Mechanically linking the cleanup to future work is the way. We did something similar with ArgoCD sync waves - the hotfix deploys immediately, but the "lint-cleanup" app won't sync until a follow-up PR merges. It's a tangible blocker in the UI.
The hard part isn't the tech, it's convincing people that the 'cleanup' is part of the original fix, not a separate chore. If you treat it as optional debt, it'll never get paid.
Yes! This is the key part people miss. It's not just about blocking new features, it's about making the debt visible and blocking *progress* on the specific service or pipeline that was hotfixed. The sync wave tactic is brilliant because it's a concrete, visual reminder in the workflow they use every day.
That visibility changes it from a nagging chore to a real status item. Otherwise it's just another ticket that sinks to the bottom of the backlog.
measure twice, ship once
Oh man, this hits close to home, but from a completely different angle. You're talking about CI/CD pipelines, and I'm over here having flashbacks to CRM data migration scripts where a mandatory formatting gate would have been a disaster.
Your point about a tool making a line *less* readable is spot on. I once had a similar rigidity with a data validation step in a Salesforce-to-HubSpot migration pipeline. The validation rule flagged a perfectly usable, slightly unconventional phone number format in a legacy record. The whole automated flow halted. For a hotfix migration of a single account, we had to bypass the entire process manually because the "mandatory" check couldn't be overridden. The bottleneck felt artificial, just like your CI waiting for a re-format commit.
Maybe the parallel is that any mandatory gate, whether it's for code style or data shape, needs a real, practical emergency lane. Not just a theoretical one, but a documented, low-friction override that doesn't make you feel like you're breaking the law. If the team culture treats that override as failure, you'll just end up with ugly code or, in my world, a spreadsheet of orphaned data.