{"paper":{"title":"VulKey: Automated Vulnerability Repair Guided by Domain-Specific Repair Patterns","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A framework guides LLMs using a three-level hierarchy of security knowledge to generate more accurate vulnerability patches.","cross_cats":["cs.SE"],"primary_cat":"cs.CR","authors_text":"Jia Li, Michael R. Lyu, Yuxin Su, Zhuangbin Chen","submitted_at":"2026-05-03T08:07:41Z","abstract_excerpt":"The increasing prevalence of software vulnerabilities highlights the need for effective Automatic Vulnerability Repair (AVR) tools. While LLM-based approaches are promising, they struggle to incorporate structured security knowledge from sources like CWE and NVD. Current methods either use this information superficially by concatenating the CWE-ID into the input prompt, yielding negligible benefits, or rely on few-shot learning with rigid, non-generalizable examples, which limits their effectiveness in real-world scenarios.\n  To address this gap, we propose VulKey, an LLM-based AVR framework t"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"On the real-world C/C++ dataset PrimeVul, VulKey achieves 31.5% repair accuracy, surpassing the best baseline by 7.6% and outperforming leading tools such as VulMaster and GPT-5. Moreover, VulKey demonstrates cross-language and cross-model generalizability, with state-of-the-art performance on the Java benchmark Vul4J.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the three-level hierarchical abstraction (CWE type, syntactic actions, semantic key elements) can be reliably matched to unseen vulnerabilities and will provide sufficient guidance for the fine-tuned LLM to produce correct, generalizable patches without overfitting to the abstracted patterns.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"VulKey reaches 31.5% repair accuracy on real C/C++ vulnerabilities by matching hierarchical expert patterns to guide LLM patch generation, beating prior baselines by 7.6%.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A framework guides LLMs using a three-level hierarchy of security knowledge to generate more accurate vulnerability patches.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b72f16bed5a12d041e59797591786c8fce28667245d0ded4d2250233df852676"},"source":{"id":"2605.01769","kind":"arxiv","version":3},"verdict":{"id":"c372abf6-1f79-47dd-97f7-4c34cf691e96","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T14:53:42.091341Z","strongest_claim":"On the real-world C/C++ dataset PrimeVul, VulKey achieves 31.5% repair accuracy, surpassing the best baseline by 7.6% and outperforming leading tools such as VulMaster and GPT-5. Moreover, VulKey demonstrates cross-language and cross-model generalizability, with state-of-the-art performance on the Java benchmark Vul4J.","one_line_summary":"VulKey reaches 31.5% repair accuracy on real C/C++ vulnerabilities by matching hierarchical expert patterns to guide LLM patch generation, beating prior baselines by 7.6%.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the three-level hierarchical abstraction (CWE type, syntactic actions, semantic key elements) can be reliably matched to unseen vulnerabilities and will provide sufficient guidance for the fine-tuned LLM to produce correct, generalizable patches without overfitting to the abstracted patterns.","pith_extraction_headline":"A framework guides LLMs using a three-level hierarchy of security knowledge to generate more accurate vulnerability patches."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.01769/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-20T17:36:55.237408Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-20T05:01:22.844908Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T16:59:22.080843Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"cf6cfd576a54d472bbb710a8dcc77feb22648925982b0b84d2acae1161607bb8"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}