TradingAgents/tradingagents/agents/signal_detectors/pump_detection_agent.py

134 lines
4.7 KiB
Python

"""
Pump Detection Agent
Analyzes stocks for potential pump signals and pre-pump opportunities.
Integrates into the langgraph agentic architecture.
"""
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from typing import Callable
def create_pump_detection_agent(llm: Callable) -> Callable:
"""
Creates a pump detection agent node for the trading graph.
Identifies stocks likely to experience pump moves using 5 detection signals.
Args:
llm: Language model to use for analysis
Returns:
Callable agent node function that processes AgentState
"""
def pump_detection_node(state: dict) -> dict:
"""
Pump detection agent node.
Analyzes a given stock for pump probability signals.
Args:
state: Current agent state containing:
- messages: Conversation history
- company_of_interest: Stock ticker to analyze
- trade_date: Current trading date
Returns:
Updated state with pump analysis results
"""
from tradingagents.agents.utils.pump_detection_tools import (
detect_volume_spike,
detect_price_acceleration,
detect_social_sentiment_surge,
detect_oversold_bounce,
detect_catalyst_event,
calculate_pump_score,
)
# Tools available to the pump detection agent
tools = [
detect_volume_spike,
detect_price_acceleration,
detect_social_sentiment_surge,
detect_oversold_bounce,
detect_catalyst_event,
calculate_pump_score,
]
ticker = state.get("company_of_interest", "")
trade_date = state.get("trade_date", "")
system_message = (
f"You are a Pump Detection Specialist analyzing {ticker} on {trade_date}. "
"Your goal is to identify if this stock is likely to experience sudden price increases (pumps)."
"\n\n"
"**Pump Detection Analysis:**"
"\n"
"Run these detection tools in order:"
"\n"
"1. Use `detect_volume_spike` - Find abnormal trading volume (2x+ average)"
"\n"
"2. Use `detect_price_acceleration` - Check for rapid price gains"
"\n"
"3. Use `detect_social_sentiment_surge` - Scan for social media buzz"
"\n"
"4. Use `detect_oversold_bounce` - Check technical oversold setup (RSI < 30)"
"\n"
"5. Use `detect_catalyst_event` - Find upcoming catalysts"
"\n"
"6. Use `calculate_pump_score` - Combine all signals into final pump score"
"\n\n"
"**Scoring Guide:**"
"\n"
"- 70+: 🔴 VERY HIGH pump probability → Strong entry candidate"
"\n"
"- 50-69: 🟠 HIGH pump probability → Good entry with risk management"
"\n"
"- 30-49: 🟡 MODERATE → Wait for confirmation"
"\n"
"- <30: 🟢 LOW → Skip this stock"
"\n\n"
"**Final Output:**"
"\n"
"Provide a concise pump analysis with:"
"\n"
"- Pump Probability Score (0-100)"
"\n"
"- Key detected signals"
"\n"
"- Trading recommendation (BUY/WAIT/SKIP)"
"\n"
"- Risk level assessment"
)
prompt = ChatPromptTemplate.from_messages(
[
(
"system",
"You are a helpful AI assistant specialized in pump detection. "
"You have access to the following tools: {tool_names}.\n{system_message}",
),
MessagesPlaceholder(variable_name="messages"),
]
)
prompt = prompt.partial(system_message=system_message)
prompt = prompt.partial(tool_names=", ".join([tool.name for tool in tools]))
# Bind tools to the LLM
chain = prompt | llm.bind_tools(tools)
# Initialize messages if needed
if not state.get("messages"):
user_input = f"Analyze {ticker} for pump opportunities using all detection methods."
state["messages"] = [("user", user_input)]
# Invoke the chain
result = chain.invoke(state)
# Add pump analysis to state
return {
"messages": state["messages"] + [result],
"pump_report": result.content if hasattr(result, 'content') else str(result),
}
return pump_detection_node