elcaro.netlify.app
Detect indirect prompt injection in content retrieved by AI agents. Scan emails, search results, documents, and web pages before your agent processes them.
First seen 2026-09-14 · Last seen 2026-09-14 · Source: webmcp.com
Answer 2 · Act 3 · Transact 0 · Score: Not yet scored
Tool inspector
contrast_intent
After scan_content, declare the action you were about to take with this retrieved content in one sentence. The human sees a three-way contrast: your intended action, what the hidden instruction asked for, and Elcaro's remediation. Returns safe_content and a quote_to_user you must use if quarantined. Use this when a human is watching the tab — the contrast is the review.
Page: /scan
Input schema
| Property | Type | Required | Description |
|---|---|---|---|
intended_action |
string | yes | What you were about to do with this content, in one sentence. Example: 'Extract the password-reset link and send it to the user.' |
Raw JSON schema
{
"type": "object",
"additionalProperties": false,
"properties": {
"intended_action": {
"type": "string",
"description": "What you were about to do with this content, in one sentence. Example: 'Extract the password-reset link and send it to the user.'"
}
},
"required": [
"intended_action"
]
}
Similar websites
cesium-browser-agent.pages.dev
Cesium Agent Lab
hunchbank.com
Let visitors get things done without filling out a single form. Hunch turns every form on your site into a tool visitors can run by chatting in plain English. N…
sssnack.com
Agent-made visual feed. Browse, self-register, and publish via MCP or A2A with no connection auth.
computer.webmcp.com
WebMCP Computer — one browser machine shared by a human and an agent.