Oh My Pi has a proactive hard-blocking system that prevents it from running certain models on tasks they can’t handle. But the blocking was too aggressive — it was blocking models that were perfectly capable.
The Problem
The hard-blocking system was blanket-banning models based on broad categories. A model that was great at code generation was being blocked from code tasks because it was in a “creative” category. The guard was doing what it was designed to do — but it was using the wrong criteria.
The Fix
Oh My Pi now exempts specific models from proactive hard-blocking. Instead of blocking by category, it blocks by demonstrated capability. Models that pass certain capability checks are allowed through, even if they belong to a category that would otherwise be blocked.
The change makes the system smarter:
- If a model has proven it can handle code tasks, it’s exempted from code blocking
- If a model has demonstrated reasoning ability, it’s exempted from reasoning blocking
- If a model is new and unproven, it starts with narrower exemptions
Why This Matters for Agent UX
Hard-blocking is a safety feature. But safety features that block legitimate use are worse than no safety features — they frustrate users and train them to find workarounds.
Oh My Pi’s approach — exempt by capability, not category — is the right balance. It keeps the safety guardrails in place while reducing false positives. The agent gets out of your way when it can, and steps in only when it needs to.
This is the kind of UX polish that comes from real-world usage. Oh My Pi has been iterating on this system across multiple versions, and the exemptions approach shows maturity.