self-correction-loop

To stop repeated coding mistakes, convert specific user corrections into generalized, class-level project rules written to MEMORY.md. This deduplicates existing guidance and applies the updated rules proactively at session start to prevent regressions.

To enforce project standards for code style automatically, detect user corrections during code review, generalize the feedback into broad rules, and write them to MEMORY.md by category. This ensures architecture and style expectations are applied consistently.

You prevent Claude from repeating code generation errors by initiating an explicit memory-writing workflow when corrections occur. This captures the fix, generalizes it into a broader rule, and stores it in MEMORY.md to be reviewed at the start of future sessions.

Yes, the self-correction loop deduplicates overlapping rules in MEMORY.md. When capturing a correction, it scans existing project memory, updates overlapping guidance, and writes categorized rules to prevent memory bloat and maintain concise standards.

To apply existing MEMORY.md rules when starting a new coding session, use session initialization to review stored project standards proactively. This ensures past corrections and generalized rules are loaded to prevent regressions immediately.

You should avoid using a self-correction loop for project memory management if your corrections are one-off fixes that do not need to be generalized into class-level rules. It is designed for iteratively reducing error rates over time through deduplicated memory updates.