Strict Architectural Review for AI Agents

Where this fits

When your main agent reviews a pull request, it often drowns in style nits and test details, missing the real risk: a new file that bypasses the dependency graph or shifts the module layering. You need a reviewer that only cares about architecture-bearing changes. This skill gives you exactly that—a strict, blocking sub-agent that flags violations with zero tolerance, so you can safely automate PR reviews without losing architectural control.

Why agents benefit

• Scoped focus: It ignores routine implementation, performance, and test quality, so the agent spends its entire context on architectural rules.
• Sub-agent isolation: Load it as a sub-agent with a medium-intelligence model, keeping your main context clean and your stronger model for deeper reasoning.
• Git-native scoping: Works with a branch, commit range, or PR, letting you plug it directly into existing CI or agentic review pipelines.
• Enforced pass/fail: Every violation is blocking—no vague suggestions, so the agent produces a clear verdict you can act on mechanically.
• Scope-creep guard: It refuses to expand into legacy cleanup, and after two rejected passes it pushes the caller to ask the user for direction.

Practical scenarios

PR gate for a monorepo

Before merging a feature branch, your agent invokes this skill to review the diff against the repo's architecture rules. If a developer moved a class from core/ to adapters/, the sub-agent reports a blocking violation, preventing a subtle dependency inversion.

Commit-range audit

During a weekly refactor, you ask your agent to review the last 50 commits. The skill isolates only the architecture-changing commits, skipping the 40 that touch tests or docs, and flags the three that alter the plugin interface.

Onboarding new contributors

A new dev adds a file that imports from an internal package not listed in the allowlist. The sub-agent catches it automatically, giving immediate, rule-based feedback instead of waiting for a senior human reviewer.

Add it to your agent workflow

Install it with:

npx skills add sesoi-ai/sesoi_apps_monorepo --skill architecture-implementation-review

Then invoke it as a sub-agent in your prompt:

You are a sub-agent. Use the architecture-implementation-review skill on branch feature/xyz. Call the tool and return only the verdict (pass or fail) plus the list of violations.

Run that sub-agent with a medium-intelligence model to balance speed and accuracy.

Compare before adopting

Assess the skill's quality by reading its rule set—does it match your monorepo's actual architectural constraints? Check maintenance freshness on its GitHub repository and whether issues are addressed. Compare against generic review tools that may be too broad or too lax, and verify it fits your workflow—it prefers Git scopes, so file-based spot checks are second-class. For more context, see the skill’s full listing.

Why it is worth tracking

With 105 stars and a clear, opinionated scope, this skill fills a real gap: most agentic review tools try to cover everything and fail at enforcing architecture. Its strict pass/fail behavior and sub-agent pattern make it easy to integrate, and its novelty means early adopters can shape its rule set through community feedback. Watch for updates—if the rule language grows to support custom architecture policies, it becomes a strong candidate for your production pipeline.