Current LLM-based coding agents often repeat the same types of mistakes across sessions due to a lack of mechanisms to retain human review feedback. We propose a closed-loop framework where every accepted review comment is encoded as a persistent behavioral rule, progressively expanding the set of error classes the agent can self-detect. This framework combines an accumulating rule set in a version-controlled instruction file, a self-review checklist executed before code submission, and automated validation to ensure rule set integrity as it grows.
Deployed across a microservices platform with over 35 services, the rule set expanded from 5 to 18 behavioral rules, over 15 language-specific standards, and a 15-item self-review checklist, all derived from real review feedback. We present empirical results from 11 recorded working sessions covering code generation, PR review, incident investigation, and cross-service refactoring. Accumulated rules shift review effort from low-level correctness to design-level validation, achieving a measured 0% recurrence rate for ruled-against error classes, and transferring across heterogeneous agent interfaces.
We compare our approach against related work in experiential LLM learning (Reflexion, ExpeL, Voyager) and automated code review (CodeReviewer, SWE-bench agents), showing that our framework achieves persistent cross-session learning without weight updates, operates on production codebases rather than synthetic benchmarks, and addresses an orthogonal dimension (behavioral consistency over time) that existing benchmarks do not measure. The result is a coding agent that improves with every review cycle, accumulating the engineering wisdom of its human collaborators without changing a single model weight.
Blogger's Review: This framework effectively addresses the common issue of error recurrence in LLM coding agents by transforming human review feedback into persistent behavioral rules, showcasing the potential for self-improvement. This approach not only enhances code quality but also sets new standards for future coding agents, emphasizing the importance of human-machine collaboration. By enabling continuous learning in real environments without altering model weights, the agent accumulates engineering wisdom over time.