fraud-detection

I was curious about how you'd approach fraud detection in Claude workflows—turns out there's some solid reasoning here, but the architecture could use a bit more breathing room for edge cases and I've got some thoughts on the scoring.

Links:

• GitHub
• Your agentic skill in the SkillzWave marketplace

The TL;DR

You're at 67/100, which lands you in D territory. This is based on Anthropic's PDA framework and the 5-pillar rubric. Your writing style is genuinely strong (8/10)—clear and technical—but Progressive Disclosure Architecture is dragging you down (16/30). You're also missing some workflow specificity that would make this immediately actionable.

What's Working Well

• Writing clarity: Your tone is consistently professional and instructional. No marketing fluff, just technical guidance.
• Solid problem framing: The 5-step framework (Signal Collection → Scoring → Decision Making → Response → Feedback Loop) actually maps to real fraud detection work.
• Appropriate scope: You're not over-promising. The "When to Use" section correctly identifies referral program fraud scenarios.

The Big One

You're missing the Progressive Disclosure structure entirely. Right now everything's in a single 31-line file—there's no room for templates, examples, or runbooks. You mention "Fraud monitoring dashboard outline," "Investigation log template," and "Policy update checklist" but provide zero actual templates in a references/ directory.

This is costing you 6 points. Create:

• references/fraud-monitoring-dashboard.md — actual dashboard layout with metrics (e.g., "Daily signups by IP", "Referral velocity per user")
• references/investigation-runbook.md — step-by-step investigation procedure
• references/policy-checklist.md ...