Content
7%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This skill is essentially a high-level role description rather than actionable guidance. It tells Claude to be a 'master software architect' but provides no concrete review frameworks, checklists, example outputs, or specific evaluation criteria. The content reads more like a job posting than an instruction set, and nearly everything stated is knowledge Claude already possesses.
Suggestions
Replace the abstract 4-step workflow with a concrete architectural review checklist including specific questions to evaluate (e.g., 'Does the design have a single point of failure?', 'Are bounded contexts properly defined?') with example outputs.
Add at least one concrete example showing an input (e.g., a system design description) and the expected architectural review output format with specific findings, risk ratings, and recommendations.
Remove the 'Expert Purpose' section entirely—it restates the description and adds no actionable information. Use that space for a concrete review template or decision record format (e.g., an ADR template).
Add validation steps to the workflow, such as 'Verify the design against these quality attributes: [list]' and 'Confirm the design handles these failure modes: [list]' to make the process rigorous and repeatable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The content is verbose and explains concepts Claude already knows well (SOLID principles, microservices, DDD, clean architecture). The 'Expert Purpose' section largely restates the description. Phrases like 'Elite software architect' and 'Masters modern architecture patterns' are filler that add no actionable value. The 'Use this skill when' and 'Do not use this skill when' sections describe obvious scenarios. | 1 / 3 |
Actionability | The instructions are entirely abstract and vague: 'Gather system context,' 'Evaluate architecture decisions,' 'Recommend improvements.' There are no concrete examples, no specific commands, no code snippets, no templates for architectural reviews, no checklists, and no example inputs/outputs. Claude would not know what to do differently after reading this. | 1 / 3 |
Workflow Clarity | The four-step workflow ('Gather context, Evaluate, Recommend, Document') is extremely high-level with no validation checkpoints, no feedback loops, no concrete criteria for evaluation, and no guidance on what 'validation plans' look like. For architectural reviews involving potentially high-risk decisions, the lack of any concrete verification steps is a significant gap. | 1 / 3 |
Progressive Disclosure | The skill does reference 10 sub-skill files with clear one-level-deep links, which is good structure. However, no bundle files were provided, so we cannot verify these references resolve to actual content. The main SKILL.md itself is thin on substance, making the references feel like they're deferring all value to files that may not exist. The section headers ('Capabilities', 'Knowledge Modules') add structure but the overview content is too vague to serve as a useful entry point. | 2 / 3 |
Total | 5 / 12 Passed |