sj-agent-review

What problem does it solve?

Manually auditing business agent codebases for structural flaws, missing guardrails, and design inconsistencies is time-consuming and prone to human error, often letting critical issues slip through until they cause runtime failures or security vulnerabilities in production. This Skill automates the audit process using a standardized 10-axis design rubric, ensuring no key quality checks are missed before deployment.

Core Features & Use Cases

• 10-Axis Structured Audit: Scores agent projects across 10 core design dimensions (runtime loop, orchestration, role separation, tool hierarchy, context management, guardrails, observability, memory layer, evaluation, graph topology) with clear 0-10 point scales and PASS/WARN/FAIL verdicts.
• Prioritized Issue Reporting: Flags critical, high, and medium priority issues with exact file:line references and concrete fix directions, so teams can address the most impactful problems first.
• Flexible Scope: Supports full project audits or focused reviews of single design axes (e.g., tool hierarchy, guardrails) based on user request, making it useful for both full pre-deployment checks and targeted deep dives.
• Use Cases: Pre-deployment quality gates for new agent projects, structural audits of legacy agent codebases, and code review support for engineering teams building business AI agents.

Quick Start

Use the sj-agent-review skill to audit the agent project in the ./my-business-agent directory against all 10 design axes and receive a prioritized list of improvement actions.