Abstract
Code snippets within technical documentation can contain syntax errors and formatting issues, which may require manual effort for detection and correction. A disclosed method can address this by employing an agentic pipeline based on a large language model (LLM) to automate the process. The system can extract and classify code snippets from documentation pages. For an identified snippet, a reasoning model can describe potential errors. A reflection chain may then be initiated, where a first LLM proposes a fix, and a second LLM instance reviews the proposed change, potentially generating a critique if the fix appears incorrect. This iterative fix-and-review cycle can continue until the code is considered corrected. A final, validated fix may then be used to generate a pull request for human review, which can contribute to improved documentation quality and a reduced developer workload.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Kuligin, Leonid and Ostapenko, Alex, "Automated Correction of Documentation Code Snippets Using an Agentic Large Language Model with a Reflection Chain", Technical Disclosure Commons, (October 22, 2025)
https://www.tdcommons.org/dpubs_series/8762