Bundie Sol

Bundie Solana — Agent Skills

The Bundie Solana CLI (bundie-sol, npm: @bundie/sol-cli) is the agent-facing surface for composing strategies, opening prediction markets, and resolving outcomes on Solana devnet.

Humans back strategies and take prediction positions through the webapp (wrapped as a Seeker TWA). Agents do the rest — composition, market-making, resolution — through this CLI.

Two usage modes, same CLI:

• Mode 1 — Human-triggered. A human prompts Claude Desktop, Cursor, or any MCP-capable assistant with the skills file available. The agent reads this file, invokes bundie-sol through its shell tool, and returns results. The human is in the loop for intent and approval.
• Mode 2 — Autonomous. A cron job, ZerePy worker, elizaOS action, long-running headless claude -p loop, or custom Rust daemon operates on its own triggers (new strategy crossing TVL threshold, time elapsed, volatility below X) and invokes bundie-sol without a per-action human prompt.

The demo leads with Mode 2 (two autonomous agents in split-screen terminals) because it lands the "agents do the hard work" positioning hardest. First-time users typically start in Mode 1 via Claude Desktop.

Archetypes

Three agent roles exist on the protocol. Any wallet can play any combination, subject to the creator self-exclusion rule (a strategy's creator cannot take positions on markets opened against their own strategy).

Humans on the webapp handle the two consumption roles: Backer (buy-shares) and Predictor (predict). The CLI supports those commands too for autonomous backer / predictor agents, but they are not the demo hero.

Prerequisites

1. Install

npm install -g @bundie/sol-cli
# or run ad-hoc
npx @bundie/sol-cli --help

Binary: bundie-sol.

2. Wallet keypair

solana-keygen new --outfile ~/.config/solana/id.json

Default path is ~/.config/solana/id.json; override with --keypair <path>.

3. Devnet SOL (for tx fees)

solana airdrop 2 --url devnet

4. Devnet USDC (for deposits and predictions)

Devnet USDC mint: 4zMMC9srt5Ri5X14GAgXhaHii3GnPAEERYPJgZJDncDU

1. Create the ATA if you don't have one:

   spl-token create-account 4zMMC9srt5Ri5X14GAgXhaHii3GnPAEERYPJgZJDncDU --url devnet
   

2. Faucet USDC at https://faucet.circle.com (select "Devnet").

Minimum: 1 USDC per strategy deposit; any positive amount for predictions.

Global CLI flags

Archetype flows

Strategy-creator

Composes DeFi strategies. The on-chain program now wraps 33 Beethoven protocols on devnet (browse them at https://app.solana.bundie.fi/protocols):

• Lending / vaults (9): Kamino, Jupiter Earn, Drift, MarginFi, Marinade, Solend, SPL Stake Pool (covers BlazeStake / Jito-SOL / custom LSTs), Drift Vaults, Meteora Dynamic Vaults
• DEX swaps (22): Orca Whirlpools, Raydium CPMM/CLMM/AMM v4, Meteora DLMM/DAMM v2, Phoenix, OpenBook v2, Manifest, Aldrin v1+v2, Heaven, Hadron, Gamma, Futarchy, Omnipair, Perena, Scale AMM/VMM, SolFi v1+v2
• Perpetuals (2): Mango v4, Zeta Markets

Pass --protocol <name> (e.g. --protocol marinade, --protocol orca-whirlpools, --protocol mango-perps) to target a specific one. Default remains --protocol kamino for backwards compatibility.

A strategy can compose multiple protocols at once — issue subsequent rebalance calls (CLI command pending) to deposit / swap / open perps inside the strategy's wallet PDA.

Mode 1 — Human-triggered in Claude Desktop

User prompt:

"Compose a conservative USDC lending strategy on Kamino with a 10% performance fee. Seed it with 100 USDC."

Expected agent behavior:

1. Agent reads this skills file.
2. Agent runs:

   bundie-sol create-strategy \
     --name "Conservative Stable Yield" \
     --protocol kamino \
     --fee-bps 1000 \
     --deposit 100 \
     --min-deposit 1
   

3. Agent parses stdout for the strategy: address and returns it to the user.

User prompt with less context:

"Launch a yield strategy for me."

Expected agent behavior: ask one clarifying question (protocol? fee? name?) or pick sensible defaults (Kamino, 10% fee, agent-generated name) and confirm before calling create-strategy.

Mode 2 — Autonomous loop

A scheduled worker that composes a new strategy whenever observed devnet Kamino supply APY drops below a threshold:

#!/usr/bin/env bash
# cron: every 6 hours
APY=$(curl -s https://api.kamino.fi/devnet/reserves/usdc | jq -r .supply_apy)
if (( $(echo "$APY < 5.0" | bc -l) )); then
  bundie-sol create-strategy \
    --name "Kamino Recovery $(date +%s)" \
    --protocol kamino \
    --fee-bps 1000 \
    --deposit 50
fi

Or as a long-running claude -p loop that reads this file, observes its own triggers, and acts:

claude -p --model sonnet \
  --append-system "You are a strategy-creator agent. Read SKILLS.md. Every 6 hours, check Kamino APY via the RPC and compose a new strategy if conditions warrant. Only call bundie-sol create-strategy."

Hard constraint

The wallet you create with becomes the strategy's authority. That wallet is then blocked from taking positions on any prediction market opened against this strategy (on-chain address-equality check). This is why the market-maker archetype is a separate wallet by design.

Market-maker

Opens LS-LMSR prediction markets on strategies that other agents have created. Seeds initial liquidity so the market has depth from trade one.

Must be a different wallet from the strategy's creator. The create-market instruction itself does not enforce this, but subsequent predict / buy_shares calls on the market will revert with CreatorCannotPredictOnOwnStrategy if the market-maker is the strategy creator.

Mode 1 — Human-triggered

User prompt:

"Open a market on whether strategy 93amC4... exceeds 10% APY by June 1."

Expected agent behavior:

1. Compute resolution-days from the target date.
2. Run:

   bundie-sol create-market \
     --strategy 93amC41qp6YfTQwsugjgRG21eUtmbqiCx4VJjzE2xZYF \
     --question "Will Kamino Seed Strategy exceed 10% APY by June 1?" \
     --threshold-bps 1000 \
     --resolution-days 40 \
     --initial-subsidy 10 \
     --fee-bps 100
   

3. Return the new market: address to the user.

Mode 2 — Autonomous watch-loop

Poll list-strategies on an interval and open a market whenever a new strategy crosses $500 in TVL:

#!/usr/bin/env bash
# cron: every 1 hour
SEEN_FILE="/var/cache/bundie-markets-seen.txt"
touch "$SEEN_FILE"

bundie-sol list-strategies | awk '$3 >= 500 { print $1 }' | while read STRAT; do
  if ! grep -q "$STRAT" "$SEEN_FILE"; then
    bundie-sol create-market \
      --strategy "$STRAT" \
      --question "Will this strategy exceed 10% APY in 30 days?" \
      --threshold-bps 1000 \
      --resolution-days 30 \
      --initial-subsidy 10
    echo "$STRAT" >> "$SEEN_FILE"
  fi
done

Choosing parameters

• --threshold-bps — The APY target. 1000 = 10%. Pick a level where the strategy's actual performance is plausibly either side.
• --resolution-days — How long backers and predictors have to trade before settlement. 7–30 days is typical.
• --initial-subsidy — USDC put into the LS-LMSR curve at creation. More subsidy = tighter spreads early, but locks capital. 10–50 USDC is typical.
• --fee-bps — Trading fee. 100 bps = 1%. Competitive range 50–200.

Resolver

Permissionless. Anyone can call resolve-market once market.resolution_slot has been reached. The program reads the strategy's NavOracle PDA (seeds ["nav", strategy]) and compares the TWAP value against the market's threshold. No external oracle in the resolution path.

Mode 2 — Keeper cron

#!/usr/bin/env bash
# cron: every 10 minutes
# Resolve every market past its resolution slot. Markets already resolved
# will simply error out and we move on.

bundie-sol list-markets 2>/dev/null | awk '$STATUS == "active" && $RESOLUTION_SLOT <= $CURRENT_SLOT { print $MARKET }' | while read M; do
  bundie-sol resolve-market --market "$M" || true
done

list-markets is not yet a CLI command (post-hackathon); for now, keepers track markets via indexing events or a seeded address list.

Mode 1 — Manual resolution

User prompt:

"Resolve market 2XfBNn... if it's past its resolution slot."

Expected agent behavior:

bundie-sol resolve-market --market 2XfBNnZ5YEH23tB4QdCNzXu2bJmeY5WwmMuikGgZP8wu

Agent returns the decoded outcome (YES / NO) from stdout.

Participation commands (for autonomous backer / predictor agents)

These are reference — the demo and pitch show humans doing these via the webapp. Autonomous agents that want to express views on-chain still use the CLI.

buy-shares (Backer)

Invest USDC into an existing strategy.

bundie-sol buy-shares \
  --strategy <pubkey> \
  --amount <usdc>

Output includes the tx signature.

predict (Predictor)

Buy YES or NO shares on an LS-LMSR market.

bundie-sol predict \
  --market <pubkey> \
  --side <yes|no> \
  --amount <usdc>

Blocked (by CreatorCannotPredictOnOwnStrategy) if the calling wallet is the strategy's on-chain authority. This is the v5 spec's first-party address-equality check.

Full command reference

create-strategy

bundie-sol create-strategy \
  --name <name> \
  [--protocol <kamino|marginfi|jupiter|<pubkey>>] \
  [--fee-bps <bps>] \
  [--deposit <usdc>] \
  [--min-deposit <usdc>] \
  [--usdc-mint <pubkey>]

Sample output:

Creating strategy "Alpha Momentum" on devnet...
  wallet: <your-pubkey>

  create_strategy: <tx-sig>
  strategy: <strategy-address>
  mint:     <share-mint-address>

✓ Strategy created successfully.

create-market

bundie-sol create-market \
  --strategy <pubkey> \
  --question <text> \
  --threshold-bps <bps> \
  [--resolution-days <days>] \
  [--initial-subsidy <usdc>] \
  [--fee-bps <bps>] \
  [--strategy-b <pubkey>]

resolve-market

bundie-sol resolve-market --market <pubkey>

Permissionless. Reverts if market is not Active or resolution slot not reached.

list-strategies

bundie-sol list-strategies

Read-only. Returns a formatted table of every Strategy account on devnet:

address                                       | name                       | nav (USDC) | shares    | investors | fee
─────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
93amC41qp6YfTQwsugjgRG21eUtmbqiCx4VJjzE2xZYF  | Kamino Seed Strategy       |   1.043210 |   1000.00 |         3 | 500 bps

(1 strategies)

buy-shares

See Participation commands above.

predict

See Participation commands above.

nav

bundie-sol nav --strategy <pubkey>

Read-only. Prints the strategy's current NAV, share price, TVL, investor count, and estimated APY.

Demo sequences

Two-agent creation-side walkthrough (Scene 2)

Terminal 1 (strategy-creator.sol):

bundie-sol --keypair ./strategy-creator.json create-strategy \
  --name "Stable Compounder" --protocol kamino --fee-bps 1000 --deposit 50
# Outputs: strategy: <SA>

Terminal 2 (market-maker.sol, different wallet):

# Watch for the new strategy to appear in list-strategies
bundie-sol list-strategies | grep "Stable Compounder"
# Then open a market on it
bundie-sol --keypair ./market-maker.json create-market \
  --strategy <SA> \
  --question "Will Stable Compounder exceed 8% APY in 7 days?" \
  --threshold-bps 800 \
  --resolution-days 7 \
  --initial-subsidy 10

The strategy-creator wallet is now blocked from taking positions on this market. The market-maker wallet can trade freely.

End-to-end loop (for testing + demo seeding)

# 1. Create a strategy
bundie-sol create-strategy --name "E2E Test" --fee-bps 1000 --deposit 10
# -> <strategy>

# 2. Back it from a second wallet
bundie-sol --keypair ./backer.json buy-shares --strategy <strategy> --amount 25

# 3. Open a market from a third wallet
bundie-sol --keypair ./maker.json create-market \
  --strategy <strategy> --question "..." --threshold-bps 500 --resolution-days 1

# 4. Predict from a fourth wallet
bundie-sol --keypair ./predictor.json predict \
  --market <market> --side yes --amount 5

# 5. (one day later, past resolution slot) Resolve
bundie-sol resolve-market --market <market>

Error cases

All commands print errors to stderr and exit with code 1.

Seeded addresses (devnet)

After the 2026-04-23 vanity-address migration, all earlier seeded strategies and markets were created against the old Y13k* programs and are now orphaned. Run bundie-sol list-strategies for a fresh list of strategies on the new Bun4* programs.

Program addresses (devnet)

The prediction-market IDL is uploaded on chain (anchor IDL PDA 9PTHYfhdxmQGunPLsUheQbRDka8UnzTnfkCn7GA4DDbf); strategy-token is Pinocchio and exposes its layout via the @bundie/sol-cli source rather than an IDL.

The previous Y13k* program IDs are abandoned (rent-locked, no upgrade path). Any client still pinned to those returns empty results.

Devnet resources

Related surfaces

• Webapp (app.solana.bundie.fi) — human surface. Backers and predictors use this; it wraps as a Seeker TWA for dApp Store distribution. Does not expose strategy or market creation.
• EVM sibling (@bundie/evm-mcp) — Bundie's EVM yield routing MCP, pre-existing. Different product (vault routing, not strategy composition). Connected via mcp.bundie.fi/evm.
• Solana MCP — deferred. CLI + this skills file cover both Mode 1 and Mode 2 today; MCP is a UX polish for Claude Desktop users, not a new capability.

License

MIT.