"Vibe-Trading: Your Personal Trading Agent"
Python ai-agentalgorithmic-tradingbacktestingfintechllmmcpmulti-agentpythonquantitative-financetrading
"""Policy helpers for finance research goals."""
from __future__ import annotations
import re
_EXECUTION_PATTERNS = (
re.compile(r"\b(place|submit|execute|send)\b.{0,40}\b(order|trade)\b", re.I),
# No bare asset noun (shares/contracts/btc/...) here: this is a research
# product whose whole purpose is evaluating whether to buy/sell/short a
# security, so "buy NVDA shares" alone in a research objective is the
# normal case, not an execution request. Only fire on an actual
# execution-imminent cue alongside the buy/sell verb.
# "long-term" / "short-term" (and "long term" / "short term") are horizons.
# A hyphen or space is a word boundary, so "update the long-term thesis
# immediately" used to match long ... immediately and was rejected as an order.
re.compile(
r"\b(?:buy|sell|short(?![\s-]*term\b)|long(?![\s-]*term\b))\b.{0,40}\b(?:now|immediately|right away|market order|limit order)\b",
re.I,
),
# Same cue, reversed order ("immediately buy TSLA", "right away sell
# AAPL"): the pattern above only fires when the verb comes first.
re.compile(
r"\b(?:now|immediately|right away)\b[\s,:-]+(?:please\s+)?(?:buy|sell|short(?![\s-]*term\b)|long(?![\s-]*term\b))\b",
re.I,
),
re.compile(r"(下单|市价单|限价单|马上买|立即买|现在买|马上卖|立即卖|现在卖)"),
# What the asset-noun list above used to catch, without its false
# positives: an objective that IS an order -- it opens with the verb
# ("Buy NVDA", "Please sell my TSLA") or names a quantity right after it
# ("then buy 100 shares", "short 2 ES", "买入100股"). A research goal
# frames the verb ("whether to buy", "should sell", "a short thesis").
# A leading "short" is left out: "Short interest ...", "Short-term ..."
# open research goals, so a bare "Short TSLA" passes.
re.compile(r"^\s*(?:please\s+)?(?:buy|sell)\b", re.I),
re.compile(r"\b(?:buy|sell|short)\s+\d", re.I),
re.compile(r"(?:买入|卖出|做空|买|卖)\s*\d+(?:\.\d+)?\s*(?:股|手|张|份|个)"),
)
def normalize_required_text(value: str, field_name: str) -> str:
"""Strip and validate a required text field.
Args:
value: User supplied text.
field_name: Field name for the error message.
Returns:
The stripped value.
Raises:
ValueError: If the stripped value is empty.
"""
text = value.strip()
if not text:
raise ValueError(f"{field_name} cannot be empty")
return text
def reject_live_execution_objective(objective: str) -> None:
"""Reject direct live-trading or order-execution goal text.
Args:
objective: Research goal objective.
Raises:
ValueError: If the objective looks like an execution request.
"""
text = objective.strip()
for pattern in _EXECUTION_PATTERNS:
if pattern.search(text):
raise ValueError("live trading or execution goals are not supported")