{"paper":{"title":"Evo-Harness: Context-to-Harness Skill Compilation for Self-Evolving Agents","arxiv_id":"2608.15071","primary_cat":"cs.AI","submitted_at":"2026-08-15T06:43:56+00:00","version_compiled":"v2"},"notice":"Every quoted sentence is the authors' text from the arXiv listing abstract. Labels and probabilities are the outputs of classifier-1.13.0 on the question bank abstract-p1-20260923, exactly as returned. No sentence in this record was written by a language model.","schema":"pith.compiled-record/v0","status":"compiled","abstract":{"source":"arXiv listing abstract v2","sentences":[{"i":0,"p":1.0,"id":"s0","own":false,"role":"background","text":"Learning from experience is critical for developing capable, self-improving large language model (LLM) agents.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.05,"main_claim_p":0.03,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":1.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":1,"p":1.0,"id":"s1","own":false,"role":"background","text":"Existing methods typically extract knowledge from accumulated trajectories via reflection, memory, rules, or skills.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.03,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":1.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":2,"p":0.7,"id":"s2","own":false,"role":"background","text":"However, agents in realistic environments continuously encounter novel tasks, often offering only a one-shot opportunity to improve.","hedge":{"p":0.95,"label":"asserted"},"own_p":0.07,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.27,"role_probabilities":{"gap":0.3,"method":0.0,"result":0.0,"background":0.7,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":3,"p":0.86,"id":"s3","own":false,"role":"background","text":"These executions yield rich but highly noisy contexts, entangling broadly useful lessons with task-specific artifacts.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.19,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.08,"role_probabilities":{"gap":0.13,"method":0.0,"result":0.0,"background":0.86,"limitation":0.01,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":4,"p":1.0,"id":"s4","own":false,"role":"gap","text":"Critically, prior works rarely validate their effectiveness on complex real-world tasks or isolate the underlying drivers of improvement.","hedge":{"p":0.99,"label":"asserted"},"own_p":0.06,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.24,"role_probabilities":{"gap":1.0,"method":0.0,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":5,"p":0.56,"id":"s5","own":true,"role":"contribution","text":"To address these gaps, we formulate online harness learning, where a frozen agent improves by continually updating a structured harness across sequential tasks.","hedge":{"p":0.99,"label":"asserted"},"own_p":0.58,"main_claim_p":0.09,"quantitative":false,"quantitative_p":0.05,"role_probabilities":{"gap":0.0,"method":0.44,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.56,"interpretation":0.0}},{"i":6,"p":0.94,"id":"s6","own":true,"role":"contribution","text":"This formulation enables a systematic study of key self-improvement factors through our proposed Evo-Harness.","hedge":{"p":0.86,"label":"asserted"},"own_p":0.66,"main_claim_p":0.08,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.0,"method":0.06,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.94,"interpretation":0.0}},{"i":7,"p":0.86,"id":"s7","own":true,"role":"method","text":"At its core, context-to-harness skill compilation distills noisy, single-shot executions into reusable skill harnesses for cross-domain and topic-level adaptation.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.67,"main_claim_p":0.14,"quantitative":false,"quantitative_p":0.13,"role_probabilities":{"gap":0.0,"method":0.86,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.14,"interpretation":0.0}},{"i":8,"p":0.76,"id":"s8","own":true,"role":"method","text":"To demonstrate the efficacy of one-shot skill compilation, we evaluate across five realistic benchmarks (TerminalBench2, SWE-bench, CL-Bench, -bench, WebArena-Infinity).","hedge":{"p":0.98,"label":"asserted"},"own_p":0.95,"main_claim_p":0.05,"quantitative":true,"quantitative_p":0.97,"role_probabilities":{"gap":0.0,"method":0.76,"result":0.01,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.23,"interpretation":0.0}},{"i":9,"p":0.79,"id":"s9","own":true,"role":"result","text":"Our extensive analysis demonstrates the effectiveness of Evo-Harness and provides a principled understanding of how LLM agents can effectively learn on the fly.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.96,"main_claim_p":0.82,"quantitative":false,"quantitative_p":0.05,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.79,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.05,"interpretation":0.16}},{"i":10,"p":0.84,"id":"s10","own":false,"role":"contribution","text":"Our code is available at https://github.com/A-EVO-Lab/a-evolve/tree/release/evo-harness.","hedge":{"p":0.96,"label":"asserted"},"own_p":0.4,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.0,"method":0.04,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.12,"contribution":0.84,"interpretation":0.0}}]},"compiled":{"main_claim":{"p":1.0,"text":"Our extensive analysis demonstrates the effectiveness of Evo-Harness and provides a principled understanding of how LLM agents can effectively learn on the fly.","location":"abstract s9","hedge_strength":{"p":1.0,"label":"asserted"}},"provenance":{"route":{"via":"TypeSafe API","model":"classifier-1.13.0","endpoint":"https://api.typesafe.ai/v1/systemone","provider":"TypeSafe"},"packing":"one request per paper","requests":1,"input_tokens":8097,"cost_usd_reported":0.0},"compiled_at":"2026-09-23","bank_version":"abstract-p1-20260923","model_version":"classifier-1.13.0","abstract_states":{"claims_first":0.08,"states_limitation":0.12,"human_participants":0.05,"controlled_comparison":0.12,"randomized_assignment":0.02,"states_quantitative_result":0.38,"compares_with_prior_results":0.26,"mentions_code_or_data_availability":0.99},"evidential_basis":{"p":0.98,"value":"observation or experiment","probabilities":{"argument":0.0,"literature":0.0,"proof_or_derivation":0.0,"analysis_of_existing_data":0.0,"observation_or_experiment":0.98,"simulation_or_computation":0.02}},"contribution_type":{"p":1.0,"value":"method or tool","probabilities":{"review":0.0,"method_or_tool":1.0,"empirical_study":0.0,"simulation_study":0.0,"theoretical_result":0.0,"dataset_or_benchmark":0.0,"position_or_commentary":0.0}},"contribution_statement":{"p":0.83,"text":"To address these gaps, we formulate online harness learning, where a frozen agent improves by continually updating a structured harness across sequential tasks.","location":"abstract s5"}},"coverage":{"tier":"abstract","levels_run":["P abstract"],"not_examined":["full text","figures and tables","references and citing sentences","relations to other papers"],"sources_read":["arXiv listing abstract v2"],"levels_not_run":["S sections","T body sentences","C citations","R cross-paper relations"],"sentences_examined":11,"sentences_in_abstract":11},"measured_accuracy":{"url":"https://pith.science/accuracy","measured_on":"2026-10-02","method":"two independent model families labelled a fresh random sample blind; the figure is how often the published label matches their consensus","sentence_role_agreement":0.92,"main_claim_agreement":0.908,"typed_citation_agreement":0.933,"typed_citation_agreement_95_interval":[0.917,0.945],"review_comments_accurate":0.758,"review_comments_accurate_95_interval":[0.684,0.82],"review_caught_admitted_error_found_or_related":0.448,"review_comments_judged":{"judged_by":["gpt-5.6-luna","grok-4.7"],"judged_against":"Pith's own text copy of each paper: text extracted from the PDF, the first 90,000 characters","measured_on":"2026-10-04","accurate":113,"decided":149,"sampled_reviews":200,"split_by_lane":false,"by_lane":"https://pith.science/accuracy#by-lane"}}}