loading…
Scenarios where tool outputs and fetched documents contain hidden instructions attempting to hijack the agent's tool-calling behavior.
/api/mcp/srv-prompt-injection-defenseMCP enable + token are only for calling from an external client — the playground below needs neither.
Perform deep analysis of potential prompt injection bypass attempts using semantic similarity and adversarial detection
{
"type": "object",
"required": [
"payload"
],
"properties": {
"payload": {
"type": "string"
},
"analysis_depth": {
"enum": [
"fast",
"standard",
"deep"
],
"type": "string"
},
"baseline_prompt": {
"type": "string"
},
"include_variants": {
"type": "boolean"
},
"similarity_threshold": {
"type": "number",
"maximum": 1,
"minimum": 0
}
}
}Deploy new prompt safeguard policy with input validation rules and output filtering controls
{
"type": "object",
"required": [
"name",
"enforcement_mode"
],
"properties": {
"name": {
"type": "string"
},
"priority": {
"type": "integer",
"maximum": 100,
"minimum": 1
},
"input_filters": {
"type": "array",
"items": {
"type": "string"
}
},
"model_targets": {
"type": "array",
"items": {
"type": "string"
}
},
"enforcement_mode": {
"enum": [
"monitor",
"warn",
"block"
],
"type": "string"
},
"output_sanitizers": {
"type": "array",
"items": {
"type": "string"
}
}
}
}Generate comprehensive incident report for prompt injection events with forensic details and mitigation timeline
{
"type": "object",
"required": [
"incident_ids",
"format"
],
"properties": {
"format": {
"enum": [
"json",
"csv",
"pdf",
"stix"
],
"type": "string"
},
"incident_ids": {
"type": "array",
"items": {
"type": "string"
}
},
"include_payloads": {
"type": "boolean"
},
"include_responses": {
"type": "boolean"
},
"classification_level": {
"enum": [
"public",
"internal",
"confidential",
"restricted"
],
"type": "string"
}
}
}Fetch detailed prompt injection detection event by detection_id including full payload and mitigation actions taken
{
"type": "object",
"required": [
"detection_id"
],
"properties": {
"detection_id": {
"type": "string"
},
"include_payload": {
"type": "boolean"
},
"include_response": {
"type": "boolean"
}
}
}Retrieve prompt injection detection events with filtering by time range, severity, and attack type
{
"type": "object",
"required": [],
"properties": {
"limit": {
"type": "integer",
"maximum": 1000,
"minimum": 1
},
"offset": {
"type": "integer",
"minimum": 0
},
"severity": {
"enum": [
"info",
"low",
"medium",
"high",
"critical"
],
"type": "string"
},
"time_range": {
"enum": [
"1h",
"6h",
"24h",
"7d",
"30d"
],
"type": "string"
},
"attack_type": {
"enum": [
"jailbreak",
"role_manipulation",
"context_injection",
"system_override",
"data_exfiltration"
],
"type": "string"
}
}
}Isolate suspicious chat session and prevent further model interactions until manual review
{
"type": "object",
"required": [
"session_id",
"reason"
],
"properties": {
"reason": {
"type": "string"
},
"session_id": {
"type": "string"
},
"auto_expire_hours": {
"type": "integer",
"maximum": 168,
"minimum": 1
}
}
}Analyze incoming user input for prompt injection patterns and return risk assessment with detected attack vectors
{
"type": "object",
"required": [
"input_text"
],
"properties": {
"input_text": {
"type": "string"
},
"model_context": {
"enum": [
"chat",
"completion",
"embedding",
"classification"
],
"type": "string"
},
"detection_rules": {
"type": "array",
"items": {
"type": "string"
}
},
"sensitivity_level": {
"enum": [
"low",
"medium",
"high",
"critical"
],
"type": "string"
}
}
}Query prompt injection indicators of compromise using pattern matching and behavioral signatures
{
"type": "object",
"required": [
"query"
],
"properties": {
"limit": {
"type": "integer",
"maximum": 500,
"minimum": 1
},
"query": {
"type": "string"
},
"time_range": {
"enum": [
"1h",
"6h",
"24h",
"7d",
"30d"
],
"type": "string"
},
"indicator_type": {
"enum": [
"pattern",
"behavioral",
"semantic",
"statistical"
],
"type": "string"
},
"confidence_threshold": {
"type": "number",
"maximum": 1,
"minimum": 0
}
}
}Modify existing detection rule configuration including patterns, thresholds, and response actions
{
"type": "object",
"required": [
"rule_id"
],
"properties": {
"name": {
"type": "string"
},
"action": {
"enum": [
"block",
"sanitize",
"flag",
"monitor"
],
"type": "string"
},
"enabled": {
"type": "boolean"
},
"rule_id": {
"type": "string"
},
"patterns": {
"type": "array",
"items": {
"type": "string"
}
},
"threshold": {
"type": "number",
"maximum": 1,
"minimum": 0
}
}
}