Add AgentRisk plugin

skills/base-mcp/plugins/agentrisk.md

@@ -0,0 +1,62 @@
+---
+title: "AgentRisk Plugin"
+description: "Pre-trade honeypot/tax risk scoring for Base tokens via x402-paid HTTP API, checked before a swap."
+tags: [dex, swap, trading, discovery, ai-agents]
+name: agentrisk
+version: 0.1.0
+integration: http-api
+chains: [base]
+requires:
+  shell: none
+  allowlist: [agentrisk.dev]
+  externalMcp: null
+  cliPackage: null
+auth: none
+risk: []
+---
+
+# AgentRisk Plugin
+
+> [!IMPORTANT]
+> Run Base MCP onboarding first (see SKILL.md). No additional session setup required — each call is paid independently via x402.
+
+## Overview
+AgentRisk is a self-hosted pre-trade risk API for Base Mainnet. It combines direct on-chain checks (ownership, pausability, LP-burn), GoPlus Security data, and DexScreener liquidity data into a single risk score. It returns an advisory JSON decision (`shouldExecute`, `riskScore`, `riskLevel`, `verdict`, `reasons`) — it does not build or submit any calldata itself.
+
+Beyond the base risk score, AgentRisk also flags:
+- **Deployer wallet freshness** — newly-created wallets used for one-off token launches
+- **Brand impersonation** — tokens named after known companies (Apple, Google, Meta, etc.)
+- **Data source disagreement** — cases where third-party APIs and our own on-chain checks conflict
+- **Human-readable verdict** — one plain-English sentence summarizing the decision
+- Repeat scans within 5 minutes are served from cache in under 1ms
+
+## Surface Routing
+| Capability | Harness with HTTP/shell (Claude Code, Codex, Cursor) | Chat-only (Claude.ai, ChatGPT) |
+| --- | --- | --- |
+| Check token risk (read) | Call the endpoint directly via the harness HTTP tool; harness's x402-aware client handles the 402 payment automatically. | Call via Base MCP `web_request` (`agentrisk.dev` is on the allowlist); Base MCP's native x402 payment capability handles the 402 automatically. |
+
+## Endpoints
+- `GET https://agentrisk.dev/scan?token=<CONTRACT_ADDRESS>` — x402-gated (0.15 USDC, Base mainnet, `eip155:8453`). Returns the full risk report JSON (`risk_score`, `risk_level`, `findings`, `is_honeypot`, `liquidity_usd`, etc.).
+- `POST https://agentrisk.dev/mcp/tools/check_token_risk` — body `{"token_address": "<CONTRACT_ADDRESS>"}`. Same underlying analysis, x402-gated identically. Returns `{"status": "success", "decision": {"shouldExecute": bool, "riskScore": int, "riskLevel": string, "reasons": [string]}}`.
+
+## Orchestration
+1. User asks to buy/swap an unknown or untrusted token on Base.
+2. Agent calls one of the endpoints above with the token contract address.
+3. Endpoint responds `402 Payment Required` with x402 payment requirements (0.15 USDC, `eip155:8453`, exact scheme).
+4. The agent's x402-aware client (harness or Base MCP) signs and submits the USDC payment authorization, then retries the request with the payment header.
+5. AgentRisk verifies and settles the payment through its own self-hosted x402 facilitator, then runs the analysis and returns the decision.
+6. Agent reads `shouldExecute`. If `false`, do not proceed — explain `reasons` to the user. If `true`, the agent may separately call Base MCP's own `swap` tool to execute the trade.
+
+## Submission
+`none`. AgentRisk's endpoints return an advisory JSON decision only — they do not produce calldata for `send_calls`. The only value that reaches Base MCP directly is the x402 USDC payment itself, handled by Base MCP's native x402 payment capability, not by this plugin.
+
+## Example Prompts
+1. "Buy $50 of token 0xABC... on Base, I'm not sure if it's safe." → Agent calls `GET /scan?token=0xABC...`, pays the x402 fee, reads `shouldExecute`. If `false`, explains the reasons and stops. If `true`, calls Base MCP `swap`.
+2. "Is this new token a honeypot before I ape in?" → Same flow; agent surfaces `reasons` regardless of the verdict so the user sees the evidence.
+3. Same request on a chat-only surface (Claude.ai) → identical flow, routed through Base MCP `web_request` instead of a direct HTTP call.
+
+## Notes
+- Price: 0.15 USDC per call, charged identically on both endpoints.
+- Facilitator: self-hosted (not Coinbase CDP) — verification and settlement happen on Base mainnet directly.
+- Only Base Mainnet (`eip155:8453`) is supported; no testnet.
+- Source: https://github.com/Neurobyteio/agentrisk