feat: add plugin-sentinel for AI safety validation

[sentinel-seed]

Summary

Add Sentinel Safety Plugin implementing the THSP (Truth-Harm-Scope-Purpose) protocol for AI agent transaction validation.

Features

• Four-gate validation: Truth, Harm, Scope, Purpose
• Configurable thresholds: Max amounts, confirmation limits, blocklists
• LLM-ready actions: SENTINEL_VALIDATE_TRANSACTION, SENTINEL_CHECK_SAFETY, etc.
• Risk assessment: LOW/MEDIUM/HIGH/CRITICAL levels
• Pattern detection: Catches suspicious operations (drain, sweep, unlimited approvals)

Use Case

Protects AI agents from executing harmful, unauthorized, or suspicious transactions on Solana by requiring every operation to pass through safety validation before execution.

Usage

import { SolanaAgentKit } from "solana-agent-kit";
import SentinelPlugin from "@solana-agent-kit/plugin-sentinel";

const agent = new SolanaAgentKit(privateKey, rpcUrl)
  .use(SentinelPlugin);

const result = await agent.methods.validateTransaction({
  action: "transfer",
  amount: 50,
  recipient: "...",
  purpose: "Payment for services",
});

if (result.shouldProceed) {
  // Safe to execute
}

Links

• Website: https://sentinelseed.dev
• npm (standalone): @sentinelseed/solana-agent-kit

[sentinel-seed]

Hi team! Just following up on this PR. Is there anything else needed from our side for this to be reviewed?

The plugin implements the THSP protocol for transaction safety validation, which could help protect autonomous agents from executing problematic transactions.

Happy to make any adjustments if needed!