إنتقل إلى المحتوى الرئيسي

Evaluation And Testing

The agent library is validated at three levels.

Registry Validation​

  • every prompt parses against the schema
  • every workflow ships codex, claude, and gemini
  • no duplicate prompt IDs
  • every prompt has verification commands, stop conditions, and handoff requirements

Content Validation​

  • prompts must mention Sham Bus repo reality
  • prompts must include repo scopes and required reads
  • Codex prompts must be implementation-oriented
  • Claude prompts must show regression or assumption awareness
  • Gemini prompts must show phase or structure awareness

Golden Coverage​

We specifically validate prompts for:

  • ecosystem audit
  • shared-core extraction
  • dashboard/admin hardening
  • mobile flow fixing
  • CI false-green triage
  • partner architecture drafting