pump-fun-strategy

Declarative strategies. One JSON spec → live run or backtest, sharing the same evaluator so they cannot drift.

Spec

{
  "scan":   { "kind": "newTokens", "limit": 20 },
  "filters": [
    "holders.total > 50",
    "holders.topHolderPct < 20",
    "creator.rugCount == 0"
  ],
  "entry": { "side": "buy", "amountSol": 0.05 },
  "exit": [
    { "if": "position.pnlPct > 50",       "do": { "side": "sell", "percent": 50 } },
    { "if": "holders.topHolderPct > 40",  "do": { "side": "sell", "percent": 100 } },
    { "if": "position.ageSec > 1800",     "do": { "side": "sell", "percent": 100 } }
  ],
  "caps": { "sessionSpendCapSol": 0.5, "perTradeSol": 0.05, "maxOpenPositions": 5 }
}

Predicate language

<lhs> <op> <rhs> where lhs is a dotted path into the runtime view, op is one of > >= < <= == !=, and rhs is a number (trailing % allowed for readability). The view exposes holders.*, creator.*, curve.*, token.*, position.*, trades.* — see dsl.js:buildView.

Tools

Tool	Use it for
validateStrategy	Parse + type-check before running
runStrategy	Live (with simulate: true for dry-run)
backtestStrategy	Replay against historical trades from getTokenTrades

Why this matters

Compose-skill loops are imperative and hard to share. The DSL turns a strategy into a portable artifact you can:

• Backtest before risking SOL
• Hand to another agent to copy-trade
• Diff between versions ("v2 added the 30-min stop")
• Publish on the agent passport as a signed track record