Fact Check Agent
★Trusted◇Sandbox optionalMIT★Gold Verified100Agentby AgentNode · published 2 months ago · agent
Verify claims against multiple web sources and produce a fact-check verdict with supporting evidence.
Verify a given claim by searching multiple sources, extracting relevant evidence, cross-referencing facts, and producing a verdict with confidence level. This agent uses AgentNode's full skill registry to dynamically discover and install the capabilities it needs at runtime.
Quick Start
agentnode install fact-check-agentRuns in a subprocess with filtered environment by default. Declared permissions are policy-checked, not sandboxed.
Usage
Generated templatefrom fact_check_agent.agent import run
result = run()
print(result)Runs locally on your machine. No execution data is sent to AgentNode. Permissions are checked before execution. Learn how this works
AAgent Configuration
Tier
LLM + ToolsGoal
Verify a claim by searching for supporting and contradicting evidence, then producing a verdict.
Agent Behavior
description onlyYou are a fact-checking analyst. Verify claims by searching multiple web sources, cross-referencing evidence, assessing source credibility, and producing a verdict (confirmed/unconfirmed/false) with supporting evidence.
Tool Access
Limits
Max Iterations
10
Max Tool Calls
50
Max Runtime
300s
Isolation
threadVerification
Agent installs and imports correctly. runtime checks passed. 2/2 verification cases passed. manifest completeness 10/10.
This package was executed and validated by AgentNode before listing. Install, import, and runtime checks passed.
Verified in real_auto mode
Last verified 5d ago· Runner v2.0.0
Version History
Capabilities
Verify claims against multiple web sources and produce a fact-check verdict with evidence.
Input Schema
{
"type": "object",
"required": [
"goal"
],
"properties": {
"goal": {
"type": "string",
"description": "The objective for the agent"
}
}
}Output Schema
{
"type": "object",
"properties": {
"done": {
"type": "boolean"
},
"result": {
"type": "object"
}
}
}Permissions
Declared by the publisher. Checked before execution by the policy gate.
Permissions are policy-checked before execution. For trusted and curated packages that run on the host, network and filesystem access are policy-checked but not OS-sandboxed. When runtime isolation is required for untrusted/community code, AgentNode uses sandbox-or-fail-closed if the required container runtime and pinned image are available. Learn more
Privacy
All tool execution happens locally on your machine. AgentNode never receives:
- • Tool inputs or outputs
- • Execution logs
- • Data your agent processes
Only install events and search queries are sent to the registry.
agentnode install fact-check-agentFiles (4)
License
MITStats
Compatibility
Frameworks
Runtime
pythonPython Version
>=3.10Trust & Security
Publisher
AgentNode
@agentnode