agentleFS
ResgridResgrid/Core/.claude/skills/code-review-workflow/SKILL.md

code-review-workflow

Skill229 starsChanged 3 months agoApache-2.0View on GitHub

Structured code review workflow for .NET projects using Roslyn MCP tools. Multi-dimensional review covering correctness, security, performance, architecture compliance, and test coverage. Load when: "review PR", "review code", "code review", "PR review", "review changes", "review my code", "check code quality".

We split this skill into the rules it gives an agent

10 rules. 0 are common, 6 are distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 6

    Distinctive

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use for non-trivial PRs (3+ files changed, new features, refactors).

    line 25
  2. Distinctive:

    Run MCP tools on changed files:

    line 34
  3. Distinctive:

    Read changed files for things tools can't catch:

    line 68
  4. Distinctive:

    Use for small changes (1-2 files, bug fixes, config changes).

    line 107
  5. Very specific:

    Run detect_antipatterns on changed files

    line 110
  6. Distinctive:

    Read the changed code for correctness

    line 112
  7. Very specific:

    Run get_project_graph — visualize the full dependency tree

    line 131
  8. Distinctive:

    Verify dependency rules per architecture:

    line 132
  9. Very specific:

    Run find_references on module/layer boundary types to verify encapsulation:

    line 141
  10. Very specific:

    Run detect_circular_dependencies to find cycles:

    line 148

“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Code Review Workflow

Core Principles

  1. MCP-first analysis — Use Roslyn MCP tools before reading source files. detect_antipatterns catches more than manual scanning, get_diagnostics finds what the compiler knows, and find_references reveals blast radius. Only read files for context that tools can't provide.
  2. Structured output — Every review follows the same format: Summary → Critical → Warnings → Suggestions → Architecture Compliance → Test Coverage → What's Good. Consistent structure makes reviews actionable and scannable.
  3. Severity-based findings — Categorize every finding as Critical (must fix before merge), Warning (should fix, creates tech debt), or Suggestion (nice to have). Never mix severities — a cosmetic issue next to a security bug buries the important finding.
  4. Actionable suggestions — Every finding includes: what's wrong, why it matters, and how to fix it. "This is bad" is not a review comment. "This creates N+1 queries because X. Fix by adding .Include() or using a projection" is.
  5. Acknowledge good work — Always include a "What's Good" section. Positive reinforcement of good patterns is as important as flagging bad ones.

Patterns

Full PR Review Flow

Use for non-trivial PRs (3+ files changed, new features, refactors). Execute steps in order:

Step 1: Understand the change scope Get changed files from git diff or user input. Categorize:

  • New files (features, tests, configs)
  • Modified files (which layers? domain, application, infrastructure, API?)
  • Deleted files (was anything depending on them?)

Step 2: Automated analysis Run MCP tools on changed files:

→ detect_antipatterns (file: each changed .cs file)
  Catch: async void, sync-over-async, DateTime.Now, new HttpClient(), broad catch, etc.

→ get_diagnostics (scope: file, path: each changed file)
  Catch: new compiler warnings, nullability issues, unused variables

→ get_public_api (typeName: each modified type)
  Check: API surface changes — new public members, removed members, signature changes

Step 3: Blast radius assessment For each changed public API:

→ find_references (symbolName: changedMethod)
  Count callers. High count = high risk. Flag breaking changes.

Step 4: Architecture compliance

→ get_project_graph
  Verify: dependency direction is correct (Domain → nothing, Infra → Domain, Api → Application)
  Flag: circular references, wrong-direction dependencies

Step 5: Test coverage check

→ get_test_coverage_map (projectFilter: changed project)
  Check: do test files exist for every changed type?
  Flag: new types without tests, modified logic without test updates

Step 6: Manual review Read changed files for things tools can't catch:

  • Business logic correctness
  • Naming clarity and consistency
  • Error handling completeness
  • Concurrency safety
  • Security: input validation, authorization checks, data exposure

Step 7: Produce review

## Review Summary
[1-2 sentence overall assessment: scope, risk level, recommendation]

## Critical (must fix)
- **[File:Line] [Title]** — [What's wrong]. [Why it matters]. [How to fix].
- ...

## Warnings (should fix)
- **[File:Line] [Title]** — [What's wrong]. [Impact if not fixed]. [Suggested fix].
- ...

## Suggestions (nice to have)
- **[File:Line] [Title]** — [Current approach]. [Better alternative]. [Why].
- ...

## Architecture Compliance
[Dependency direction check results. Layer violation findings. Module boundary enforcement.]

## Test Coverage
[Which changed types have tests. Which are missing. Specific test scenarios to add.]

## What's Good
- [Positive finding 1 — reinforce good patterns]
- [Positive finding 2]
- ...

Quick Review

Use for small changes (1-2 files, bug fixes, config changes). Lightweight — skip blast radius and architecture checks.

Steps:

  1. Run detect_antipatterns on changed files
  2. Run get_diagnostics on changed files
  3. Read the changed code for correctness
  4. Produce abbreviated review (Summary + Issues + What's Good)
## Quick Review
[1 sentence assessment]

### Issues
- [Finding with severity tag: 🔴 Critical / 🟡 Warning / 🔵 Suggestion]

### What's Good
- [Positive note]

Architecture Compliance Check

Standalone check for architecture-level concerns. Use when reviewing project structure changes, new project additions, or module boundary modifications.

Steps:

  1. Run get_project_graph — visualize the full dependency tree
  2. Verify dependency rules per architecture:
ArchitectureRuleViolation Example
VSAFeatures don't reference each otherFeature A imports from Feature B
Clean ArchitectureDomain has zero project referencesDomain references Infrastructure
DDDAggregates don't reference other aggregatesOrder aggregate imports Product aggregate
Modular MonolithModules communicate only via integration eventsModule A directly references Module B's DbContext
  1. Run find_references on module/layer boundary types to verify encapsulation:
→ find_references(symbolName: "OrdersDbContext")
  Should only be referenced within the Orders module.
  External references = module boundary violation.
  1. Run detect_circular_dependencies to find cycles:
→ detect_circular_dependencies(scope: projects)
  Flag any project-level cycles.

→ detect_circular_dependencies(scope: types, projectFilter: "MyApp.Application")
  Flag type-level cycles within the application layer.

Anti-patterns

Reviewing Without MCP Tools

# BAD — Reading every file manually, missing patterns across the codebase
"Let me read OrderService.cs... looks fine to me."
# Missed: 3 DateTime.Now usages, 1 async void, 2 compiler warnings
# GOOD — MCP-first, then targeted file reads
→ detect_antipatterns: Found 3 DateTime.Now (AP004), 1 async void (AP001)
→ get_diagnostics: 2 CS8600 warnings in OrderService.cs
"I found 6 issues via static analysis. Let me read the files for business logic review..."

Vague Feedback

# BAD
"The code could be better."
"This doesn't look right."
"Consider refactoring this."
# GOOD
"OrderService.cs:47 — `DateTime.Now` should be `TimeProvider.GetUtcNow()`.
DateTime.Now is untestable and uses local timezone. Inject TimeProvider
via primary constructor and call GetUtcNow()."

Missing Security Checks

# BAD — Only checking code style and patterns
"Code looks clean, approved!"
# Missed: SQL injection in raw query, missing authorization attribute, exposed PII in logs
# GOOD — Security is a review dimension
"## Critical
- **OrderController.cs:23** Missing `[Authorize]` — endpoint exposes order data without auth
- **SearchService.cs:45** SQL injection — user input concatenated into raw SQL. Use parameterized query.
## Suggestions
- **LoggingMiddleware.cs:12** PII exposure — email logged at Information level. Mask or use Debug level."

Blocking on Style, Ignoring Substance

# BAD — 10 comments about naming, 0 about the race condition
"Rename `svc` to `service`. Use `var` instead of explicit type. Add XML docs."
# GOOD — Prioritize by impact
"## Critical
- Race condition in OrderService.ProcessAsync — concurrent calls can double-charge
## Suggestions
- Consider renaming `svc` to `service` for clarity"

Decision Guide

ScenarioReview TypeMCP Tools
Feature PR (3+ files)Full PR ReviewAll tools
Bug fix (1-2 files)Quick Reviewdetect_antipatterns, get_diagnostics
Config/infra changesQuick Review + Manualget_project_graph
New project/module addedArchitecture Complianceget_project_graph, detect_circular_dependencies
Refactor PRFull PR Review + ArchitectureAll tools + find_references (blast radius)
Security-sensitive changeFull PR Review → escalate to security-auditordetect_antipatterns + manual security review
Test-only changesQuick Reviewget_diagnostics only
Performance-critical pathFull PR Review → escalate to performance-analystget_diagnostics + manual review

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

Reviews can't be read right now.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.