Plugin: SafeAgent — Token safety checks before trading

[Aigen-Protocol]

Plugin Proposal: SafeAgent Token Safety

Problem

AI agents using GOAT to trade on DEXes have no way to check if a token is safe before buying. Honeypots, rug pulls, and scam tokens cause significant losses.

Solution

SafeAgent provides pre-trade safety checks:

const { checkTokenSafety } = require('safeagent-sdk');

// Before any swap
const report = await checkTokenSafety(tokenAddress, 'base');
if (report.safety_score < 60) {
  console.log('Token unsafe:', report.flags);
  return; // Skip trade
}

Features

• Honeypot detection via DEX swap simulation
• 17 scam pattern checks on source code
• LP lock verification
• 6 EVM chains (Base, Ethereum, Arbitrum, Optimism, Polygon, BSC)
• Score 0-100 with risk flag bitmask
• Sub-second response (<1s cached, <1.5s fresh)

Integration

Available as:

• npm package: npm install CryptoGenesisSecurity/erc-token-safety-score
• MCP server: SSE at https://cryptogenesis.duckdns.org/mcp/sse
• REST API: https://cryptogenesis.duckdns.org/token/scan?address=0x...&chain=base
• On-chain oracle (Optimism): 0x3B8A6D696f2104A9aC617bB91e6811f489498047

Standard

Implements the ERC Token Safety Score — an open standard for token safety on EVM chains.

Comparison

GOAT already has Rugcheck for Solana SPL tokens. SafeAgent covers 6 EVM chains with deeper analysis (honeypot simulation, source code audit, LP lock check).

Happy to build the GOAT plugin if there's interest.

GitHub: https://github.com/CryptoGenesisSecurity/erc-token-safety-score

[mgopal20]

Hi — PaladinFi (Mallesh) here. Aigen-Protocol's #562 proposal targets token-contract risk; we ship a different scope (wallet-side primitive), so wanted to ask maintainers whether both could land as orthogonal plugins or whether you'd prefer one consolidated proposal.

For context, we shipped an in-tree action provider to Coinbase AgentKit last week (PR #1245) using a similar decision-only shape. PaladinFi's trust primitive:

• Wallet-OFAC SDN screen (/v1/trust-check/ofac): free, anonymous, no signup, ~1 r/s rate-limited, daily Treasury XML refresh. Returns allow/block with _real: true for downstream-agent verification. Wallet-side scope, complementary to Aigen-Protocol's contract-side proposal.
• Multi-source token trust (/v1/trust-check): single decision from OFAC + GoPlus + Etherscan + anomaly heuristics, fail-closed (any source unreachable → degrades to warn, never silent-allow). $0.001/call via x402 on Base.
• Production endpoint at https://swap.paladinfi.com (v0.11.77; see /health for current version; fail-closed-contract test suite; trust composition v1.1). The npm plugins @paladinfi/eliza-plugin-trust and @paladinfi/agentkit-actions are MIT-licensed and verifiable.

paladin-trust would land at typescript/packages/plugins/paladin-trust/ mirroring the rugcheck shape (no wallet client; pure decision surface). Constructor-only config; enablePaidMode default false so taker data never leaves on the free path.

Two questions if anyone has bandwidth:

1. Is the repo currently triaging plugin PRs? Latest commit on main is 2025-08-19 — can defer effort if there's a window when contributions are easier to review.
2. Would paladin-trust and safeagent work as orthogonal plugins (different scope), or would maintainers prefer one consolidated proposal? Either's fine; want to defer the architecture call to whoever is closest to plugin-tree composition.