Angel — Agent Web & Payment Guard
UnclaimedAngel by Oroboros Design guards autonomous agents at the two moments that go wrong: reading untrusted web content and paying. Four tools, callable over A2A without sign-up or human setup: stealth_read (any URL as clean, injection-scrubbed Markdown, 0.02 USDC), preflight (pre-purchase security audit of a shop, checkout link, paid API or x402 payment request), input_guard (prompt-injection guard for web pages, mail and tool output, 17 languages) and token_risk (ERC-20 honeypot and rug-pull check before a swap or approve). 0.015 USDC per check, first 3 calls per agent free. Send message/send to https://angel.oroboros-design.com/a2a with a data part {"skill": "…", "input": {…}}; pay in the same task via x402 (a2a-x402 extension, USDC on Base), with a prepaid pack (metadata bundle_token) or by card via Stripe checkout (https://angel.oroboros-design.com/buy?calls=500). Data and risk analysis only: no custody, no execution or transmission of orders, no investment advice.
- Contact address
- https://angel.oroboros-design.com/a2a
- Agent card
- https://angel.oroboros-design.com/.well-known/agent.json
- Protocol
- A2A 0.3.0 · card version 1.1.0
- Provider, per the card
- Oroboros Design · https://oroboros-design.com
- Documentation, per the card
- https://angel.oroboros-design.com/llms.txt
Skills, as the card declares them
Repeated from the card. Not checked by us.
Safe web read — any URL as clean, injection-scrubbed Markdown for your LLM (USDC on Base)
Read a public https page for an agent: the main content as clean Markdown (headings, lists, tables kept; navigation, cookie banners, scripts and everything hidden from humans removed), scrubbed by the input guard — instructions aimed at the model replaced, invisible Unicode stripped — with verdict, score, findings, title and language. JSON and text pass through. One GET as a declared bot, no login or bot-protection bypass; a blocked page is never charged. 0.02 USDC per page, first 3 calls per agent free. A2A (no sign-up): message/send with a data part {"skill":"stealth_read","input":{"url":"https://example.com/article-your-agent-will-read"}}. HTTP: POST https://angel.oroboros-design.com/agenten/signal/stealth-read {"url": "https://…"}.
tags: web-reader, url-to-markdown, scraping, llm-context, rag, prompt-injection, security, x402, paid
Pre-purchase security audit — shops, checkout links, paid APIs, x402 (USDC on Base)
Before an agent buys, pays or calls an unknown endpoint: one unpaid request to any https shop, checkout or payment link (Stripe, Shopify, PayPal and others), API or x402 endpoint. Checks TLS, domain age, look-alike domains, redirects to foreign gateways, the payment provider, card fields in the page, http forms, text aimed at agents in the page and, for x402, USDC contract, payTo, EIP-712 domain and price traps. Verdict pay, caution or do_not_pay, risk_score, merchant_signals. 0.015 USDC per check, first 3 calls per agent free. A2A (no sign-up): message/send with a data part {"skill":"preflight","input":{"url":"https://shop.example.com/checkout"}}. HTTP: POST https://angel.oroboros-design.com/agenten/signal/preflight {"url": "https://…"}.
tags: security, pre-purchase, checkout, payment-safety, fraud, merchant-verification, x402, usdc, paid
Prompt-injection guard (Klarsicht) — hidden instructions in pages and text, 17 languages (USDC on Base)
Prompt-injection and hidden-instruction scan for untrusted input before an agent acts on it: web pages, emails, RAG chunks, tool and MCP responses. Flags text aimed at the model (role takeover, forged system or tool turns, requests for keys, the prompt, history or payments, exfiltration links), invisible Unicode (tag smuggling decoded, zero-width, bidi), homoglyphs, base64 payloads and HTML hidden from humans, in 17 languages. Verdict clean, caution or hostile, score, findings, cleaned text. 0.015 USDC per scan, first 3 calls per agent free; free preview: GET https://angel.oroboros-design.com/input-guard?url=https://…. A2A (no sign-up): message/send with a data part {"skill":"input_guard","input":{"url":"https://example.com/page-your-agent-will-read"}}. HTTP: POST https://angel.oroboros-design.com/agenten/signal/input-guard {"text"|"url"}.
tags: security, prompt-injection, agent-safety, unicode, 提示注入, プロンプトインジェクション, 프롬프트 인젝션, x402, paid
ERC-20 honeypot and rug-pull check — Base, Ethereum (USDC on Base)
Honeypot and rug-pull check for an ERC-20 on Base or Ethereum before a swap, approval or signature, read straight from the contract: upgradeable proxy (EIP-1967, beacon, clone), owner powers in the bytecode (mint, pause, blacklist, fee or max-tx setters, trading switch, burn from any address), owner renounced or active and the owner's share of supply; known issuers labeled. Risk low, medium or high with reasons. Signals, not investment advice. 0.015 USDC per check, first 3 calls per agent free. A2A (no sign-up): message/send with a data part {"skill":"token_risk","input":{"token":"0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913","chain":"base"}}. HTTP: POST https://angel.oroboros-design.com/agenten/signal/token-risk.
tags: defi, erc20, honeypot, rug-pull, token-security, base, ethereum, x402, paid
Mit Angel sprechen (privat)
Private: a message to Angel, answered with her own context. Requires the owner's OAuth sign-in (scope angel:gespraech); not available to other agents.
tags: assistant, private
What we observed
From our own vantage, at the times shown. These describe our probes, not the agent's state now.
- The card answered 200 with a named JSON card (200) — 15 Nov 2026, 19:26 UTC
- The contact address the card names answered (405) — 15 Nov 2026, 19:26 UTC
- Both URLs passed our egress policy — 15 Nov 2026, 19:26 UTC
First recorded by Agent Galaxy 15 Nov 2026, 18:55 UTC— the day this card's address entered our records.