agiprolabs/claude-trading-skills/jito-bundles

Claude Code and agent skill by agiprolabs/claude-trading-skills

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Jito bundle submission for MEV protection on Solana — bundle building, tip strategies, block engine endpoints, and landing rate optimization

Install

npx vskill@latest install agiprolabs/claude-trading-skills/jito-bundles

SKILL.md

Jito Bundle Submission for Solana

Jito bundles allow you to submit up to 5 Solana transactions that execute atomically — either all land in the same slot or none do. This is the primary mechanism for MEV protection and competitive transaction execution on Solana. Approximately 85%+ of Solana validators run the Jito-modified client, making bundles the standard for reliable, front-run-resistant execution.

EXECUTION SKILL — SAFETY WARNING: Submitting bundles spends real SOL on tips. Always test with --demo mode first. Never submit bundles with real funds without explicit confirmation. Default to simulation/dry-run in all scripts and examples.

When to Use Bundles

ScenarioUse Bundle?Why
Swap on illiquid tokenYesPrevents sandwich attacks
Multi-step arbitrageYesAtomic execution prevents partial fills
LiquidationYesCompetitive — tip determines priority
Simple SOL transferNoPriority fees are cheaper and sufficient
Time-insensitive swapMaybeBundles cost tips; priority fees may suffice
NFT mint / competitive actionYesGuarantees ordering within the slot

Core Concepts

Bundle Anatomy

A Jito bundle is a JSON-RPC request containing 1-5 base58-encoded signed transactions. The transactions execute sequentially and atomically within a single slot.

Bundle = [Tx1, Tx2, ..., TxN] (N <= 5)
- All transactions must be signed
- Transactions execute in order: Tx1 → Tx2 → ... → TxN
- If ANY transaction fails, the ENTIRE bundle is dropped
- The tip instruction goes in the LAST transaction (last instruction)
- Bundle has ~2 slots (~800ms) to land before expiry
Tip Mechanism

Tips are SOL transfers to one of Jito's 8 tip accounts. The tip incentivizes validators to include your bundle.

python
# Tip is a standard SOL transfer instruction
tip_instruction = transfer(
from_pubkey=your_wallet,
to_pubkey=tip_account, # One of 8 Jito tip accounts
lamports=tip_amount # Tip in lamports (1 SOL = 1e9 lamports)
)
# Add as the LAST instruction of the LAST transaction in the bundle

Tip accounts are fetched dynamically via getTipAccounts. Rotate through them to distribute load.

Block Engine Endpoints

Jito operates geographically distributed block engines. Choose the one closest to your infrastructure:

RegionEndpoint
New Yorkhttps://mainnet.block-engine.jito.wtf
Amsterdamhttps://amsterdam.block-engine.jito.wtf
Frankfurthttps://frankfurt.block-engine.jito.wtf
Tokyohttps://tokyo.block-engine.jito.wtf

All endpoints accept JSON-RPC over HTTPS on port 443. The /api/v1/bundles path handles bundle operations.

API Methods

sendBundle

Submit a bundle of up to 5 transactions.

python
import httpx
BLOCK_ENGINE = "https://mainnet.block-engine.jito.wtf"
payload = {
"jsonrpc": "2.0",
"id": 1,
"method": "sendBundle",
"params": [
[tx1_base58, tx2_base58], # List of base58-encoded signed txs
]
}
resp = httpx.post(f"{BLOCK_ENGINE}/api/v1/bundles", json=payload)
data = resp.json()
bundle_id = data["result"] # UUID string
getBundleStatuses

Check the landing status of submitted bundles (up to 5 bundle IDs per request).

python
payload = {
"jsonrpc": "2.0",
"id": 1,
"method": "getBundleStatuses",
"params": [[bundle_id]]
}
resp = httpx.post(f"{BLOCK_ENGINE}/api/v1/bundles", json=payload)
statuses = resp.json()["result"]["value"]
# Each status: {bundle_id, status, slot, transactions: [{signature, ...}]}
# status: "Invalid", "Pending", "Failed", "Landed"
getTipAccounts

Fetch the current list of Jito tip accounts.

python
payload = {
"jsonrpc": "2.0",
"id": 1,
"method": "getTipAccounts",
"params": []
}
resp = httpx.post(f"{BLOCK_ENGINE}/api/v1/bundles", json=payload)
tip_accounts = resp.json()["result"] # List of 8 base58 pubkeys
getInflightBundleStatuses

Check status of bundles that haven't landed yet (in-flight).

python
payload = {
"jsonrpc": "2.0",
"id": 1,
"method": "getInflightBundleStatuses",
"params": [[bundle_id]]
}
resp = httpx.post(f"{BLOCK_ENGINE}/api/v1/bundles", json=payload)
# status: "Pending", "Failed", "Landed"

Bundle Construction Pattern

A typical bundle for a protected swap:

python
from solders.transaction import VersionedTransaction
from solders.message import MessageV0
from solders.instruction import Instruction
from solders.system_program import transfer, TransferParams
from solders.pubkey import Pubkey
import random
def build_protected_swap_bundle(
swap_ix: Instruction,
payer: Pubkey,
tip_lamports: int,
tip_accounts: list[str],
recent_blockhash: str,
) -> list[VersionedTransaction]:
"""Build a 1-tx bundle: swap + tip in the same transaction.
For simple swaps, a single-transaction bundle is sufficient.
The tip instruction is appended as the last instruction.
"""
# Pick a random tip account
tip_account = Pubkey.from_string(random.choice(tip_accounts))
# Tip instruction
tip_ix = transfer(TransferParams(
from_pubkey=payer,
to_pubkey=tip_account,
lamports=tip_lamports,
))
# Build transaction with swap + tip
msg = MessageV0.try_compile(
payer=payer,
instructions=[swap_ix, tip_ix],
address_lookup_table_accounts=[],
recent_blockhash=recent_blockhash,
)
tx = VersionedTransaction(msg, [keypair])
return [tx]

Tip Sizing Guide

ScenarioTip Range (lamports)Tip Range (SOL)
Normal swap (low urgency)1,000 - 10,0000.000001 - 0.00001
Normal swap (standard)10,000 - 50,0000.00001 - 0.00005
Competitive action (arb, liquidation)50,000 - 500,0000.00005 - 0.0005
Highly competitive (NFT mint, MEV)500,000 - 5,000,0000.0005 - 0.005
Emergency (must land this slot)5,000,000+0.005+

Dynamic tip calculation based on recent tip levels:

python
def calculate_dynamic_tip(
base_tip: int = 10_000,
urgency_multiplier: float = 1.0,
recent_tip_percentile_50: int = 15_000,
) -> int:
"""Calculate tip based on urgency and recent network tips.
Args:
base_tip: Minimum tip in lamports.
urgency_multiplier: 1.0 = normal, 2.0 = urgent, 5.0 = critical.
recent_tip_percentile_50: Median tip from recent bundles.
Returns:
Tip amount in lamports.
"""
dynamic_tip = max(base_tip, int(recent_tip_percentile_50 * urgency_multiplier))
# Cap at 0.01 SOL to prevent accidents
return min(dynamic_tip, 10_000_000)

Common Errors and Fixes

ErrorCauseFix
Bundle dropped (slot expired)Bundle didn't land within 2 slotsRetry with fresh blockhash; consider higher tip
Transaction simulation failedA tx in the bundle would fail on-chainSimulate each tx individually to find the failing one
Bundle already processedDuplicate bundle IDExpected on retry; check status instead
Rate limitedToo many requests to block engineBack off; rotate between block engine endpoints
Invalid transactionMalformed or unsigned transactionVerify all txs are signed and base58-encoded
Blockhash not foundStale blockhashUse getLatestBlockhash with finalized commitment

Landing Rate Optimization

Strategies to maximize bundle landing probability:

  1. Multi-region submission: Send the same bundle to multiple block engines simultaneously. The first to reach the current leader wins.
  1. Fresh blockhash: Use getLatestBlockhash with confirmed commitment immediately before building. Stale blockhashes are the #1 cause of dropped bundles.
  1. Retry with backoff: If a bundle doesn't land within 2-3 seconds, rebuild with a fresh blockhash and resubmit. Do NOT resubmit with the same blockhash.
  1. Adequate tipping: Under-tipped bundles are deprioritized. Monitor the network's tip distribution and tip at or above the 50th percentile for your urgency level.
  1. Minimal bundle size: Fewer transactions = less simulation time = higher landing rate. Use single-transaction bundles when possible.
python
async def submit_with_retry(
bundle_txs: list[str],
endpoints: list[str],
max_retries: int = 3,
) -> str | None:
"""Submit bundle to multiple endpoints with retry logic.
Returns bundle_id if submitted, None if all retries exhausted.
"""
for attempt in range(max_retries):
# Submit to all endpoints in parallel
async with httpx.AsyncClient() as client:
tasks = [
client.post(
f"{ep}/api/v1/bundles",
json={
"jsonrpc": "2.0", "id": 1,
"method": "sendBundle",
"params": [bundle_txs],
},
timeout=5.0,
)
for ep in endpoints
]
# Process first successful response
for resp in asyncio.as_completed(tasks):
result = (await resp).json()
if "result" in result:
return result["result"]
# Wait before retry with fresh blockhash
await asyncio.sleep(0.5 * (attempt + 1))
return None

Safety Checklist (Execution)

Before submitting any bundle with real funds:

  • [ ] Simulated all transactions individually via simulateTransaction
  • [ ] Verified tip amount is reasonable (not accidentally SOL instead of lamports)
  • [ ] Confirmed blockhash is fresh (< 60 seconds old)
  • [ ] Verified all transactions are properly signed
  • [ ] Checked wallet balance covers all transaction costs + tip
  • [ ] Tested with devnet or --demo mode first
  • [ ] Set maximum tip cap to prevent accidental overpayment

Files

References
  • references/bundle_api.md — Complete JSON-RPC API reference with request/response schemas and error codes
  • references/tip_strategies.md — Tip calculation strategies, dynamic tipping, cost optimization
  • references/best_practices.md — Bundle construction patterns, landing rate optimization, common pitfalls
Scripts
  • scripts/build_bundle.py — Bundle construction with tip instruction; --demo mode builds but does not submit
  • scripts/check_bundle_status.py — Bundle status checking and tip account fetching; --demo mode uses mock responses

Security

  • Security scanPassed, 95 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

7-day trend score
0

Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/agiprolabs/claude-trading-skills/jito-bundles/badge)](https://verified-skill.com/skills/agiprolabs/claude-trading-skills/jito-bundles)

Something wrong with this skill or its page? Tell us.