docs: add VerifierAgent pattern for output verification

docs/examples/verification/verifier_agent_crew.md

@@ -0,0 +1,101 @@
+# VerifierAgent Pattern: Validating Agent Outputs in CrewAI
+
+## Overview
+
+In multi-agent systems, Large Language Model (LLM) agents may generate outputs that are
+factually incorrect, inconsistent, or hallucinated. While CrewAI enables powerful
+agent orchestration, it does not enforce output validation by default.
+
+This example demonstrates a **VerifierAgent pattern**, where a secondary agent is used
+to evaluate and validate the output of a primary agent before it is accepted or used
+downstream.
+
+---
+
+## Problem
+
+- LLM agents can hallucinate facts
+- Incorrect outputs may propagate through agent pipelines
+- High-stakes workflows require an additional validation step
+
+Without verification, agent-based systems risk producing unreliable results.
+
+---
+
+## Solution: VerifierAgent Pattern
+
+The VerifierAgent pattern introduces a dedicated agent responsible for:
+- Fact-checking
+- Logical validation
+- Hallucination detection
+- Confidence scoring
+
+This separates **generation** from **verification**, improving overall reliability.
+
+---
+
+## Architecture
+
+```text
+User Query
+   ↓
+Generator Agent
+   ↓
+Verifier Agent
+   ↓
+Decision (Accept / Reject / Revise)
+```
+## How It Works
+
+A Generator Agent produces an initial response.
+
+A Verifier Agent reviews the response for factual accuracy and consistency.
+
+The Verifier Agent returns a structured evaluation.
+
+The result can be used to accept, reject, or revise the output.
+
+## Example Implementation
+
+A runnable example is provided in:
+
+verifier_agent_crew.py
+This example includes:
+
+One Generator Agent
+
+One Verifier Agent
+
+Sequential task execution
+
+
+## Sample Output
+
+
+{
+  "verdict": "incorrect",
+  "confidence": 0.45,
+  "issues": [
+    "Incorrect factual claim regarding climate change causes"
+  ],
+  "suggested_fix": "Revise the response using verified scientific sources"
+}
+
+
+## When to Use This Pattern
+Research assistants
+
+Enterprise decision systems
+
+RAG pipelines
+
+High-stakes AI workflows
+
+Multi-agent collaboration systems
+
+## Limitations
+Verification quality depends on the underlying LLM
+
+This pattern reduces risk but does not guarantee correctness
+
+For critical applications, combine with external tools or trusted data sources

docs/examples/verification/verifier_agent_crew.py

@@ -0,0 +1,48 @@
+from crewai import Agent, Task, Crew, Process
+
+# -----------------------------
+# Agents
+# -----------------------------
+generator_agent = Agent(
+    role="Content Generator",
+    goal="Generate a clear and accurate answer to the given question",
+    backstory="Expert assistant generating initial responses",
+    verbose=True
+)
+
+verifier_agent = Agent(
+    role="Verifier Agent",
+    goal="Verify factual accuracy and detect hallucinations",
+    backstory="Critical evaluator of AI outputs",
+    verbose=True
+)
+
+# -----------------------------
+# Tasks
+# -----------------------------
+generate_task = Task(
+    description="Explain the primary causes of climate change in simple terms.",
+    expected_output="A factually correct explanation of climate change causes.",
+    agent=generator_agent
+)
+
+verify_task = Task(
+    description=(
+        "Review the previous response for factual accuracy and hallucinations.\n"
+        "Return JSON with verdict, confidence, issues, suggested_fix."
+    ),
+    expected_output="A JSON verification report.",
+    agent=verifier_agent
+)
+
+# -----------------------------
+# Crew
+# -----------------------------
+crew = Crew(
+    agents=[generator_agent, verifier_agent],
+    tasks=[generate_task, verify_task],
+    process=Process.sequential
+)
+
+if __name__ == "__main__":
+    print(crew.kickoff())