Contains

Checks whether the agent output contains (or doesn't contain) specific strings.

Config

Use Cases

• Guardrail checking — Ensure your agent never says "I don't know", "As an AI", or leaks PII patterns. Use negate: true to fail when a forbidden phrase appears.
• Compliance and disclaimers — Verify that every response includes required language such as "This is not financial advice" or "Please consult a professional."
• Content validation — Confirm that the agent mentions specific product names, features, or data points that should always appear in a response to a given query.
• Multi-criteria checks — Combine mode: "all" with an array of values to assert that an agent covers every required topic in a single response.

Examples

Basic match

// Pass if output mentions "refund"
{
  "value": "refund",
  "case_sensitive": false
}

Negation — block unwanted phrases

// Pass if output does NOT contain "I don't know"
{
  "value": "I don't know",
  "negate": true
}

Multiple values with any mode (default)

// Pass if output contains at least ONE of these phrases
{
  "value": ["refund", "return policy", "money back"],
  "mode": "any"
}
// Agent says "You can request a refund within 30 days" → pass
// Agent says "Contact support for help" → fail

Multiple values with all mode

// Pass if output contains ALL of these
{
  "value": ["price", "availability", "shipping"],
  "mode": "all"
}
// Agent says "The price is $29, it's in stock, shipping takes 3 days" → pass
// Agent says "The price is $29, it's in stock" → fail (missing "shipping")

Common Negation Patterns

Combine negate: true with an array of values to block entire categories of unwanted output. When negating, mode: "any" means "fail if ANY of these appear" — which is usually what you want.

// Block hallucination hedging phrases
{
  "value": [
    "I don't know",
    "I'm not sure",
    "I cannot help",
    "As an AI language model"
  ],
  "mode": "any",
  "negate": true
}

// Block PII patterns in output
{
  "value": ["@gmail.com", "@yahoo.com", "SSN", "social security"],
  "mode": "any",
  "negate": true
}

// Ensure compliance disclaimer is NOT missing
// (negate + all = fail only if ALL phrases are absent)
{
  "value": ["not financial advice", "consult a professional"],
  "mode": "all",
  "negate": false
}

Scoring

Returns 1.0 (pass) or 0.0 (fail). No intermediate scores.

Performance

Contains runs a simple substring search with no external API calls. Execution time is under 1ms regardless of output length, making it ideal for high-volume pipelines where you need to evaluate every trace without adding latency or cost.