signal-scoring

I've been exploring signal-scoring implementations, and I'm curious how you're approaching the scoring logic—particularly how you're weighting different signal types to avoid false positives. The skill scores well on utility (68/100), but there's room to tighten up the specification and make the scoring heuristics more transparent.

Links:

• GitHub
• Your agentic skill in the SkillzWave marketplace

TL;DR

You're at 68/100, which puts you in solid D territory. This is based on Anthropic's progressive disclosure architecture standards for agentic skills. Your Writing Style is your strongest pillar (8/10)—the framework is clear and well-organized. The weakest spots are Spec Compliance (11/15) and PDA (18/30), mainly because you're missing trigger phrases in the description and the templates you mention aren't actually provided as reference files.

What's Working Well

• Clear framework structure – Your 5-step approach (Source Inventory → Weighting → Decay Logic → Tier Definition → Governance) is logical and easy to follow. The progression makes sense for someone building a scoring model.
• Strong writing voice – You're using imperative language consistently ("list", "assign", "set") and keeping the tone instructional rather than marketing-y. That's exactly what developers want.
• Practical governance angle – The governance section with "audit tiers," "change log," and "review cadence" addresses a real pain point—most scoring frameworks ignore the human side of maintenance.

The Big One: Missing Template Files

Your biggest friction point is right here: you mention three templates (Scoring worksheet, Tier definition matrix, Change log) but don't actually provide them. This tanks your PDA score and leaves developers hanging at the most critical moment—when they're ready to apply the framework.

The fix: Create three reference files alongside SKILL.md:

• references/scoring-worksheet.md – A markdown table template with columns for Source, Weight, Decay Function, Threshold, Notes
• references/tier-matrix.md – A populated example showing tiers (e.g., "Hot" vs "Warm" vs "Cold") with signal combinations and recommended plays
• references/change-log.md – A template showing how to document changes with Date, Changed Weights, Rationale, Data Evidence, Approvals

This alone gets you +5 points and suddenly developers have actual scaffolding to work from.

Other Things Worth Fixing

1. Add trigger phrases to the description – Right now it's "Use to design composite intent scoring models..." but developers searching for this skill need explicit hooks like "score intent signals," "calculate account scores," or "design scoring model." Add these to the frontmatter description (+2-3 points).

2. Break the framework into numbered sub-steps – Your five phases are solid, but "Source Inventory – list each signal type..." needs to become a mini-checklist (list signals, document freshness, calculate coverage %, assess reliability). This makes it actionable (+3 points).

3. Add a Validation section – Right now there's no guidance on testing your scoring logic. Add: "Test with historical data sample," "Run A/B test on 10% of accounts," "Check edge cases (zero signals, expired signals)." Feedback loops matter (+2 points).

Quick Wins

• Create 3 reference files with actual templates (biggest impact: +5 points)
• Rewrite description with trigger phrases like "score intent signals" or "composite scoring" (+2-3 points)
• Convert framework phases to numbered sub-steps with checkboxes (+3 points)
• Add validation/testing guidance section (+2 points)

Those four changes alone put you at 80+. You're close—just need to bridge the gap between "here's the framework" and "here's how you actually use it."

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.