{"slug": "rider-2026-2-ai-agent-skills-that-know-your-codebase", "title": "Rider 2026.2: AI Agent Skills That Know Your Codebase", "summary": "JetBrains released Rider 2026.2 on July 22, introducing Agent Skills that expose IDE analysis capabilities—profiler data, coverage maps, refactoring intelligence, and framework knowledge—directly to external AI coding agents. Three skills ship in this release: dottrace-analyze, which improved agent diagnosis accuracy from 4.71 to 8.15 out of 10 across 80 test scenarios; finding-tests, which reduces token consumption by up to 50% per test-writing session; and code quality check hooks that validate every AI agent change before the agent can continue. The update also includes performance improvements such as 2.8 seconds faster debugger launch on Windows, 2–3× faster branch switching, and 7–8% lower backend memory usage, plus native GitHub Copilot integration and third-party AI completion providers.", "body_md": "AI coding agents have a context problem. Every time Claude Code, Codex, or Copilot touches your .NET codebase, it starts from scratch — scanning files, guessing at test locations, inferring performance hotspots from code structure rather than actual runtime data. The IDE already knows all of this. The agent just can’t see it. [Rider 2026.2](https://blog.jetbrains.com/dotnet/2026/07/22/rider-2026-2-release/), released July 22, changes that.\n\n## Agent Skills: The IDE Opens Its Intelligence Layer\n\nThe headline feature in this release is not an AI chat panel or a smarter autocomplete. It’s something more structural: Rider now exposes its own analysis capabilities — profiler data, coverage maps, refactoring intelligence, framework knowledge — directly to external coding agents. JetBrains calls these “Agent Skills,” and they’re bundled with the IDE rather than living in a plugin marketplace.\n\nThree skills ship in 2026.2 and each solves a real pain point.\n\n**dottrace-analyze** gives agents access to dotTrace performance snapshots. Instead of guessing which method is the bottleneck by reading code, an agent can now read the actual call tree from a profiling session. JetBrains ran 80 test scenarios comparing agent diagnoses with and without the skill. Accuracy went from 4.71 to 8.15 out of 10. In one Avalonia UI test, the agent scored 1.6/10 cold — and hit 10/10 across all ten runs once it could read the snapshot. That’s the difference between a plausible guess and a correct answer.\n\n**finding-tests** uses dotCover coverage data to tell an agent exactly which test file to write into. Without this, agents wander through project directories. With it, they get the precise file path, read surrounding tests for naming conventions, and write something consistent with the rest of the codebase. JetBrains reports up to 50% fewer tokens consumed per test-writing session — less wandering, less cost, better output. The [full finding-tests writeup](https://blog.jetbrains.com/dotnet/2026/05/22/claude-codex-ai-agent-skill-for-writing-tests/) is worth reading if you run AI agents against large test suites.\n\n**Code quality check hooks** put Rider between the agent and the codebase. Every change an AI agent makes gets validated before the agent can continue. Errors block task completion outright. Warnings return as structured feedback the agent can act on. You set the quality bar; the IDE enforces it. Claude Code is supported now, with Codex coming. This is the closest thing .NET has to an automated review gate that runs before the agent pushes changes — not after. More detail in the [code quality hooks announcement](https://blog.jetbrains.com/dotnet/2026/06/08/rider-2026-2-code-quality-check-hooks-for-ai-agents/).\n\n## Performance: The Numbers That Affect Every Session\n\nThe AI agent story is the headline, but Rider 2026.2 also ships improvements that every .NET developer will notice on day one.\n\n| Metric | Improvement |\n|---|---|\n| Debugger launch (Windows, .NET apps) | ~2.8 seconds faster |\n| Branch switching (Roslyn solutions) | 2–3× faster |\n| Unreal Engine C++ indexing | ~2× faster vs 2026.1 |\n| Backend memory usage | 7–8% lower |\n\nFor anyone who has waited on Rider to finish indexing a large solution before anything else is usable, the branch switching improvement alone is worth the update.\n\n## GitHub Copilot Goes Native, Third-Party Providers Added\n\nCopilot is now a natively integrated agent in Rider — OAuth sign-in, no registry editing required. JetBrains also added third-party AI completion providers: LM Studio, llama.cpp, and Mercury are configurable as local completion backends. TypeScript 7 support arrives with a noticeable impact: Kibana-scale TypeScript projects load in roughly 3 seconds versus 12 seconds previously.\n\n## Game Developers: Godot Tooling Gets Serious\n\nRider 2026.2 is the first release where Godot support feels genuinely professional. Natvis debugging support now works on Linux and macOS (previously Windows-only). A configurable GDScript formatter ships with the IDE. JetBrains published an official Rider Integration addon to the Godot Asset Store. Scene nodes can be dragged directly into code as paths or variable declarations.\n\nOn the Unreal Engine side, UInterface navigation and Gameplay Tag usage tracking work across C++ and Blueprints simultaneously. ISPC receives first-class editor support. The [dottrace-analyze deep dive](https://blog.jetbrains.com/dotnet/2026/06/25/performance-profiling-agent-skill-in-rider/) covers how agents interact with profiling data in the context of game workloads as well.\n\n## How to Update\n\nRider 2026.2 is available now through JetBrains Toolbox App or from the release page. Agent Skills require AI Assistant enabled in Rider settings. The dottrace-analyze skill needs an existing dotTrace snapshot — profiling sessions captured in Rider or standalone dotTrace both work. Quality check hooks for Claude Code activate automatically once Claude Code is configured as your agent.\n\nThe pattern JetBrains is establishing across its 2026.2 cycle — IDEs that expose structured intelligence to external agents rather than treating agents as file-browsing scripts — is the right direction. Rider 2026.2 makes the case with actual numbers, not architecture diagrams.", "url": "https://wpnews.pro/news/rider-2026-2-ai-agent-skills-that-know-your-codebase", "canonical_source": "https://byteiota.com/rider-2026-2-ai-agent-skills-that-know-your-codebase/", "published_at": "2026-07-29 07:12:22+00:00", "updated_at": "2026-07-29 07:23:57.918076+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-tools", "developer-tools", "ai-agents", "large-language-models"], "entities": ["JetBrains", "Rider 2026.2", "Claude Code", "Codex", "GitHub Copilot", "dotTrace", "dotCover", "Godot"], "alternates": {"html": "https://wpnews.pro/news/rider-2026-2-ai-agent-skills-that-know-your-codebase", "markdown": "https://wpnews.pro/news/rider-2026-2-ai-agent-skills-that-know-your-codebase.md", "text": "https://wpnews.pro/news/rider-2026-2-ai-agent-skills-that-know-your-codebase.txt", "jsonld": "https://wpnews.pro/news/rider-2026-2-ai-agent-skills-that-know-your-codebase.jsonld"}}