aidlc-workflows bugfix: AI audit overturns ticket's 1000-character fix

The case shows how aidlc-workflows' bugfix scope, which runs only 7 of 32 stages and skips market research and architecture design, still forced a pause before implementation. The reverse-engineering stage surfaced that the menu and service limits were already different in the screen (1000 vs 300 characters) and handled by separate APIs, so the ticket's symmetric fix looked doubtful. In requirements-analysis, the AI first cited a deprecated API and used 30 days of zero error logs to argue GBP did not enforce its 250-character recommendation; Mikami pointed out the DB query covered all data regardless of creation date, so the AI withdrew the claim. A full re-investigation of official docs, API call paths, DB, a year of error logs, and the GBP admin screen confirmed service descriptions at 300 and menu descriptions at 1000 characters, and also revealed a 120-character service name limit not in the ticket. Later, a review AI warned that adding OpenAPI limits could block saves for stores with over-limit data, but a DB check found no such cases, so the limits were added. The author notes many of his challenges to the AI mirrored his usual review habits with team members, suggesting AI collaboration extends existing practices rather than replacing them.