# ==========================================
# Crilio Configuration (crilio.yaml)
# The CI/CD quality gate for AI.
# ==========================================

settings:
  # Provider is automatically inferred from your API key (OPENAI_API_KEY or ANTHROPIC_API_KEY)
  provider: openai

  # The AI model you are testing (Target)
  target_model: gpt-4o-mini

  # The fast/cheap model that grades the tests (Judge)
  judge_model: gpt-4o-mini

  # Hard limit on API spending per month (in USD) to prevent surprises
  max_monthly_budget_usd: 10.0

# ==========================================
# Test Cases
# ==========================================

tests:
  # Test 1: Semantic Rule Check (API)
  - name: "Refund Policy Check"
    prompt: "How long do I have to return a product?"
    rules:
      - "Must mention the 30-day return window."
      - "Must NOT mention competitor names like Amazon or Walmart."
    tags:
      - "refund"
      - "policy"
      - "customer service"

  # Test 2: JSON Formatting Enforcement (API)
  - name: "JSON Format Check"
    prompt: "Return my user status as JSON."
    rules:
      - "Must return valid JSON with keys 'status' and 'user_id'."
      - "Must NOT include apologies or extra prose."
    tags:
      - "json"
      - "formatting"

  # Test 3: Local Model Execution (Zero API Cost, Total Privacy)
  # Uncomment the lines below to test a local model via Ollama!
  # - name: "Local Llama 3 Check"
  #   prompt: "Say hello in a professional tone."
  #   target:
  #     command: "ollama run llama3 '{{prompt}}'"
  #   rules:
  #     - "Must contain the word 'Hello' or 'Hi'."
  #   tags:
  #     - "local"
  #     - "llama"

Keys:

• provider: openai or anthropic
• model / judge_model: override defaults
• settings.max_monthly_budget_usd: monthly cap — run stops when exceeded, shows Budget: $spent / $cap (pct) after each test
• system: global system prompt (passed to Target); per-test system overrides
• tests[].name must be unique
• tests[].rules are natural language — Judge is literal. Use Must / Must NOT.
• tests[].tags optional list — filter with crilio run --tag smoke (no flag = all, missing tags = skipped when filtered)
• tests[].target.command — local CLI e.g. python bot.py '{{prompt}}' or ollama run llama3 '{{prompt}}', {{prompt}} → shlex.quote, no placeholder → stdin pipe, timeout 30s, $0 target, injection-safe
• tests[].skip — set skip: true to pause a test (shows SKIPPED)
• Per-test provider/model/judge_model/system/tags/target/skip overrides global (advanced, needs its own key).
• Leak Guard: crilio.yaml containing sk-... or api_key: fails validate/run (exit 2) — use .env / Secrets, never commit keys.

# 1. ollama serve & ollama pull llama3  (or mistral, qwen2, etc.)
# 2. crilio.yaml:
settings:
  target_model: gpt-4o
  judge_model: gpt-4o-mini

tests:
  - name: Ollama Refund
    prompt: "How long do I have to return a product?"
    target:
      command: "ollama run llama3 '{{prompt}}'"  # any model: mistral, qwen2, gemma
      # no placeholder also works → stdin: command: "ollama run llama3"
    rules:
      - "Must mention the 30-day return window."
    tags: ["ollama", "local"]

# 3. export OPENAI_API_KEY="sk-..."  # Judge still API
# 4. crilio run --tag ollama --verbose