Evaluate Agent Skills against agentskills.io specification with three progressive modes and smart visual reports. (1) Static Analysis - SKILL.md-only review for quality and spec compliance, outputs score /60. (2) Semi-Static Analysis - adds environment and user fit assessment without execution, outputs score /100. (3) Full Analysis - complete evaluation with security scanning, trigger testing, and dynamic verification, outputs score /130. Supports Bento-ready JSON report output for visual dashboards with auto-scaling blocks based on issue severity. Trigger phrases include "evaluate skill", "review skill", "audit skill", "is this skill good", "should I use/install this skill", "skill quality check", "rate this skill", "score this skill", "bento report", "visual report", "技能评估", "评测 skill", "审核 skill", "可视化报告". Use when this capability is needed.
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Um comando direto ignora o prompt de revisão. Verifique a origem antes de executá-lo.
Instruções da origem · Visualização somente leitura
name
skill-evaluator
description
Evaluate Agent Skills against agentskills.io specification with three progressive modes and smart visual reports. (1) Static Analysis - SKILL.md-only review for quality and spec compliance, outputs score /60. (2) Semi-Static Analysis - adds environment and user fit assessment without execution, outputs score /100. (3) Full Analysis - complete evaluation with security scanning, trigger testing, and dynamic verification, outputs score /130. Supports Bento-ready JSON report output for visual dashboards with auto-scaling blocks based on issue severity. Trigger phrases include "evaluate skill", "review skill", "audit skill", "is this skill good", "should I use/install this skill", "skill quality check", "rate this skill", "score this skill", "bento report", "visual report", "技能评估", "评测 skill", "审核 skill", "可视化报告". Use when this capability is needed.
metadata
{"author":"alterxyz"}
Skill Evaluator
Evaluate Agent Skills against agentskills.io specification and best practices.
Core Philosophy
Good Skill = Expert-only Knowledge − What the LLM Already Knows
Knowledge Type
Treatment
Example
Expert (LLM doesn't know)
Keep — this is value
"mediabox not cropbox for PDF size"
Activation (LLM may forget)
Keep if brief
"Always validate XML before packing"
Redundant (LLM knows)
Delete — token waste
"What is a PDF file"
Before Evaluating, Ask Yourself
"Does this skill capture knowledge that took someone years to learn?" — If no, low D1 score.
"Would an expert read this and nod, or roll their eyes?" — Eye-roll = redundant content.
"After reading, can the LLM do something it couldn't before?" — If just faster/reminded, marginal value.
Mode Selection
Mode
When to Use
Time
Input
Output
Static
Quick check, bulk screening, first pass
~2 min
SKILL.md only
/60
Semi-Static
Install decision, fit check
~5 min
+ Environment/User info
/100
Full
Production deploy, security audit
~15 min
+ Complete package
/130
Default workflow: Static first → If score >60 AND installation considered → Semi-Static → If deploying to production → Full.
Look for: Diagnostic questions, priority rules, "The expert's first question is always..."
D3: Anti-Patterns (10 pts)
Score
Indicator
9-10
Comprehensive NEVER list with non-obvious reasons
6-8
Specific warnings, some reasons
3-5
Vague warnings ("Be careful with...")
0-2
No anti-patterns mentioned
What counts: Specific, actionable, with surprising consequences. "NEVER use Inter font — dead giveaway of AI-generated" beats "Choose fonts carefully."
Any HIGH severity finding = instant fail. Do not proceed.
For semantic analysis (obfuscation, prompt injection, data flow): Read {baseDir}/references/security-scan-llm.md and use the LLM scan prompt.
Step 3: Trigger Testing (10 pts)
Test 5 prompt types:
Type
Example
Expected
Direct
"Use [skill-name] to..."
Trigger
Keyword
"[feature word] my file"
Trigger
Indirect
"[Problem the skill solves]"
Trigger
Ambiguous
Vague related request
Maybe trigger
Negative
Unrelated task
NOT trigger
Score
Criteria
9-10
Reliable triggers, zero false positives
7-8
Usually triggers, rare false positives
4-6
Sometimes triggers, some false positives
0-3
Unreliable or excessive false positives
Step 4: Functional Tests (20 pts)
Category
Points
What to Test
Happy path
/8
Core use cases work correctly
Edge cases
/6
Unusual inputs, boundary conditions
Error handling
/3
Graceful failures, helpful messages
Output quality
/3
Results match expert expectations
Full report template: See {baseDir}/references/templates.md#full-report
NEVER Do (Evaluator Anti-Patterns)
NEVER score D1 high for tutorials — "How to use library X" is not expert knowledge, even if well-written.
NEVER ignore token cost — An 800-line skill that could be 200 lines is wasting 75% of context window.
NEVER pass security for "educational" shell=True — Legitimate purposes don't justify vulnerabilities.
NEVER assume environment — A skill perfect for CLI is worthless in claude.ai web.
NEVER conflate "comprehensive" with "good" — More content ≠ more value. Density matters.
NEVER skip gate checks — A skill with invalid YAML shouldn't get quality scores.
Common Failure Patterns → Fixes
Pattern
Symptom
Root Cause
Fix
Tutorial
Low D1, "What is X" sections
Author wrote for humans, not LLMs
Delete all content LLM already knows
Dump
>800 lines, no structure
No progressive disclosure
Split to references/, add loading triggers
Orphan References
references/ exists but never loaded
Missing "when to read" instructions
Add explicit "MANDATORY if [X]: Read..."
Invisible
Never triggers
Bad description
Move ALL trigger info to description, add keywords
Wrong Location
"When to Use" in body
Body loads AFTER trigger decision
Description = when, Body = how
Vague Warnings
"Be careful with X"
No actionable anti-patterns
Specific NEVER + surprising consequence
Quick Reference: Scoring Cheatsheet
D1 (Knowledge Delta): "Would deleting this make LLM worse?"
D2 (Mindset): "Does it shape HOW to think, not just WHAT to do?"
D3 (Anti-Patterns): "Specific NEVERs with surprising reasons?"
D4 (Structure): "<300 lines? Progressive disclosure? Loading triggers?"
Environment: "Will core features actually work?"
User: "Right audience? Right complexity level?"
The Meta-Question
After every evaluation, ask:
"Would an expert in this domain say: 'Yes, this captures knowledge that took me years to learn'?"
Yes → The skill has value.
No → The skill is compressing what the LLM already knows.
Bento-Ready Report Output
For visual dashboard integration, generate JSON reports where evaluation logic directly determines visual weight.
Core Principles
异常放大，正常收敛 — Information density scales with deviation severity.
Speak human, not framework — Users haven't read our evaluation docs. No D1/D2/G1 jargon.
Status
Display Strategy
Normal score (≥80%)
Compact block, headline only
Notable issue (<60%)
Expanded block with detail + action
Critical issue (<30% or security)
Prominent block with full evidence
Output Formats
Request
Output
Standard evaluation
Markdown report (see {baseDir}/references/templates.md)
"bento report" / "visual report" / "JSON report"
Bento-ready JSON
Bento Report Structure
{"meta":{"skillName":"...","totalScore":48,"maxScore":60},"summary":{"verdict":"good","oneLiner":"Ready to use with minor improvements."},"blocks":[{"id":"expert-knowledge","type":"score","importance":{"level":"normal","reason":"Good expert content with minor redundancy"},"layout":{"size":"default"},"content":{"headline":"Expert Knowledge — Good ✓","detail":"Contains valuable professional insights. Some basic tutorials could be trimmed."}}]}
Importance → Layout Mapping
Importance
Layout
Trigger
critical
prominent
Gate fail, security issue, score <30%
notable
expanded
Score <60%, outlier disparity
normal
default
Score 60-80%
minor
compact
Score ≥80%, all gates pass
Visual Assets
Add visual_asset for critical/notable blocks:
Charts: gauge (header), radar (quality overview)
Illustrations: warning style for security issues
Badges: verdict display
Always include fallback.text for image generation failures.
MANDATORY for bento reports: Read {baseDir}/references/bento-report-schema.md for complete JSON schema.
MANDATORY for bento generation: Read {baseDir}/references/bento-report-instruction.md for generation rules.
Converted and distributed by TomeVault — claim your Tome and manage your conversions.
