code-reviewer

Run a full code review on a PR or local branch — auto-detects tech stack and publishes findings to Azure DevOps

Gather full work item hierarchy context for a PR — walks ADO links from Task/Bug up to Epic level

Senior software architect agent performing comprehensive architectural design review. Detects structural issues that compound over time: layer boundary violations (Clean Architecture/Hexagonal), full SOLID principle violations (SRP god classes, DIP violations, LSP breaks, fat ISP interfaces, OCP type-dispatch chains), anemic domain models, component coupling patterns (circular, stamp, temporal, shared mutable state), DI anti-patterns (service locator, captive dependencies), bounded context breaches, composition-over-inheritance violations, Orleans-specific architectural issues, cross-cutting concern misplacement, and scalability architecture anti-patterns. Dispatch when a PR introduces new services, classes, or interfaces; changes cross-cutting architecture; adds new dependency relationships between layers; modifies DI registrations; or when the PR description mentions architecture, refactoring, or structural changes. Focuses on long-term structural health — not code style, exception handling, or duplication (covered by other agents). <example> Context: A PR adds a new service that directly references a data access layer from a controller user: "Review PR #5678 for architectural issues" assistant: "I'll dispatch the architecture-review agent to check for layer violations, dependency direction issues, and coupling problems in the new service." <commentary> The PR introduces cross-layer dependencies that need architectural review beyond class-level design. </commentary> </example> <example> Context: A PR creates three new projects and moves code between layers user: "Run a full review on PR #9012" assistant: "I'll dispatch architecture-review alongside other agents since the PR restructures project boundaries and could introduce architectural violations." <commentary> Structural changes to project layout require checking dependency direction, bounded contexts, and layer discipline. </commentary> </example>

Identifies unnecessary complexity in class, method, component, and layer design. Use when reviewing PRs for over-engineering, over-abstracted hierarchies, unnecessary layers, premature generalization, or excessive component complexity.

Finds code blocks and method chains that are more complex than they need to be. Use when reviewing PRs for overly complex control flow, verbose code patterns, unnecessary method chains, or expressions that have simpler equivalents.

Finds duplicate and near-duplicate code patterns including copy-pasted blocks, repeated logic with minor variations, and structural duplication. Use when reviewing PRs for code that should be extracted into shared abstractions.

Detects End User Identifiable Information (EUII) leaks in log statements, telemetry, error messages, and other output channels.

Shared search discipline rules for all reviewer agents. Prevents false positive and over-claimed findings by enforcing source-branch search, scoped queries, PR diff checks, build status evidence, and claim-strength matching. Invoke before making any claim about what exists, doesn't exist, or is "only/all/always" true in a codebase.

NScript client code review — AutoFire/nameof enforcement, Promise patterns, IoC registration, C# restrictions, MVVM patterns, template bindings, LESS conventions, and JS interop attributes. Load reference files on demand.

Orleans PR review skill — delegates to orleans-dev:orleans-patterns for domain knowledge, then loads review-specific bridge content (output format, PR edge cases, severity mapping).

Code-review methodology for detecting over-engineering, gold-plating, and scope creep — places where the implementation does meaningfully more than the linked task, bug, or PR title actually required. Compares delivered diff to stated intent, not raw complexity in isolation. Particularly catches what AI- and LLM-generated PRs tend to produce: speculative abstractions for hypothetical futures, single-use helper extractions, defensive code for impossible scenarios, drive-by refactors bundled with bug fixes, premature optimization without measurement, unused configuration hooks, excessive logging, tutorial-style comments, unrequested features, and duplicate code paths added next to existing ones. Use whenever a reviewer says or thinks the diff feels too big for what was asked — e.g., "this bug fix introduces a new interface for one impl", "the AI added five helpers used once each", "fix login bug PR has 14 files", "is this YAGNI?", or any audit of AI-generated code against an original prompt. NOT for general code quality, performance, security, exception handling, duplicate-finding, EUII, architecture, test generation, logging setup, or implementing work items — those have dedicated tools.

Publish PR review results to Azure DevOps — posts findings as inline/file/general comments, posts context questions as inline `[QUESTION]` comments, manages the review summary thread (replies to existing thread instead of creating duplicates), and optionally approves or merges the PR. Use this skill when the pr-review workflow reaches Step 12 and needs to publish its collected findings, questions, and verdict to ADO. Also usable standalone for any workflow that needs structured comment posting to a PR. Trigger when: "post review results", "publish findings to PR", or when invoked from pr-review Step 12. NOT for analysis or grading — this skill only handles the publishing/posting workflow.