analyzing-dotnet-performance

# .NET Performance Patterns Agent uses this skill to scan C# code for counted performance defects and concrete fixes. ## When to Use | User prompt | Use | |---|---| | User asks for a performance review of .NET code | Agent uses this skill | | User asks to audit hot paths for allocations or inefficient APIs | Agent uses this skill | | User asks for a systematic anti-pattern scan before release | Agent uses this skill | ## When Not to Use | User prompt | Route | |---|---| | User asks for algorithm redesign | Agent uses a design review workflow | | User asks for production tracing or profiling only | Agent uses tracing or profiling skills | | User has no performance requirement and no hot path | Agent skips speculative optimization | ## Required Inputs | Input | Required | Notes | |---|---|---| | Source scope | Yes | Agent scans files, folders, or code blocks. | | Hot-path context | Recommended | Agent raises severity for known hot paths. | | Target framework | Recommended | Agent uses runtime facts for fix viability. | | Scan depth | No | Agent uses `critical-only`, `standard`, or `comprehensive`. | ## Workflow | Step | Agent action | Test | Pass | |---|---|---|---| | 1 | Agent identifies hot paths, target framework, and scan depth. | Read the prompt and scope note. | Scope note states hot-path status, TFM, and scan depth. | | 2 | Agent detects category signals and selects scan recipes. | Run `rg -n "async|await|Task|ValueTask|Span<|Memory<|Regex|Dictionary<|JsonSerializer|HttpClient|Stream" <scope>`. | Agent has a category list with counted hits. | | 3 | Agent runs recipe groups and records hit counts. | Run the selected `rg` commands from the recipe matrix. | Recipe log lists every command and a hit count, including zero. | | 4 | Agent applies inverse counts for ratio findings. | Count both positive and negative patterns, such as sealed and unsealed classes. | Ratio findings report `N of M` values. | | 5 | Agent classifies findings by severity and hot-path context. | Review the finding list. | Each finding has severity, count, files, and a concrete fix. | | 6 | Agent writes a compact summary table. | Read the output footer. | Summary table lists severity counts and the top defect. | ## Recipe Matrix | Category | Agent action | |---|---| | Strings | Agent scans for `.IndexOf(`, `.StartsWith(`, `.EndsWith(`, `.Contains(`, `.ToLower()`, `.ToUpper()`, `.Substring(`, and `.Replace(`. | | Collections and LINQ | Agent scans for `.Where(`, `.Select(`, `.OrderBy(`, `.GroupBy(`, `.Any(`, `.All(`, `new List<`, and `new Dictionary<`. | | Regex | Agent scans for `new Regex(`, `RegexOptions.Compiled`, and `GeneratedRegex`. | | Async | Agent scans for `Task`, `ValueTask`, sync-over-async calls, and hidden allocations in closures. | | I/O and serialization | Agent scans for `JsonSerializer`, `HttpClient`, `Stream`, and file I/O on hot paths. | | Structural | Agent counts `sealed class` and non-sealed class declarations. | ## Severity Matrix | Severity | Agent action | |---|---| | Critical | Agent reports deadlocks, crashes, severe regressions, and security-sensitive performance defects first. | | Moderate | Agent reports counted API and allocation defects that affect hot paths or appear at scale. | | Info | Agent reports lower-impact patterns outside hot paths. | | Scale escalation | Agent raises repeated patterns with 11+ instances and flags 50+ instances as systematic. | ## Output Contract | Field | Agent action | |---|---| | Title | Agent writes a short defect title with instance count. | | Impact | Agent states one-line performance effect. | | Files | Agent lists `File.cs:L42` locations inline. | | Fix | Agent states the concrete code change. | | Caveat | Agent adds a caveat only when runtime or correctness risk matters. | ## Validation Checklist - [ ] Agent logged each scan recipe and hit count. - [ ] Agent reported zero-hit recipes when they add evidence. - [ ] Agent reported `N of M` ratios for inverse patterns. - [ ] Agent attached a concrete fix to each finding. - [ ] Agent grouped findings by severity. - [ ] Agent ended with a summary table. ## Common Pitfalls | Pitfall | Agent fix | |---|---| | Agent flags every `Dictionary` as a `FrozenDictionary` defect | Agent flags only read-only dictionaries. | | Agent suggests `Span<T>` inside async code | Agent uses `Memory<T>` for async paths. | | Agent reports LINQ outside hot paths as a universal defect | Agent ties severity to hot-path scope. | | Agent suggests `ValueTask` everywhere | Agent limits `ValueTask` to measured hot paths with frequent synchronous completion. |