v1.0.0 — CIGE Agentic Test Automation Guidelines

Agentic Validation Skills v1.0.0

Guidelines and a Claude Code skill for authoring self-healing, goal-driven
agentic test cases using the CIGE standard.

What's included

CLAUDE.md — always-on behavioral guidelines (6 principles)

• Establish Context before testing — narrow the agent's decision space
• Anchor every test to Intent — stable outcome over mutable steps
• Define Guardrails before executing — bounded autonomy is safe autonomy
• Treat Execution as adaptive guidance — goal fidelity over procedural fidelity
• Load context in layers — filter before fetching, never all upfront
• Self-heal by field — classification determines what each agent may touch

/agentic-validation-skills:cige-guidelines skill

• CIGE test case format with structured Context (including specRef), Intent,
  Guardrails, and ID-tagged Execution steps
• Three self-healing agents with explicit field mutation rules:
  • EnvironmentRecoveryAgent — restores infrastructure, touches nothing in the test
  • StaleExecutionAgent — rewrites Execution[] only, with human approval
  • ProductDefectAgent — escalates bugs; proposes Intent updates against the BRD,
    with human approval
• specRef pattern — lazy-fetched BRD pointer that distinguishes product defects
  from intentional product changes
• 3-layer progressive context disclosure (Orient → Bound → Execute) with per-step
  IDs and run summaries
• Quality checklist for test readiness