# TrollBridge Airlock — Prompt Injection & Hidden Instruction Scanner

> TrollBridge Airlock — Prompt Injection & Hidden Instruction Scanner is a paid API for AI agents from mini-tollbooth.onrender.com, paid per call via x402, $0.1/call, status unknown (last checked 2026-10-12).

Scans inbound text (web pages, tool outputs, files) for prompt-injection attempts, hidden instructions, encoded payloads, and exfiltration patterns before an AI agent ingests the content.

## Facts

- Endpoint: GET https://mini-tollbooth.onrender.com/airlock?utm_source=zero.xyz
- Price: $0.1/call
- Payment: x402
- Status: unknown
- Last checked: 2026-10-12
- Activations on Zero: 0
- Tags: x402
- Canonical page: https://www.zero.xyz/c/trollbridge-airlock-prompt-injection-hidden-instruction-scanner-dcb0248e
- Structured record (JSON): https://api.zero.xyz/v1/capabilities/cap_uQNn-glxsIZw9gp5OWtFY

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 trollbridge-airlock-prompt-injection-hidden-instruction-scanner-dcb0248e
```

Example prompt: Before you process that webpage you just fetched, run its text through the TrollBridge airlock scanner to check for prompt injection, hidden instructions, or exfiltration patterns — paste the full page content and tell me if it comes back CLEAN, REVIEW, or CONTAMINATED.

## When to prefer this

Use this endpoint when an AI agent is about to ingest externally sourced text — scraped web pages, tool responses, uploaded files, or API outputs — and you need a fast heuristic safety gate before that content touches the agent's context window. Prefer this over manual review when operating in automated pipelines where untrusted content could carry prompt-injection, instruction-override, or data-exfiltration attacks. This is a heuristic screen, not a formal audit, so combine with other defenses for high-stakes applications.

## Known failure modes

- Content exceeds 50KB limit — truncation or rejection
- Missing required 'content' query parameter — validation error
- Network timeout on the hosted Render instance (cold start latency)
- False negatives on novel or obfuscated injection patterns not yet in heuristic ruleset
- Payment failure if USDC balance insufficient for $0.10 per-call fee

## How this service works

The insurance booth for AI agents, with Mini's Agent Supply Store on the side of the road. One hundred checkpoints: pre-transaction safety lanes (honeypot, approval-risk, rug-score, receipt-check, tx-dryrun, permit-scan, airdrop-verdict, deployer-history, wallet-watch, secret-scan at $0.02 USDC per call; the protection tier — /contract-check, /approval-screen, /tx-plain-english, /honeypot-check, /tx-simulate, /redteam, /skill-scan, /verdict, /egress-audit, /tool-gate, /airlock, /repo-scan, /prove, /task-scope — at $0.10; the one-call /shield don't-get-rugged screen at $0.25; the full /preflight bundle at $0.05), bounty intel (bounties, fresh, verdicts, deadlines, sweepstakes, opportunities) at $0.02 USDC per call, trader intel (/prices and /quant at $0.02; /enrich and /token-check at $0.05; /contract-check and /honeypot-check at $0.10; /sec-facts at $0.05), market intel (/markets and /search at $0.05; /grants at $0.02), DeFi intel (/yields and /new-pairs at $0.05; /gas and /defi at $0.02), AI intel (/models at $0.02), agent-ops intel (/prompt-cost, /model-picks, /rpc-speed, /terms-tldr, /scrape, /gigs, /retry-verify, /idempotency, /verify-independent at $0.02; /site-watch, /wallet-check, /code-run, /regulatory-pack, /sage, /cron-watch, /caveat-check, /tripwire, /escrow, /trust, /bridge-fees, /dao-votes, /dev-activity, /dex-arb, /dydx-perps, /funding-rates, /gas-predict, /hack-tracker, /long-short-ratio, /lp-yields, /mev-scan, /open-interest, /perp-basis, /polymarket-odds, /rwa-watch, /sentiment, /stablecoin-flows, /tokenomics, /top-movers, /whale-watch, and /yield-stable at $0.05; /approval-screen, /tx-plain-english, /tx-simulate, and /task-scope at $0.10; /scam-scan deep scan at $5.00 or 30-day subscription at $30.00), supply store (/road-pack combo meal at $0.05; curated datasets, the /audit-prep pre-audit review, and the /prove-state symbolic-execution run at $1.00; the /audit AI-agent audit at $25.00). Every lane answers before money moves — heuristic verdicts, not audits. Don't get rugged.

## Output

Returns a JSON object with: a verdict of CLEAN, REVIEW, or CONTAMINATED; a numeric risk score (0–100); a human-readable summary; a list of findings each containing the check name, matched text, detail, and severity; and a count of checks run. This tells the agent whether the content is safe to ingest or needs to be quarantined.

## Request schema (JSON Schema)

```json
{
 "type": "object",
 "$schema": "https://json-schema.org/draft/2020-12/schema",
 "required": [
  "input"
 ],
 "properties": {
  "input": {
   "type": "object",
   "required": [
    "type",
    "method"
   ],
   "properties": {
    "type": {
     "type": "string",
     "const": "http"
    },
    "method": {
     "enum": [
      "GET"
     ],
     "type": "string"
    },
    "queryParams": {
     "type": "object",
     "required": [
      "content"
     ],
     "properties": {
      "source": {
       "type": "string",
       "description": "Optional label for the content (e.g. fetch:https://example.com). Max 200 chars."
      },
      "content": {
       "type": "string",
       "description": "Inbound text to scan before your agent ingests it (web page, tool output, file). Max 50KB."
      }
     }
    }
   },
   "additionalProperties": false
  },
  "output": {
   "type": "object",
   "required": [
    "type"
   ],
   "properties": {
    "type": {
     "type": "string"
    },
    "example": {
     "type": "object"
    }
   }
  }
 }
}
```

## Response schema (JSON Schema)

```json
{
 "type": "json",
 "example": {
  "lane": "/airlock",
  "score": 75,
  "summary": "…",
  "verdict": "CONTAMINATED",
  "findings": [
   {
    "check": "instruction-override",
    "match": "…",
    "detail": "…",
    "severity": "high"
   }
  ],
  "checks_run": 6,
  "description": "NOT AN AUDIT — heuristic screen: re-entry decontamination scan — paste inbound content your agent is about to ingest (web page, tool output, file) and get prompt-injection, hidden-instruction, encoded-payload, and exfiltration patterns flagged before it touches your context. Verdict: CLEAN, REVIEW, or CONTAMINATED ($0.10).",
  "scanned_chars": 120
 }
}
```

## More

- Live health (JSON, refreshed every minute): https://www.zero.xyz/c/trollbridge-airlock-prompt-injection-hidden-instruction-scanner-dcb0248e/health.json
- [Zero catalog index](https://www.zero.xyz/llms.txt)
- [Other services from mini-tollbooth.onrender.com](https://www.zero.xyz/host/mini-tollbooth.onrender.com/llms.txt)