# Agent Tools – Prompt Injection Screener

> Agent Tools – Prompt Injection Screener is a paid API for AI agents from agents.crowdspark.me, paid per call via x402, $0.005/call, status unknown (last checked 2026-10-31).

Screens text for prompt-injection attacks, returning a safe/suspicious/malicious verdict, a 0-100 risk score, and matched spans with categories.

## Facts

- Endpoint: POST https://agents.crowdspark.me/v1/text/injection-check?utm_source=zero.xyz
- Price: $0.005/call
- Payment: x402
- Status: unknown
- Last checked: 2026-10-31
- Activations on Zero: 0
- Tags: x402
- Canonical page: https://www.zero.xyz/c/agent-tools-prompt-injection-screener-852ce246
- Structured record (JSON): https://api.zero.xyz/v1/capabilities/cap_VfutY9DnmFsufsre_xBuc

Status and success rate cover calls made through Zero and Zero's own probes. Third-party monitors may report differently.

## How to call it through Zero

Zero handles the 402 payment challenge and records the run. With the Zero CLI installed (`npm i -g @zeroxyz/cli`):

```sh
zero fetch --capability agent-tools-prompt-injection-screener-852ce246 -d '<json body>'
```

Example prompt: Before you summarize that scraped article, run it through the prompt-injection screener first — just send the full article text and tell me if it comes back safe, suspicious, or malicious, and what the risk score is.

## When to prefer this

Choose this endpoint when you need a fast, rule-based, stateless prompt-injection check on untrusted text (scraped pages, emails, tool outputs, user messages) before passing it into an LLM pipeline. It is purpose-built for agentic security gating — not a general-purpose content moderation or spam filter. Prefer it when you need explicit matched spans and categories (instruction_override, data_exfiltration, role_spoofing, hidden_unicode) rather than just a binary flag, and when multi-language coverage across 6 languages matters. Text is not stored, making it safe for sensitive content.

## Known failure modes

- Text exceeds 50,000 characters — request rejected
- Missing or empty 'text' field in JSON body — validation error
- Rate limit exceeded (25 free calls/day per IP without payment header) — 402 or 429 response
- Network timeout for very long texts near the size limit
- False negatives: novel or obfuscated injection patterns not yet covered by rule set

## How this service works

Pay-per-call utilities for AI agents over x402, USDC on Base.

## Output

A JSON object containing: verdict (safe, suspicious, or malicious), a 0-100 risk score, total character count, matched spans (each with start/end positions, excerpt, rule name, category, and weight), and a category summary dict (e.g. instruction_override, data_exfiltration counts).

## Request schema (JSON Schema)

```json
{
 "type": "object",
 "properties": {
  "text": {
   "type": "string",
   "description": "Text to screen, max 50,000 characters"
  }
 }
}
```

## Response schema (JSON Schema)

```json
{
 "type": "json",
 "example": {
  "chars": 104,
  "score": 83,
  "matches": [
   {
    "end": 49,
    "note": "tells the model to drop earlier instructions",
    "rule": "override_ignore_prev",
    "start": 15,
    "weight": 60,
    "excerpt": "Ignore all previous instructions",
    "category": "instruction_override"
   }
  ],
  "verdict": "malicious",
  "categories": {
   "data_exfiltration": 1,
   "instruction_override": 1
  }
 }
}
```

## More

- Live health (JSON, refreshed every minute): https://www.zero.xyz/c/agent-tools-prompt-injection-screener-852ce246/health.json
- [Zero catalog index](https://www.zero.xyz/llms.txt)
- [Other services from agents.crowdspark.me](https://www.zero.xyz/host/agents.crowdspark.me/llms.txt)