By ByteBot
JetBrains shipped Rider 2026.2.1 yesterday, and the headline item is a bundled skill that hands AI agents direct access to Rider’s refactoring engine. Median task time falls from 157.9 seconds to 26.6 seconds. Cost per task drops from $0.52 to $0.19. Build invocations across a 15-task evaluation collapsed from 163 to 3. If you’re running Claude Code or Codex against a .NET codebase, this patch is worth picking up today.
Why AI Agents Were Bad at Refactoring
Refactoring sounds like something AI coding agents should handle easily — rename a class, extract a method, move a type. The problem is that agents do it the hard way. Without a semantic model of your code, an agent has to read a file, guess what to change, make a text edit, run a build, catch what broke, and repeat. JetBrains measured 17 tool calls and 163 build invocations across their evaluation before the skill existed. That’s not a workflow — that’s thrashing.
The deeper issue: structural code changes require understanding that a renamed symbol and a nameof() reference to it are the same thing, that a signature change has a call site three directories away, that a XAML binding will break if you move a type. Language models don’t natively maintain that kind of resolved reference graph. IDEs do.
What the refactoring-code Skill Does
The refactoring-code skill bridges the gap by giving agents access to the same engine Rider uses for its own inspections and navigation. When you ask Claude Code to extract an interface from a class, the agent calls the skill instead of editing text. Rider resolves the target through its reference index and rewrites every affected site — nameof() references, XML documentation, XAML bindings, Razor views, and C++/Unreal cross-language references — as a single semantic operation.
Eight refactoring types are supported: symbol rename, method extraction, interface derivation, base class creation, API signature modification across call sites, type relocation with namespace repair, namespace reorganization, and safe symbol deletion. Coverage extends across the entire solution, not just the file the agent happens to have open.
The most dramatic improvement in JetBrains’ own testing was base class extraction: a task that previously took 337 seconds and 24 tool calls now takes 20 seconds and 3 calls. That’s not an optimization — it’s a different operation entirely.
The Numbers
| Metric | Before Skill | After Skill | Change |
|---|---|---|---|
| Median task time | 157.9 seconds | 26.6 seconds | 83% faster |
| Cost per solved task | $0.52 | $0.19 | 64% cheaper |
| Tool calls per task | 17.0 | 6.2 | 63% fewer |
| Build invocations (total eval) | 163 | 3 | 98% fewer |
The cost number deserves attention. At current usage patterns, agentic refactoring tasks compound quickly — a team running agents continuously sees those per-task costs accumulate into real monthly spend. A 64% reduction per task isn’t negligible.
How to Enable It
There’s nothing to configure. The skill ships bundled with Rider 2026.2.1 and activates automatically when an agent is asked to perform a supported refactoring. Update the IDE, and Claude Code or Codex picks it up on the next refactoring request.
One practical tip from JetBrains: be specific in your prompts. “Rename UserService to AccountService across the solution” gives the skill a clear target. “Clean up this class” does not. The more precisely you describe the refactoring, the more directly the skill can route the request to the right engine operation.
The Rest of the Patch
Rider 2026.2.1 also ships a debugging-code skill, which lets AI agents investigate runtime issues rather than guessing from static code. Agents can set breakpoints, step through execution, and inspect actual values — across C#, F#, C++, Unity, and Unreal Engine projects. On the ReSharper side, out-of-process mode is now the default, cutting Visual Studio UI freezes from 26 seconds to 10.1 seconds on large solutions; this was held back until dotCover gained OOP support, and that dependency is now resolved. Junie gains Claude Opus 5 as a model option, and quality-check hooks now cover Codex alongside the existing Claude Code support.
Who Should Update Now
If you’re using Claude Code or Codex for any .NET development — refactoring legacy code, modernizing APIs, restructuring namespaces — upgrade to 2026.2.1 immediately. The refactoring-code skill is bundled, costs nothing extra, and the improvement is large enough that waiting doesn’t make sense. Teams that have been avoiding agentic refactoring because of the time and cost overhead now have a concrete reason to revisit that call.
If you’re a ReSharper user who hasn’t touched AI agents yet, the OOP-default change alone is worth the update for the reduction in Visual Studio freezes. The full release notes cover the bug fix list. The dedicated skill post goes deeper on the refactoring benchmarks if you want the technical breakdown.
