implementation-review

I took a look at your implementation-review skill and noticed you're using a confidence-based filtering approach to surface high-priority issues—that's a smart pattern for code review at scale, especially when dealing with noisy linters. The 84 score reflects solid fundamentals, but there are a few architectural decisions worth reconsidering if you want to push this toward excellence.

Links:

• GitHub
• Your agentic skill in the SkillzWave marketplace

The TL;DR

You're at 84/100, squarely in B-grade territory. This is based on Anthropic's skill best practices across five pillars. Your Utility scores strong at 18/20—the three-mode precision approach (quick/standard/thorough) gives real flexibility—but Progressive Disclosure Architecture (24/30) and Spec Compliance (11/15) are where you're leaving points on the table.

What's Working Well

• Solid utility design: Your three-mode system is actually elegant. Quick/standard/thorough gives developers real flexibility without overengineering.
• Strong feedback loop: The consecutive-pass validation approach (run once, check, fix, run again) is exactly what you'd want for implementation review—it catches the "but wait, did I actually fix it?" problem.
• Clear trigger phrases: "implementation review", "run implementation review"—easy for users to discover and invoke.
• Project-type awareness: Breaking down checks by Python, TypeScript, and Go shows you thought about real-world usage.

The Big One: Single-File Architecture Bottleneck

Your entire 407-line skill lives in one SKILL.md file. This hurts both token economy and discoverability. The mode explanations appear twice—once in a table (lines 27-33), again in prose (lines 43-59)—which is redundant when you're trying to keep token count tight.

Fix this: Extract to a references/ directory:

• references/checklist-templates.md — Move detailed project-type checks here
• references/mode-guide.md — Deep-dive on mode selection logic
• references/overlooked-issues.md — The "what devs forget" section

Keep SKILL.md lean (150-200 lines max) with just the essential workflow. This gains you +3 points on PDA and actually makes the skill faster to load.

Other Things Worth Fixing

1. Add trigger phrases to the description: Right now it's "Self-feedback review loop to verify..." —missing the activation keywords. Change to: "Performs implementation review operations. Use when asked to 'implementation review', 'run implementation review', or 'implementation review help'." That's +2 points on spec compliance.

2. Switch to imperative voice: You've got "If you (Claude) reviewed it..." and "You remember WHY you wrote it..." Replace with "When the implementing agent reviews its own work" and "The agent remembers implementation rationale..." Removes the second-person assumption and is cleaner. +1 point on writing style.

3. Pseudocode clarity: Lines 134-157 mix comments, pseudocode, and prose. Pick one: either pure numbered steps or pure pseudocode, not a blend. Consider adding a simple flowchart in references/ if the logic is genuinely complex.

4. Missing workflow example: You show JSON templates and partial snippets, but no end-to-end "here's a real spec input → here's what the subagent outputs → here's the fix → final report" example. Add examples/full-review-cycle.md with a complete walkthrough. +1 point on utility.

Quick Wins

• ✅ Add 4-5 trigger phrases to frontmatter description
• ✅ Create references/ directory and split out mode guide + checklists
• ✅ Remove redundant mode explanation prose (keep table only)
• ✅ Add one complete workflow example
• ✅ Switch second-person references to agent-centric language

Those changes get you to ~91/100 (A-grade territory) with maybe 30 minutes of restructuring.

Checkout your skill here: SkillzWave.ai | SpillWave We have an agentic skill installer that install skills in 14+ coding agent platforms. Check out this guide on how to improve your agentic skills.