{"paper":{"title":"Runtime-Structured Task Decomposition for Agentic Coding Systems","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Runtime-structured task decomposition reduces retry costs in agentic coding systems by rerunning only failed subtasks.","cross_cats":["cs.AI"],"primary_cat":"cs.SE","authors_text":"Bing Zhang, Chad DeLuca, Hima Patel, Ruchi Mahindru, Shubhi Asthana","submitted_at":"2026-05-14T21:16:23Z","abstract_excerpt":"Agentic coding systems increasingly use large language models (LLMs) for software engineering tasks such as debugging, root cause analysis, and code review. However, many existing systems encode task logic, execution flow, and output generation inside monolithic prompts. This design creates brittle behavior, limited debuggability, and high retry costs because failures often require rerunning the full workflow.\n  We present runtime-structured task decomposition, an architectural approach in which task partitioning and execution flow are managed through executable control logic rather than promp"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"The runtime-structured approach reran only failed subtasks, reducing retry costs to 436 +/- 132 tokens for root cause analysis and 460 tokens for debugging, achieving up to 51.7% lower retry cost than monolithic systems and 73.2% lower retry cost than static decomposition baselines.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That output validation against predefined schemas reliably catches errors and that subtask failures remain sufficiently localized to enable partial reruns without cascading effects or requiring full workflow restarts, as implied by the description of the runtime-structured configuration.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Runtime-structured task decomposition reduces retry costs in agentic coding systems by up to 51.7% versus monolithic prompts by rerunning only failed subtasks on two software engineering workloads.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Runtime-structured task decomposition reduces retry costs in agentic coding systems by rerunning only failed subtasks.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"568f2ece6ac694226891ab70341e88b10af3291f23230413772199f12951b565"},"source":{"id":"2605.15425","kind":"arxiv","version":1},"verdict":{"id":"9354b60e-c7ab-4f5e-beae-260c306ad2c0","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T14:39:46.845058Z","strongest_claim":"The runtime-structured approach reran only failed subtasks, reducing retry costs to 436 +/- 132 tokens for root cause analysis and 460 tokens for debugging, achieving up to 51.7% lower retry cost than monolithic systems and 73.2% lower retry cost than static decomposition baselines.","one_line_summary":"Runtime-structured task decomposition reduces retry costs in agentic coding systems by up to 51.7% versus monolithic prompts by rerunning only failed subtasks on two software engineering workloads.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That output validation against predefined schemas reliably catches errors and that subtask failures remain sufficiently localized to enable partial reruns without cascading effects or requiring full workflow restarts, as implied by the description of the runtime-structured configuration.","pith_extraction_headline":"Runtime-structured task decomposition reduces retry costs in agentic coding systems by rerunning only failed subtasks."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.15425/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"cited_work_retraction","ran_at":"2026-05-19T15:55:05.144682Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"citation_quote_validity","ran_at":"2026-05-19T15:50:39.309795Z","status":"completed","version":"0.1.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T15:01:17.683496Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T14:50:24.728530Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T14:21:54.137555Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T13:33:22.700143Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"3eb3a2210f605fe96a7305bd8f8d3db9db94e2d19e62393eb57f44a163f96633"},"references":{"count":64,"sample":[{"doi":"","year":null,"title":"30th USENIX Security Symposium (USENIX Security 21) , pages=","work_id":"23f94b87-69cd-4613-96c4-8d02e8ad6cce","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Quantifying Memorization Across Neural Language Models","work_id":"35487ec1-b90b-4ace-95bd-1bce30064b2e","ref_index":2,"cited_arxiv_id":"2202.07646","is_internal_anchor":true},{"doi":"","year":null,"title":"Differentiation","work_id":"8e363b98-7d95-4d00-90a3-fc7e43376fa3","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"AIDev: A Large-Scale Dataset of Real-World","work_id":"857a80b8-3ad6-4d05-83f2-5bbc11e97ade","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Panel: Privacy Challenges and Opportunities in \\ LLM-Based \\ Chatbot Applications , author=","work_id":"f4ef9998-570a-4251-8e51-d770fcb65a6a","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":64,"snapshot_sha256":"ef096b7b491be90c94170f7fc62f7606032931835f03ce899dfbb659b3507725","internal_anchors":12},"formal_canon":{"evidence_count":2,"snapshot_sha256":"e26097ef306c24babd8f7cc53f7d2a53e2fbf0af8ed0a85f4a147d9e6c528fef"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}