solana-wallet-aware — правило для Cursor: защита кошелька

---
description: "Guidelines for writing Solana-native code with wallet-security awareness, isolated signer subprocesses, MEV defense, oracle gates, and transaction safety checks"
globs: **/*.{ts,tsx,js,jsx,py,rs}
alwaysApply: false
---
# Solana Wallet-Aware Coding

When writing Solana on-chain or off-chain code, apply these wallet-safety and transaction-safety rules.

## Wallet architecture

- Never store a raw private key in `.env`, config files, or source. Encrypt with a passphrase-derived key (HKDF-SHA256 + AES-256-GCM, or `eth-account`'s scrypt V3 keystore, or `libsodium`'s sealed-box).
- Use a three-tier wallet split at $1k+ scale: **hot** (bot-signing, ≤10% of AUM), **warm** (manual-top-up buffer on founder phone, ~30%), **cold** (hardware wallet or Squads 2-of-2 multisig, ~60%, untouchable ≥6 months).
- Treat the hot wallet as burnable. Any key that ever touched a `.env` file on a dev machine is compromised forever.
- Isolate the signer in a subprocess with only two capabilities: (a) receive a pre-built transaction over a local Unix socket / stdin, (b) return a signature. No network access, no program-scope escalation, explicit allowlist of program IDs.

## MEV defense on Solana

- Never broadcast swaps to the public mempool. Always use Jito bundles with a per-bundle tip (start at 10k lamports, scale with expected profit).
- Add an oracle gate: reject a trade if Jupiter's quoted price is > 0.5% off Pyth's spot price. Update threshold dynamically with 1-minute realized volatility.
- Maintain an illiquidity blocklist: skip any token where the deepest pool has < $1M TVL (use GeckoTerminal or Birdeye API to check).
- For limit-style orders, prefer Jupiter's limit-order program (built-in MEV protection) over composing your own.

## Program-ID allowlist pattern

When building a signer, hard-code the list of program IDs your bot may invoke. Reject any transaction whose instructions touch a program outside the allowlist. At minimum for a Solana trading bot:

```
const ALLOWED_PROGRAMS = new Set([
  "JUP6LkbZbjS1jKKwapdHNy74zcZ3tLUZoi5QNyVTaV4",  // Jupiter v6
  "opnb2LAfJYbRMAHHvqjCwQxanZn7ReEHp1k81EohpZb",  // OpenBook v2
  "TokenkegQfeZyiNwAJbNbGKPFXCWuBvf9Ss623VQ5DA",  // SPL Token
  "ComputeBudget111111111111111111111111111111",  // Compute budget
  // add your DEX IDs here — NEVER include unknown programs
]);
```

## Transaction-safety invariants

Before signing ANY transaction:
- Deserialize + inspect every instruction. No opaque "signTransaction(bytes)" without parsing.
- Enforce a max SOL outflow per transaction AND per rolling 24h window (spend-cap circuit breaker).
- Require a freshness check on the blockhash (slot age < 150 or transaction will likely expire).
- Compute budget ≤ 200k CU default; require explicit opt-in for higher.

## Simulation gate

Do not move wallet-automation code from simulation to live signing until the user has explicitly approved it and the implementation has passed fault-injection tests for signer failure, RPC failure, stale blockhashes, rejected allowlist entries, and spend-cap enforcement.

## Operational safety

- Keep deployment credentials, wallet keys, RPC credentials, and monitoring tokens out of source control.
- Add health checks for signer availability, stale blockhashes, RPC error rate, and transaction failure rate.
- Require an explicit manual approval path before raising spend caps, program allowlists, or compute-unit limits.