Designer-RSI is a continual-adaptation framework in which a frozen frontier model drives professional design software through 230+ tools while an external, natural-language procedural memory accumulates and revises reusable design skills from its own successful and failed executions, with a replay gate that admits only changes that fix failures without regressing prior successes. Across five rounds on 1,406 real user briefs, the skill bank grows from 76 to 139 procedures and raises execution success on Claude Sonnet 4 from 72.7% to 99.3%, with 61.8-67.6% win rates against a no-skill agent. The paper shows that combining memory widening (new skills) and deepening (revising existing skills) outperforms either mechanism alone.
