{"id":1190638,"date":"2026-10-21T10:26:02","date_gmt":"2026-10-21T17:26:02","guid":{"rendered":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/publication\/certified-long-horizon-code-agent-evolution-via-validation-gated-skill-optimization\/"},"modified":"2026-10-22T15:27:05","modified_gmt":"2026-10-22T22:27:05","slug":"certified-long-horizon-code-agent-evolution-via-validation-gated-skill-optimization","status":"publish","type":"msr-research-item","link":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/publication\/certified-long-horizon-code-agent-evolution-via-validation-gated-skill-optimization\/","title":{"rendered":"Certified Long-Horizon Code Agent Evolution via Validation-Gated Skill Optimization"},"content":{"rendered":"\n\n\n<p class=\"wp-block-paragraph\">Long horizon agent self-evolution without model weight updates is essential for enabling deployed agents to accumulate reusable skills and improve over time. Prior self-evolution work has focused primarily on short-horizon tasks, while repository-level software engineering remains unexplored despite being an ideal testbed for long-horizon adaptation. In this setting, agents are required to solve streams of sequential tasks, navigate complex dependencies with evolving repositories and persistently store and reuse experience. Text-based skill optimization offers an efficient, non-parametric approach for such adaptation. However, existing methods often suffer from unstable updates, performance drawdown, and agent collapse over extended deployments. In this paper, we formalize the concept of in-context self-evolution and introduce VALVE, a validated-gated framework for long-horizon skill optimization. We establish finite convergence, provide theoretical guarantees for future-task gain and drawdown, and derive the validation and evaluation holdout sizes required for a prescribed tolerance, with leading-order scaling Empirically, our pipeline, VALVE achieves stable self-improvement over evolution horizon spanning more than 1,000 SWE tasks, with average final and peak gains of <math><mn>14.9<\/mn><\/math> and <math><mn>16.5<\/mn><\/math> points across three frontier models (GPT-5.5, Claude-4.6 and MiniMax-M2.7). The validation gate reduces average drawdown by 75% and produces an 11x more compact skill bank than ungated evolution. We further present extensive ablations identifying the design choices most critical to long-horizon skill evolution.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Long horizon agent self-evolution without model weight updates is essential for enabling deployed agents to accumulate reusable skills and improve over time. Prior self-evolution work has focused primarily on short-horizon tasks, while repository-level software engineering remains unexplored despite being an ideal testbed for long-horizon adaptation. In this setting, agents are required to solve streams of [&hellip;]<\/p>\n","protected":false},"featured_media":0,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","msr-author-ordering":[{"type":"text","value":"Yifan Wang","user_id":0},{"type":"user_nicename","value":"Hao Cheng","user_id":"39922"},{"type":"text","value":"Xiaomin Li","user_id":0},{"type":"text","value":"Yuexing Hao","user_id":0},{"type":"text","value":"H. Ramesh","user_id":0},{"type":"text","value":"Dongwon Jung","user_id":0},{"type":"text","value":"Hao Tang","user_id":0},{"type":"text","value":"Keru Wang","user_id":0},{"type":"text","value":"Chenliang Zhou","user_id":0},{"type":"user_nicename","value":"Qianhui Wu","user_id":"40741"},{"type":"user_nicename","value":"Wenlin Yao","user_id":"44262"},{"type":"text","value":"Ananth Grama","user_id":0},{"type":"user_nicename","value":"Andrzej Banburski-Fahey","user_id":"43299"},{"type":"user_nicename","value":"Baolin Peng","user_id":"43779"},{"type":"user_nicename","value":"Jaron Lanier","user_id":"32148"},{"type":"user_nicename","value":"Jianfeng Gao","user_id":"32246"}],"msr_publishername":"","msr_publisher_other":"","msr_booktitle":"","msr_chapter":"","msr_edition":"","msr_editors":"","msr_how_published":"arXiv","msr_isbn":"","msr_issue":"","msr_journal":"","msr_number":"","msr_organization":"","msr_pages_string":"","msr_page_range_start":"","msr_page_range_end":"","msr_series":"","msr_volume":"","msr_copyright":"","msr_conference_name":"","msr_doi":"","msr_arxiv_id":"2609.32990","msr_mag_id":"","msr_other_authors":"","msr_other_contributors":"","msr_speaker":"","msr_award":"","msr_affiliation":"","msr_institution":"","msr_host":"","msr_version":"","msr_duration":"","msr_release_tracker_id":"","msr_highlight_type":"","msr_date_display_format":"","msr_main_download_label":"","msr_external_link_label":"","msr_doi_label":"","msr_published_date":"2026-09-26","msr_startdate":"","msr_presentation_date":"","msr_highlight_text":"","msr_notes":"","msr_longbiography":"","msr_publicationurl":"","msr_external_url":"","msr_secondary_video_url":"","msr_conference_url":"","msr_journal_url":"","msr_year":2026,"msr_month":9,"msr_day":26,"msr_microsoftintellectualproperty":false,"msr_pub_id":"372144d8ffe41bd587ebef35fc606b019d6171f7","msr_publication_uploader":[{"type":"url","title":"https:\/\/arxiv.org\/abs\/2609.32990","label_id":243109,"id":false,"viewUrl":false}],"msr_related_uploader":[],"msr_original_fields_of_study":[],"msr_s2_paper_id":"372144d8ffe41bd587ebef35fc606b019d6171f7","msr_s2_pdf_url":"","msr_citation_count_updated":"","msr_citation_count":1,"msr_influential_citations":0,"msr_reference_count":41,"msr_s2_open_access":false,"msr_s2_author_ids":[],"msr_pub_ids":[{"provider":"s2","id":"372144d8ffe41bd587ebef35fc606b019d6171f7"},{"provider":"arxiv","id":"2609.32990"},{"provider":"corpusid","id":"292435205"}],"msr_hide_image_in_river":0,"footnotes":""},"msr-research-highlight":[],"research-area":[13561,13556,13560],"msr-publication-type":[270373],"msr-publisher":[],"msr-publication-cta":[],"msr-focus-area":[],"msr-locale":[268875],"msr-post-option":[],"msr-field-of-study":[269187,246691],"msr-conference":[],"msr-journal":[],"msr-impact-theme":[],"msr-pillar":[],"class_list":["post-1190638","msr-research-item","type-msr-research-item","status-publish","hentry","msr-research-area-algorithms","msr-research-area-artificial-intelligence","msr-research-area-programming-languages-software-engineering","msr-locale-en_us","msr-field-of-study-artifical-intelligence","msr-field-of-study-computer-science"],"msr_publishername":"","msr_edition":"","msr_affiliation":"","msr_published_date":"2026-09-26","msr_host":"","msr_duration":"","msr_version":"","msr_speaker":"","msr_other_contributors":"","msr_booktitle":"","msr_pages_string":"","msr_chapter":"","msr_isbn":"","msr_journal":"","msr_volume":"","msr_number":"","msr_editors":"","msr_series":"","msr_issue":"","msr_organization":"","msr_how_published":"arXiv","msr_notes":"","msr_highlight_text":"","msr_release_tracker_id":"","msr_original_fields_of_study":"","msr_download_urls":"","msr_external_url":"","msr_secondary_video_url":"","msr_longbiography":"","msr_microsoftintellectualproperty":0,"msr_main_download":"","msr_publicationurl":"","msr_doi":"","msr_publication_uploader":[{"type":"url","title":"https:\/\/arxiv.org\/abs\/2609.32990","label_id":243109,"id":false,"viewUrl":false}],"msr_related_uploader":[],"msr_citation_count":1,"msr_citation_count_updated":"","msr_s2_paper_id":"372144d8ffe41bd587ebef35fc606b019d6171f7","msr_influential_citations":0,"msr_reference_count":41,"msr_arxiv_id":"2609.32990","msr_s2_author_ids":[],"msr_s2_open_access":false,"msr_s2_pdf_url":null,"msr_attachments":[],"msr-author-ordering":[{"type":"text","value":"Yifan Wang","user_id":0,"rest_url":false},{"type":"user_nicename","value":"Hao Cheng","user_id":39922,"rest_url":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Hao Cheng"},{"type":"text","value":"Xiaomin Li","user_id":0,"rest_url":false},{"type":"text","value":"Yuexing Hao","user_id":0,"rest_url":false},{"type":"text","value":"H. Ramesh","user_id":0,"rest_url":false},{"type":"text","value":"Dongwon Jung","user_id":0,"rest_url":false},{"type":"text","value":"Hao Tang","user_id":0,"rest_url":false},{"type":"text","value":"Keru Wang","user_id":0,"rest_url":false},{"type":"text","value":"Chenliang Zhou","user_id":0,"rest_url":false},{"type":"user_nicename","value":"Qianhui Wu","user_id":40741,"rest_url":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Qianhui Wu"},{"type":"user_nicename","value":"Wenlin Yao","user_id":44262,"rest_url":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Wenlin Yao"},{"type":"text","value":"Ananth Grama","user_id":0,"rest_url":false},{"type":"user_nicename","value":"Andrzej Banburski-Fahey","user_id":43299,"rest_url":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Andrzej Banburski-Fahey"},{"type":"user_nicename","value":"Baolin Peng","user_id":43779,"rest_url":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Baolin Peng"},{"type":"user_nicename","value":"Jaron Lanier","user_id":32148,"rest_url":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Jaron Lanier"},{"type":"user_nicename","value":"Jianfeng Gao","user_id":32246,"rest_url":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Jianfeng Gao"}],"msr_impact_theme":[],"msr_research_lab":[199565],"msr_event":[],"msr_group":[],"msr_project":[],"publication":[],"video":[],"msr-tool":[],"msr_publication_type":"misc","related_content":[],"_links":{"self":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1190638","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item"}],"about":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-research-item"}],"version-history":[{"count":4,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1190638\/revisions"}],"predecessor-version":[{"id":1190852,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1190638\/revisions\/1190852"}],"wp:attachment":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1190638"}],"wp:term":[{"taxonomy":"msr-research-highlight","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-highlight?post=1190638"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1190638"},{"taxonomy":"msr-publication-type","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-type?post=1190638"},{"taxonomy":"msr-publisher","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-publisher?post=1190638"},{"taxonomy":"msr-publication-cta","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-cta?post=1190638"},{"taxonomy":"msr-focus-area","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-focus-area?post=1190638"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1190638"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1190638"},{"taxonomy":"msr-field-of-study","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-field-of-study?post=1190638"},{"taxonomy":"msr-conference","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-conference?post=1190638"},{"taxonomy":"msr-journal","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-journal?post=1190638"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=1190638"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=1190638"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}