Quality CI

.NET Quality CI

Trigger On

• adding or tightening .NET code-quality gates in CI
• choosing analyzers, coverage, mutation, or architecture-test tooling for a .NET repo
• standardizing .editorconfig, dotnet format, and warning policy

Value

• produce a concrete project delta: code, docs, config, tests, CI, or review artifact
• reduce ambiguity through explicit planning, verification, and final validation skills
• leave reusable project context so future tasks are faster and safer

Do Not Use For

• non-.NET repositories
• generic CI/CD guidance with no .NET quality stack decisions
• framework-specific test authoring with no quality-gate change

Inputs

• the nearest AGENTS.md
• the current repo-root .editorconfig and MSBuild props
• the current CI workflow and package references
• the active test runner model: VSTest or Microsoft.Testing.Platform

Quick Start

1. Read the nearest AGENTS.md and confirm scope and constraints.
2. Run this skill's Workflow through the Ralph Loop until outcomes are acceptable.
3. Return the Required Result Format with concrete artifacts and verification evidence.

Workflow

1. Start with the repo-native baseline:
   • repo-root .editorconfig
   • dotnet format --verify-no-changes
   • SDK analyzers with explicit EnableNETAnalyzers, AnalysisLevel, and warning policy
2. Add third-party analyzers only where they close a real gap:
   • StyleCopAnalyzers
   • Roslynator
   • Meziantou.Analyzer
   • framework analyzers such as xUnit, MSTest, or TUnit analyzers
3. Separate quality gates by purpose:
   • formatting and style
   • correctness and static analysis
   • coverage and reports
   • architecture rules
   • security scanning
   • mutation testing
4. For complexity, use a composite approach:
   • CA1502 thresholding
   • maintainability limits in AGENTS.md
   • architecture tests
   • coverage and mutation where risk justifies it
5. Make ownership explicit in AGENTS.md and CI:
   • which command formats
   • which command analyzes
   • which command measures coverage
   • which runner model the tests use
6. After any .NET code change, the repo's quality pass must be runnable by agents:
   • format
   • build
   • analyze
   • focused tests
   • broader tests
   • coverage and report generation when configured
   • extra configured gates only when the repo actually enabled them
7. Route tool-specific setup through dedicated skills where possible:
   • format
   • code-analysis
   • analyzer-config
   • analyzer-pack skills such as stylecop-analyzers, roslynator, and meziantou-analyzer
   • frontend asset quality skills in mixed .NET plus Node repos such as eslint, stylelint, htmlhint, webhint, biome, sonarjs, metalint, and chous
   • coverage/reporting skills such as coverlet and reportgenerator
   • architecture/security skills such as netarchtest, archunitnet, and codeql
8. Avoid overlapping tools with conflicting ownership. If you add an opinionated formatter, define whether it replaces or complements dotnet format.

Test Output Budget

• Keep native runner progress and ANSI enabled (--progress on --ansi on for supported MTP/TUnit runners); use a PTY locally. Do not replay progress redraws into model context. Use the detected runner's flags, not MTP switches on VSTest.
• Show warnings and errors plus one final summary (counts, duration, exit code). Configure test-owned console logging at Warning; keep Information/Debug/Trace, successful-test output, and expected negative-test noise out of context. Quiet build verbosity alone does not filter application logs.
• On failure, crash, startup error, or timeout, show the failing test/resource, root exception, and relevant stack frames. Deduplicate; cap each diagnostic tool response at 80 lines / 8 KiB, whichever comes first. Never automatically dump stdout/stderr, host logs, browser console history, DOM/HTML, TRX, or crash artifacts.
• Keep necessary diagnostics in size-bounded or rotating artifacts and link them. Search by exact failure/correlation; read bounded excerpts, never whole logs. Capture/filter noisy output before tool delivery, preserve the real exit code, and disclose truncation. Monitor actual activity; silence alone does not prove a hang.