{"paper":{"title":"SkillGym: Internalizing Human Skills into LLMs for Real-World Problem Solving","arxiv_id":"2609.27717","primary_cat":"cs.CL","submitted_at":"2026-09-23T11:35:34+00:00","version_compiled":"v1"},"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 v1","sentences":[{"i":0,"p":0.7,"id":"s0","own":false,"role":"background","text":"Human-written agent skills encode rich workflows for real-world problem solving, but are typically used as external inference-time instructions rather than internalized as reusable model capabilities.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.08,"main_claim_p":0.05,"quantitative":false,"quantitative_p":0.1,"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":1,"p":0.96,"id":"s1","own":true,"role":"contribution","text":"We introduce \\texttt{SkillGym}, a framework that transforms these skills into executable, verifiable training environments for large language model agents.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.91,"main_claim_p":0.07,"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":0.0,"contribution":0.96,"interpretation":0.0}},{"i":2,"p":0.99,"id":"s2","own":true,"role":"method","text":"Its skill-to-task pipeline instantiates concrete tasks, verifies outcomes with code-based checkers, and assesses empirical skill dependence through contrastive executions.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.92,"main_claim_p":0.04,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.0,"method":0.99,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.01,"interpretation":0.0}},{"i":3,"p":0.53,"id":"s3","own":true,"role":"method","text":"We construct and release 2,756 environments across 12 categories and collect 8,364 successful trajectories from multiple models and harnesses, averaging 49 tool calls and over 60k logged text tokens.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.96,"main_claim_p":0.04,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.53,"result":0.07,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.4,"interpretation":0.0}},{"i":4,"p":0.88,"id":"s4","own":true,"role":"method","text":"These resources support supervised fine-tuning on verified workflows and reinforcement learning with outcome-based rewards.","hedge":{"p":0.85,"label":"asserted"},"own_p":0.63,"main_claim_p":0.07,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.0,"method":0.88,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.01,"contribution":0.11,"interpretation":0.0}},{"i":5,"p":1.0,"id":"s5","own":true,"role":"result","text":"Under Claude Code, supervised fine-tuning improves Qwen3.5-35B-A3B by 199 Elo on GDPval-AA v2, 19.10 percentage points on Terminal-Bench 2.1, and 28.13 and 12.38 points on SkillsBench v1.1 with and without skills, respectively.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.93,"main_claim_p":0.16,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":6,"p":1.0,"id":"s6","own":true,"role":"result","text":"Our 35B \\texttt{SkillGym-Agent} reaches 51.47\\% on skill-assisted SkillsBench, exceeding reported scores for Claude Sonnet 4.6, GPT-5.4 Mini, and DeepSeek V4 Pro.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.95,"main_claim_p":0.3,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":7,"p":0.79,"id":"s7","own":true,"role":"interpretation","text":"Without skills, it also surpasses skill-assisted bases under Codex and Claude Code, suggesting reusable procedural competence.","hedge":{"p":0.92,"label":"hedged_possibility"},"own_p":0.94,"main_claim_p":0.3,"quantitative":true,"quantitative_p":0.78,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.21,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.79}}]},"compiled":{"main_claim":{"p":0.9,"text":"Without skills, it also surpasses skill-assisted bases under Codex and Claude Code, suggesting reusable procedural competence.","location":"abstract s7","hedge_strength":{"p":0.92,"label":"hedged possibility"}},"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":6523,"cost_usd_reported":0.0},"compiled_at":"2026-09-24","bank_version":"abstract-p1-20260923","model_version":"classifier-1.13.0","abstract_states":{"claims_first":0.08,"states_limitation":0.14,"human_participants":0.11,"controlled_comparison":0.94,"randomized_assignment":0.02,"states_quantitative_result":0.99,"compares_with_prior_results":0.98,"mentions_code_or_data_availability":0.87},"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":0.92,"value":"method or tool","probabilities":{"review":0.0,"method_or_tool":0.92,"empirical_study":0.0,"simulation_study":0.0,"theoretical_result":0.0,"dataset_or_benchmark":0.08,"position_or_commentary":0.0}},"contribution_statement":{"p":1.0,"text":"We introduce \\texttt{SkillGym}, a framework that transforms these skills into executable, verifiable training environments for large language model agents.","location":"abstract s1"}},"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 v1"],"levels_not_run":["S sections","T body sentences","C citations","R cross-paper relations"],"sentences_examined":8,"sentences_in_abstract":8},"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"}}}