{"id":1191484,"date":"2026-10-27T15:52:53","date_gmt":"2026-10-27T22:52:53","guid":{"rendered":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/publication\/dpiagent-divide-protocol-isolate-for-agentic-reproduction-test-generation\/"},"modified":"2026-10-30T09:38:44","modified_gmt":"2026-10-30T16:38:44","slug":"dpiagent-divide-protocol-isolate-for-agentic-reproduction-test-generation","status":"publish","type":"msr-research-item","link":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/publication\/dpiagent-divide-protocol-isolate-for-agentic-reproduction-test-generation\/","title":{"rendered":"DPIAgent: Divide, Protocol, Isolate for Agentic Reproduction Test Generation"},"content":{"rendered":"\n\n\n<p class=\"wp-block-paragraph\">Reproduction test generation, producing a failing-then-passing test that captures a reported bug, is a critical step in automated software engineering. Existing agentic methods treat this as a monolithic loop, despite the task inherently comprising two subtasks of distinct nature: diagnosing the root cause and writing a fail-to-pass test. Without explicit separation, the agent faces a compound objective with underspecified intermediate goals, leading to goal drift. We propose DPIAgent, a structured agentic framework built on three principles, Divide, Protocol, Isolate (DPI), that mitigates compound-objective ambiguity and goal drift: it Divides the task into single-objective phases of defect exploration and test generation; enforces a handoff Protocol that records the diagnosis and test plan, preventing context loss; and Isolates each phase&#8217;s action space by tailoring the toolset to its task, preventing irrelevant tools from misleading execution. On SWT-Bench Verified, DPIAgent outperforms seven baselines across three backbone LLMs. With DPI alone it reaches 81.76% success rate on GPT-5, the highest reported among open-source methods, gaining up to 11.88 points over the strongest baseline on GPT-5-Mini; adding test selection further raises it to 86.17%. Our analysis shows that architectural structure and backbone capability are complementary axes rather than substitutes, demonstrating DPI&#8217;s generalizability across model classes.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Reproduction test generation, producing a failing-then-passing test that captures a reported bug, is a critical step in automated software engineering. Existing agentic methods treat this as a monolithic loop, despite the task inherently comprising two subtasks of distinct nature: diagnosing the root cause and writing a fail-to-pass test. Without explicit separation, the agent faces a [&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":"Hao Liu","user_id":0},{"type":"text","value":"Steven Liu","user_id":0},{"type":"text","value":"Xin Zhang","user_id":0},{"type":"text","value":"Jane Luo","user_id":0},{"type":"user_nicename","value":"Yu Kang","user_id":"39381"},{"type":"text","value":"Jie Wu","user_id":0},{"type":"user_nicename","value":"Fangkai Yang","user_id":"41425"},{"type":"user_nicename","value":"Yangyu Huang","user_id":"41488"},{"type":"text","value":"Pengfei Gao","user_id":0},{"type":"user_nicename","value":"Scarlett Li","user_id":"37736"},{"type":"user_nicename","value":"Yan Lu","user_id":"34969"}],"msr_publishername":"","msr_publisher_other":"","msr_booktitle":"","msr_chapter":"","msr_edition":"","msr_editors":"","msr_how_published":"","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":"NeurIPS 2026","msr_doi":"","msr_arxiv_id":"2608.23341","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":"2026-08-24","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":8,"msr_day":24,"msr_microsoftintellectualproperty":false,"msr_pub_id":"56242781ec72c45b8999a2bc3a93021b8a25de39","msr_publication_uploader":[{"type":"url","title":"https:\/\/arxiv.org\/abs\/2608.23341","label_id":243109,"id":false,"viewUrl":false}],"msr_related_uploader":[],"msr_original_fields_of_study":[],"msr_s2_paper_id":"56242781ec72c45b8999a2bc3a93021b8a25de39","msr_s2_pdf_url":"","msr_citation_count_updated":"","msr_citation_count":0,"msr_influential_citations":0,"msr_reference_count":30,"msr_s2_open_access":false,"msr_s2_author_ids":[],"msr_pub_ids":[{"provider":"s2","id":"56242781ec72c45b8999a2bc3a93021b8a25de39"},{"provider":"arxiv","id":"2608.23341"},{"provider":"corpusid","id":"291327981"}],"msr_hide_image_in_river":0,"footnotes":""},"msr-research-highlight":[],"research-area":[13556,13560],"msr-publication-type":[193716],"msr-publisher":[],"msr-publication-cta":[],"msr-focus-area":[],"msr-locale":[268875],"msr-post-option":[],"msr-field-of-study":[246691,265728],"msr-conference":[259048],"msr-journal":[],"msr-impact-theme":[],"msr-pillar":[],"class_list":["post-1191484","msr-research-item","type-msr-research-item","status-publish","hentry","msr-research-area-artificial-intelligence","msr-research-area-programming-languages-software-engineering","msr-locale-en_us","msr-field-of-study-computer-science","msr-field-of-study-software-engineering-15"],"msr_publishername":"","msr_edition":"","msr_affiliation":"","msr_published_date":"2026-08-24","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":"","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\/2608.23341","label_id":243109,"id":false,"viewUrl":false}],"msr_related_uploader":[],"msr_citation_count":0,"msr_citation_count_updated":"","msr_s2_paper_id":"56242781ec72c45b8999a2bc3a93021b8a25de39","msr_influential_citations":0,"msr_reference_count":30,"msr_arxiv_id":"2608.23341","msr_s2_author_ids":[],"msr_s2_open_access":false,"msr_s2_pdf_url":null,"msr_attachments":[],"msr-author-ordering":[{"type":"text","value":"Hao Liu","user_id":0,"rest_url":false},{"type":"text","value":"Steven Liu","user_id":0,"rest_url":false},{"type":"text","value":"Xin Zhang","user_id":0,"rest_url":false},{"type":"text","value":"Jane Luo","user_id":0,"rest_url":false},{"type":"user_nicename","value":"Yu Kang","user_id":39381,"rest_url":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Yu Kang"},{"type":"text","value":"Jie Wu","user_id":0,"rest_url":false},{"type":"user_nicename","value":"Fangkai Yang","user_id":41425,"rest_url":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Fangkai Yang"},{"type":"user_nicename","value":"Yangyu Huang","user_id":41488,"rest_url":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Yangyu Huang"},{"type":"text","value":"Pengfei Gao","user_id":0,"rest_url":false},{"type":"user_nicename","value":"Scarlett Li","user_id":37736,"rest_url":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Scarlett Li"},{"type":"user_nicename","value":"Yan Lu","user_id":34969,"rest_url":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Yan Lu"}],"msr_impact_theme":[],"msr_research_lab":[199560,1012650],"msr_event":[],"msr_group":[],"msr_project":[],"publication":[],"video":[],"msr-tool":[],"msr_publication_type":"inproceedings","related_content":[],"_links":{"self":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1191484","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":6,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1191484\/revisions"}],"predecessor-version":[{"id":1192212,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1191484\/revisions\/1192212"}],"wp:attachment":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1191484"}],"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=1191484"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1191484"},{"taxonomy":"msr-publication-type","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-type?post=1191484"},{"taxonomy":"msr-publisher","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-publisher?post=1191484"},{"taxonomy":"msr-publication-cta","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-cta?post=1191484"},{"taxonomy":"msr-focus-area","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-focus-area?post=1191484"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1191484"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1191484"},{"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=1191484"},{"taxonomy":"msr-conference","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-conference?post=1191484"},{"taxonomy":"msr-journal","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-journal?post=1191484"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=1191484"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=1191484"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}