Skill Quality Evaluator 📊

Score any skill on 6 dimensions. Catch the 30% of skills that look good but fail silently.

Why This Matters

Tessl Research (April 2026) found:

• 20% accuracy gain when using a good skill vs no skill
• 3X cost savings when small model + right skill matches large model
• 40% activation rate — agents often fail to use available skills
• 30% of evaluation tasks have leakage — skills that seem great but aren't

This skill helps you evaluate and improve your skills systematically.

6-Dimension Evaluation

1. Activation Reliability (0-100)

Can the agent find and activate this skill when needed?

Checklist:

• Trigger words are specific and unambiguous
• Description matches actual functionality
• No conflicting skills with similar triggers
• Skill is discovered when user asks relevant questions

Common Issues:

• Vague description → agent doesn't know when to use it
• Missing trigger words → skill never activates
• Too broad → activates when it shouldn't

Score Guide:

• 90+: Agent activates correctly 95%+ of the time
• 70-89: Activates in most relevant contexts
• 50-69: Sometimes activates, sometimes misses
• <50: Agent rarely finds/uses this skill

2. Task Coverage (0-100)

Does the skill handle the tasks it claims to cover?

Checklist:

• Each claimed capability has a usage example
• Edge cases are documented
• Known limitations are stated
• Failure modes are explained

Common Issues:

• Claims broad coverage but only handles happy path
• No examples for secondary features
• Undocumented prerequisites

Score Guide:

• 90+: All claimed tasks have working examples
• 70-89: Main tasks covered, some gaps in secondary features
• 50-69: Core functionality works but incomplete
• <50: Major claims unsupported

3. Instruction Clarity (0-100)

Can the agent follow the instructions without confusion?

Checklist:

• Instructions are step-by-step, not vague guidelines
• Decision points have clear criteria
• Output format is specified
• Anti-patterns are listed

Common Issues:

• "Do X when appropriate" → when is appropriate?
• Missing priority/precedence rules
• Contradictory instructions

Score Guide:

• 90+: Agent follows instructions correctly 95%+ of the time
• 70-89: Mostly clear, occasional confusion
• 50-69: Agent frequently asks for clarification
• <50: Instructions are ambiguous or contradictory

4. Leakage Resistance (0-100)

Does the evaluation actually test the skill, or does it leak answers?

Checklist:

• Examples don't contain verbatim solutions
• Test tasks require genuine skill application
• No shortcut paths that bypass skill content
• Evaluation criteria measure real capability

Common Issues (from Tessl Research):

• Example tasks are too similar to skill content
• Skill contains answers verbatim
• Test can be solved by pattern matching without understanding