BlueAgent skill v2 — 70 tools, Builder OS for Base

[madebyshun]

Updates BlueAgent skill from 31 tools (quantum series, outdated) to 70 live x402 tools on Bankr. New positioning: The Builder OS for Base. Correct Base URL and pricing.

[saltoriousSIG]

There are a few security issues that need to be addressed before merging. Let me know if you have any questions!

High: avoid showing @blueagent/sdk initialized with process.env.WALLET_KEY; this normalizes handing a hot private key to a third-party SDK, so this should use Bankr-managed signing, WalletConnect, KMS/WaaS, or clearly mark raw keys as dev-only.

High: add guardrails around claude mcp add blueagent https://blueagent.dev/api/mcp; remote MCPs can inject tools/instructions or exfiltrate context, so the skill should say to only install trusted MCPs, review permissions, keep secrets/workspace data out, and require confirmation for paid or wallet actions.

High: add x402 spend controls to the paid-call examples; agents can be prompt-injected into repeated paid calls, so the skill should recommend verifying host/recipient/price and setting per-session/day caps plus confirmation above a small threshold.

High: treat blue-deploy and blue-compose outputs as untrusted; remotely generated scripts/workflows can include malicious calldata, approvals, or shell commands, so the skill should require inspection, sandbox/testnet runs, address verification, and human approval before mainnet execution.

Medium: clarify that token signals, whale-copy signals, DeFi opportunities, and risk-gate are advisory only; they should not be treated as approval to trade or interact with contracts without simulation, slippage/allowance limits, address checks, and user confirmation.

Medium: add basic safety language to the $BLUEAGENT credits section; users/agents should verify the token contract independently, avoid unlimited approvals, use a limited wallet, and not buy/approve tokens automatically from the skill text alone.

[madebyshun]

thanks for the detailed review

addressed all 6 issues in the latest commit (e8a52d1):

HIGH

• SDK: marked WALLET_KEY as dev-only with note to use
  Bankr-managed signing / WalletConnect / KMS in production
• MCP: added trust guardrails before the install command
  (trusted sources only, review permissions, keep secrets out,
  confirm paid/wallet actions)
• x402: added Spend Controls section — verify host/recipient/price,
  set per-session cap ($1–5 USDC), confirm above $0.50
• blue-deploy/compose: added untrusted output warning —
  inspect before execution, testnet first, verify addresses,
  human approval before mainnet

MEDIUM

• Signals: added Advisory Disclaimer after Intelligence section —
  not financial advice, simulate/slippage/verify/confirm required
• $BLUEAGENT: added token safety note — verify contract independently,
  no unlimited approvals, limited wallet, no auto-buy from skill text

let me know if anything else needs adjusting!