skill-tester

name	skill-tester
description	Test a OpenCode skill with prompt scenarios, invocation checks, negative cases, and execution validation.

Skill Tester

Prompt-level testing tool for OpenCode skills. Verifies that a skill is selected when it should be, ignored when it should not be, and behaves correctly once invoked.

Capabilities

• discovery testing
• negative testing
• execution testing
• regression testing
• integration testing
• configuration testing
• prompt-matrix coverage analysis

Eval Workflow

Use a prompt matrix instead of only abstract capability claims.

Positive case

Prompt: "Why is my review skill not being picked for PR reviews?"
Expected: skill-debugger is selected

Borderline case

Prompt: "Check whether this skill package is ready to ship"
Expected: skill-quality-analyzer or another clearly stronger fit

Negative case

Prompt: "Implement the caching layer in src/api.ts"
Expected: skill-debugger is not selected

Input Requirements

Basic:

• skill name or path
• prompt scenario
• expected behavior

Comprehensive:

• skill path
• positive prompts
• borderline prompts
• negative prompts
• expected outputs or checks after invocation

Example Test Case

{
  "skill": "skill-debugger",
  "test": "Under-trigger diagnosis",
  "prompt": "Why is my code-review skill not being picked?",
  "expected_behavior": "skill-debugger is selected and audits name, description, and flags",
  "priority": "HIGH"
}

Output Formats

Quick Result

Testing: skill-debugger
Test: Under-trigger diagnosis
Result: PASS

Detailed Report

=== Skill Test Report ===

Skill: skill-debugger
Tests Run: 12
Passed: 10
Failed: 2
Coverage: 83%

Failures:
1. Negative case failed because description is too broad
2. Auto-selection failed because disable-model-invocation is true

Recommendations:
- tighten description
- add a negative example
- fix invocation flag

Test Types

1. Discovery Tests

Purpose: verify the skill triggers when it should.

2. Negative Tests

Purpose: verify the skill does not trigger outside its lane.

3. Execution Tests

Purpose: verify correct behavior after the skill is selected.

4. Edge Case Tests

Purpose: validate flag and routing edge cases.

5. Integration Tests

Purpose: verify the skill works correctly with adjacent skills.

6. Upstream-Conformance Tests

Purpose: for upstream-derived skills, verify that the local file still matches the vendored baseline where expected and that local adaptations are explicitly isolated.

How to Use

Quick

Test if skill-debugger triggers for under-triggering requests
Verify code-review skill works for pull requests
Test skill-quality-analyzer on itself

Comprehensive

Run a full prompt suite on skill-debugger with positive and negative cases
Test all capabilities of a custom skill package

Before Finalizing a New Skill

Write a prompt matrix for this new skill before we finalize the description
Create positive, borderline, and negative cases for the package

Coverage Goals

Skill Type	Target Coverage
Core routing or safety skill	95%+
Production skill	80%+
Experimental skill	60%+
Utility skill	50%+

Development Workflow

Phase 1: Draft Test Matrix

1. Write positive, borderline, and negative prompts
2. Document expected selection and behavior
3. Finalize description and flags

Phase 2: Continuous Testing

1. Re-run after every meaningful skill edit
2. Add new prompt classes when scope expands
3. Track regressions over time

Phase 3: Pre-Release

1. full prompt suite
2. negative and edge case validation
3. integration tests
4. user acceptance testing

Best Practices

1. Test real prompts, not only idealized summaries
2. Include negative cases so the skill stays in its lane
3. Test metadata such as disable-model-invocation, user-invocable, and allowed-tools
4. Document results for regression checking
5. For upstream-derived skills, test both baseline conformance and local adaptation behavior
6. Re-test after changing name or description

When NOT to Use This Skill

• For diagnosis and root cause, use skill-debugger
• For static quality scoring, use skill-quality-analyzer
• For first-draft authoring or upgrades, use skill-creator