Check text for prompt injection
check_prompt_injectionCheck untrusted text before your agent reads it (web pages, emails, social posts, tool results, documents): finds prompt-injection attempts such as instruction overrides, fake system or tool-call markers, hidden text, data-leak links and demands to move funds. Returns a verdict, a 0-1 risk score, each finding with its location, and a cleaned copy with hidden characters and flagged lines removed. With mode=memory it also guards long-term memory writes against planted history, standing orders and 'trusted' addresses. US$0.003 per call; 3 free calls a day.
Tool call template
Replace each placeholder with a value matching the input schema. Only required parameters are included; add optional parameters as needed.
{
"id": 1,
"jsonrpc": "2.0",
"method": "tools/call",
"params": {
"arguments": {
"text": "<text>"
},
"name": "check_prompt_injection"
}
}Input Schema
| Name | Required | Description | Default |
|---|---|---|---|
| mode | No | 'memory' when the text is about to be saved to long-term memory. | |
| text | Yes | The untrusted text (up to 20,000 characters). | |
| context | No | Where it came from, e.g. 'web page', 'tweet'. |
Output Schema
| Name | Required | Description | Default |
|---|---|---|---|
| paid | No | ||
| risk | No | ||
| trial | No | ||
| layers | No | ||
| message | No | ||
| verdict | No | ||
| findings | No | ||
| cleaned_text | No | ||
| pay_per_call | No | ||
| hidden_characters_removed | No |