{"slug":"skill-evolution","label":"Skill Evolution","item_count":3,"day_count":3,"source_count":2,"first_seen":"2026-10-19T07:29:21+00:00","last_updated":"2026-11-02T14:50:06+00:00","generated_at":"2026-11-04T00:12:28.797841+00:00","sources":["arxiv_agent_systems_research","arxiv_cs_cl"],"days":[{"date":"2026-10-19","items":[{"title":"EvoSkill v2: Fully Agentic Skill Evolution from Benchmark Adaptation to a Stateful Evolver","url":"http://arxiv.org/abs/2610.25931v1","source":"arxiv_agent_systems_research","type":"paper","summary_1line":"EvoSkill v2 is a three-agent system that edits the instructions (skills) an agent is given, leaving its model and harness intact. The user starts a run by describing a benchmark and supplying a grader, which a benchma...","why_it_matters":"Matches feed focus: agentic, harness.","sid":"3af6dffb8c0777d2","published":"2026-10-19T07:29:21+00:00","editor_note":"Skill evolution with model and harness held fixed."}]},{"date":"2026-10-26","items":[{"title":"Hierarchical Skill Evolution for Coding Agent","url":"http://arxiv.org/abs/2610.35616v1","source":"arxiv_cs_cl","type":"paper","summary_1line":"Skills provide agents with reusable experience to guide future decisions without modifying the underlying model. However, coding trajectories are often entangled with task- and repository-specific contexts, making it...","why_it_matters":"Matches feed focus: agent, harness, eval.","sid":"187923b6aa7d2a93","published":"2026-10-26T15:34:29+00:00","editor_note":"Adds hierarchy to disentangle reusable experience from repo context."}]},{"date":"2026-11-02","items":[{"title":"ScienceCUA: Resource-to-Skill Construction and Execution-Guided Skill Evolution for Scientific Computer Use Agents","url":"http://arxiv.org/abs/2611.03238v1","source":"arxiv_agent_systems_research","type":"paper","summary_1line":"Computer-use agents for scientific software require application-specific expertise. Much of this expertise is documented in human-authored software manuals and web pages. However, turning these resources into executab...","why_it_matters":"Matches feed focus: agent, evaluation.","sid":"1e71fb03632c55bc","published":"2026-11-02T14:50:06+00:00","editor_note":"Extends skill construction to documentation-driven scientific software."}]}],"editorial":{"tldr":"EvoSkill v2 (Oct 19) evolves agent skills, meaning instructions, without touching the model or harness. A Oct 26 paper extended this to coding agents with hierarchical skills.","stale":false,"whats_new":"ScienceCUA (Nov 2) builds skills from software manuals and web pages and refines them from execution feedback for scientific computer-use agents.","why_it_matters":"Skills are editable text, so agent behavior can improve without retraining; the open issue is whether evolved skills transfer across tasks and repos.","take_for_builders":"Version agent skills like code and keep a grader or benchmark, since evolution loops need a measurable signal to accept edits.","beats":[{"kicker":"LAUNCH","tone":"launch","headline":"EvoSkill v2: a three-agent system that edits skills","sids":["3af6dffb8c0777d2"]},{"kicker":"CODING","tone":"rising","headline":"Hierarchical skills separate reusable experience from repo-specific context","sids":["187923b6aa7d2a93"]},{"kicker":"NOW","tone":"now","headline":"ScienceCUA builds skills from manuals and evolves them by execution","sids":["1e71fb03632c55bc"]}],"open_questions":["Do evolved skills hold up on tasks and repositories outside their training benchmark?","Does skill evolution beat hand-written skills on a shared benchmark?"],"generated_at":"2026-11-04T00:00:00+00:00"}}