AgentSkillsCN

test-orchestrator

通过扫描命令、代理与技能,生成测试场景,并以手动验证的方式逐步执行,从而测试 ClaudeKit 工作流。

SKILL.md
--- frontmatter
name: test-orchestrator
description: Test ClaudeKit workflows by scanning commands/agents/skills, generating test scenarios, and executing step-by-step with manual verification.

Test Orchestrator

Automated testing framework for ClaudeKit Marketing workflows.

When to Use

  • Testing new commands after implementation
  • Validating agent orchestration flows
  • Verifying skill integrations work correctly
  • Regression testing after changes
  • End-to-end workflow validation

Workflow

1. Scan Components

bash
# Generate fresh catalogs
python .claude/scripts/generate_catalogs.py --all

# Or scan specific type
python .claude/skills/test-orchestrator/scripts/scan-components.py

2. Select Test Scope

ScopeDescription
commandTest single command with happy case
workflowTest multi-step workflow
integrationTest skill + agent + command together
fullComplete end-to-end test suite

3. Execute Tests

Each test step pauses for manual verification:

code
[STEP 1/5] Executing: /youtube:social "https://youtube.com/..."
─────────────────────────────────────────────────
[OUTPUT]
...generated content...
─────────────────────────────────────────────────
[VERIFY] Check output matches expected format
[PASS/FAIL?] > _

Test Case Format

yaml
name: youtube-to-social-flow
description: Convert YouTube video to multi-platform social posts
type: integration

steps:
  - name: Extract video data
    action: vidcap summary
    input: "https://youtube.com/watch?v=dQw4w9WgXcQ"
    verify:
      - Response contains video title
      - Response contains summary content

  - name: Generate social posts
    action: /youtube:social
    input: "{video_url}"
    verify:
      - Twitter post under 280 chars
      - LinkedIn post has professional tone
      - No anti-pattern hooks used

  - name: Apply writing style
    action: copywriting skill
    input: Apply casual style to Twitter post
    verify:
      - Contains contractions
      - Uses first-person

Pre-Built Test Scenarios

1. YouTube Pipeline

StepCommand/SkillInputVerify
1vidcap.py infoYouTube URLReturns title, views, duration
2vidcap.py summaryYouTube URLReturns structured summary
3/youtube:socialYouTube URLMulti-platform posts generated
4/youtube:blogYouTube URLSEO article generated
5/youtube:infographicYouTube URLVisual layout generated

2. Content Creation

StepCommand/SkillInputVerify
1/content:blogTopicArticle with frontmatter
2/content:croArticleCRO-optimized version
3/socialArticle summaryPlatform posts

3. Email Automation

StepCommand/SkillInputVerify
1/email:flowwelcome5-email sequence
2email-marketing skillSequenceTiming + decision branches
3copywriting skillEmail bodyPAS/AIDA formula applied

4. Brand Consistency

StepCommand/SkillInputVerify
1inject-brand-context.cjs-Returns brand JSON
2/brand:updatepresetTokens synced
3Content generationAnyBrand voice applied

Happy Case Prompts

Pre-validated inputs that should always succeed:

yaml
youtube_url: "https://www.youtube.com/watch?v=dQw4w9WgXcQ"
blog_topic: "10 productivity tips for remote workers"
email_flow: "welcome"
brand_preset: "ocean-professional"
social_platform: "twitter"
writing_style: "casual"

Manual Verification Checklist

At each step, verify:

  • Output format matches expected structure
  • No error messages in response
  • Content quality acceptable
  • Anti-patterns avoided (for hooks)
  • Brand voice consistent (if applicable)
  • File saved to correct path (if applicable)

Integration

Use with:

  • /test command to launch test runner
  • debugging skill for failure analysis
  • code-review skill for output validation