Abstract
Automated bug repair agents may operate in isolated sessions and can lack a memory of past engagements, which may lead to repeating failed strategies and inefficient workflows. A self-improving framework can use a reflective knowledge feedback loop to address potential inefficiencies. Following a bug-fixing attempt, for example, a reflection phase can analyze the session to extract structured, reusable knowledge, such as dead ends, learnings, and fix patterns. This knowledge may then be categorized and stored in a persistent, indexed repository for retrieval during subsequent tasks. This process can facilitate learning from experience across sessions, which may reduce the repetition of past mistakes and improve bug-fixing efficiency and effectiveness over time.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Xia, Huang; Liu, Ting; Zhang, Yulong; Yan, Zhixian; Huang, Ke; Zhou, Paul; and Yu, Jialin, "The Cross-Session Learning in Automated Bug Repair via a Reflective Knowledge Feedback Loop", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/10276