for continual evolution of agent skills that preserves structured decision history alongside skill revisions. The history records diagnoses, evaluation evidence, revisions, outcomes, and rejected alternatives, allowing later sessions to reuse prior rationale instead of rediscovering it. Role-separated subagents execute candidate skills on practice probes, use redacted reports, and propose revisions. The abstract reports gains over a commercially backed deep-research agent of 15.8 points on GAIA and 3.2 points on WebWalkerQA-EN, plus an average 18.8-point accuracy improvement across seven internal tool-mediated analysis settings. The system reportedly operates without an integrated search stack.
No heat snapshots are available in the last 24 hours.