🛡️ exSafe — NFT Community Safety Desk

Check it before you sign it. exSafe gives NFT communities one clear verdict — 🟢 SAFE / 🟡 CAUTION / 🔴 DANGER — on any link, contract address, transaction, signature request, or suspicious announcement, before a member connects a wallet or signs.

Wallet drainers steal hundreds of millions of dollars a year (see Scam Sniffer's annual reports for the latest figure). Most of it happens through one bad signature — a setApprovalForAll, a Permit, a fake mint link posted in a Discord that got hijacked. Wallets like MetaMask show you almost nothing useful at that moment. exSafe does.

---

What it does

One verdict engine ("the brain"), exposed through the places communities actually are:

Surface	What it does
Web app	Paste anything → clear verdict card with a plain-language explanation and a transparent signal checklist.
Discord bot	Auto-scans every link/address posted in a channel and replies with a verdict. /check and /report slash commands.
/api/check	The shared brain any surface (or your own tool) can call.

It understands four kinds of input, auto-detected:

• Links — phishing blocklist (MetaMask's live eth-phishing-detect), typosquat / look-alike detection, domain age, community lists.
• Contracts / addresses — GoPlus reputation (phishing, drainer, honeypot, hidden owner…), Etherscan verified-source + contract age.
• Transactions & signatures (the flex) — decodes setApprovalForAll, approve, permit, Permit2 and Seaport signatures into plain English, and cross-checks the address receiving the permission against reputation feeds. "This gives 0x00…dbad permission to transfer EVERY NFT you own — and that address has been reported as a drainer."
• Announcements — Claude flags social-engineering tactics (false urgency, "we got hacked, use this new link", claim/airdrop bait) and checks every link inside.

The community wedge

Other tools are individual browser extensions. exSafe is a community safety desk: mods verify official links (allowlist), members report scams (blocklist), and every verdict is scoped to your community (Discord guild). The bot lives where your members already are — no install required.

---

Architecture — one brain, many doors

                 detect → checkers → score → narrate (Claude)
                 ┌───────────────────────────────────────────┐
  web app  ─────▶│  lib/engine/runCheck()                     │
  discord  ─────▶│    ├─ checkers/url        (phishing, RDAP) │
  /api     ─────▶│    ├─ checkers/contract   (GoPlus, scan)   │
                 │    ├─ checkers/calldata   (viem decode)    │
                 │    ├─ checkers/announcement (Claude)       │
                 │    └─ community/store     (allow/blocklist)│
                 └───────────────────────────────────────────┘

Everything degrades gracefully: with zero API keys the app still runs (GoPlus, phishing list, RDAP and the community layer need no key; LLM narration falls back to deterministic text — set AI_API_KEY for Xiaomi MiMo / OpenAI-compatible, or ANTHROPIC_API_KEY for Claude; Etherscan signals are simply skipped and noted).

AI keys (narration)

# Xiaomi MiMo (recommended if you already farmed keys into 9Router)
AI_PROVIDER=openai
AI_API_KEY=sk-...
AI_BASE_URL=https://api.xiaomimimo.com/v1
AI_MODEL=mimo-v2.5-pro

Do not point Vercel at localhost:20128 (9Router). Use the public MiMo base URL above. Anthropic still works via ANTHROPIC_API_KEY if preferred.

---

Quickstart

npm install
cp .env.example .env.local   # optional — the app runs without any keys

# 1) Web app + API
npm run dev                  # http://localhost:3000

# 2) Discord bot (needs a bot token — see .env.example)
npm run bot

Try the buttons on the homepage, or:

curl -s localhost:3000/api/check -H 'content-type: application/json' \
  -d '{"input":"opensea-mint.net"}'

Discord setup {#discord-setup}

1. Create an app at https://discord.com/developers/applications.
2. Bot tab → enable Message Content Intent (required for auto-scan).
3. Invite with scopes bot + applications.commands (perms: Send Messages, Embed Links, Read Message History).
4. Put the token/client id (and optionally a guild id) in .env.local.
5. Set NEXT_PUBLIC_DISCORD_INVITE to the bot invite URL so the landing page shows Add to Discord.
6. Point EXSAFE_API_URL at the deployed web API, then npm run bot.

Shareable verdicts & judge demo

• Any check updates the URL: ?q=<input>&chain=8453&lang=en
• Use Copy share link on the verdict card
• Judge path: open /?demo=judge or click Run demo (official → typosquat → drainer approval → hijacked announcement)

Default chain is Base (8453) for Vibestarter / Wave 1 alignment.

---

API

POST /api/check

// request
{ "input": "0x...", "chainId": "8453", "community": "guild-123", "lang": "en" }
// response
{ "verdict": "DANGER", "score": 100, "kind": "calldata",
  "summary": "...", "explanation": "...", "recommendation": "...",
  "signals": [{ "label": "...", "severity": "danger", "detail": "...", "source": "GoPlus" }],
  "meta": { "chainName": "Base", "checkedAt": "...", "degraded": [], "aiNarrated": true } }

POST /api/report → { "value": "scam.xyz", "type": "domain", "list": "block", "community": "guild-123" }

---

Data sources

MetaMask eth-phishing-detect · GoPlus Security · Etherscan · RDAP · on-chain calldata decoding via viem · your community's own reports.

Limitations

exSafe surfaces risk signals; it is not financial advice, and no automated check is a guarantee of safety. Absence of a signal is not proof something is safe. RDAP does not cover every TLD (e.g. .io). The community store is a file-backed MVP — swap it for a database for multi-instance persistence.

Roadmap

• Full transaction simulation (Tenderly / Alchemy) showing exact asset outflow
• MetaMask Snap for in-wallet insights at signing time
• Persistent DB + mod dashboard for the community desk
• Browser extension companion