{"schema":"hottub-news/story/v1","story":"st_b2okhrownj4bggpxemcq","title":"GitHub讓AI代理接手模糊測試，可依覆蓋率自動補測未觸及程式碼","title_en":"GitHub allows AI agents to take over fuzzy testing, automatically supplementing untested code based on coverage rate","outlets":1,"brief":null,"page":"https://hottub.news/stories/github-allows-ai-agents-to-take-over-fuzzy-testing-automatically-b2okhrownj","count":1,"items":[{"id":"n_b2okhrownj4bggpxemcq","seq":591804,"kind":"article","title":"GitHub讓AI代理接手模糊測試，可依覆蓋率自動補測未觸及程式碼","title_en":"GitHub allows AI agents to take over fuzzy testing, automatically supplementing untested code based on coverage rate","url":"https://www.ithome.com.tw/news/179245","summary":"新聞 GitHub Security Lab 開源 Fuzzing Taskflow，可將大型語言模型代理與AFL++模糊測試工具串成自動化管線。開發者指定C/C++專案的GitHub儲存庫後，代理會找出適合測試的函式、分析建置方式、產生模糊測試程式並執行AFL++，再依程式覆蓋率修改測試程式或測試輸入，程式崩潰分析與漏洞報告也納入流程。GitHub提醒，漏洞判斷與修補內容仍須人工審查。 Fuzzing Taskflow建立在GitHub Security Lab Taskflow Agent框架之上，大型語言模型負責決定測試目標、測試程式內容與未覆蓋程式碼的處理方式，MCP工具負責編譯、執行AFL++、收集覆蓋率及保存崩潰資訊，流程狀態存入SQLite資料庫。…","published":"2026-09-29T04:21:47Z","seen":"2026-09-29T06:09:59Z","lang":"zh","country":"TW","source":{"id":"kite-ithome-com-tw-c23d09","name":"ithome.com.tw","domain":"ithome.com.tw","outlet":{"wikidata":"Q123398745","name":"iThome","type":"magazine","country":"TW","designations":["magazine","news website"]}},"via":"kite-ithome-com-tw-c23d09","authors":["李建興"],"entities":[{"id":"Q364","name":"GitHub","type":"org"}],"story":"st_b2okhrownj4bggpxemcq","category":"science-technology","category_p":0.9900000095367432,"sentiment":"neutral","tone":-0.019999999552965164,"political":0.03999999910593033}]}