Resisting Manipulative Bots in Memecoin Copy Trading: A Multi-Agent Approach with Chain-of-Thought Reasoning

Abstract: The launch of \$Trump coin ignited a wave in meme coin investment. Copy trading, as a strategy-agnostic approach that eliminates the need for deep trading knowledge, quickly gains widespread popularity in the meme coin market. However, copy trading is not a guarantee of profitability due to the prevalence of manipulative bots, the uncertainty of the followed wallets' future performance, and the lag in trade execution. Recently, LLMs have shown promise in financial applications by effectively understanding multi-modal data and producing explainable decisions. However, a single LLM struggles with complex, multi-faceted tasks such as asset allocation. These challenges are even more pronounced in cryptocurrency markets, where LLMs often lack sufficient domain-specific knowledge in their training data. To address these challenges, we propose an explainable multi-agent system for meme coin copy trading. Inspired by the structure of an asset management team, our system decomposes the complex task into subtasks and coordinates specialized agents to solve them collaboratively. Employing few-shot chain-of-though (CoT) prompting, each agent acquires professional meme coin trading knowledge, interprets multi-modal data, and generates explainable decisions. Using a dataset of 1,000 meme coin projects' transaction data, our empirical evaluation shows that the proposed multi-agent system outperforms both traditional machine learning models and single LLMs, achieving 73% and 70% precision in identifying high-quality meme coin projects and key opinion leader (KOL) wallets, respectively. The selected KOLs collectively generated a total profit of \$500,000 across these projects.

• The paper introduces a multi-agent LLM framework using chain-of-thought prompts to identify manipulative bots in memecoin copy trading.
• It leverages multimodal evidence from on-chain data, candlestick analysis, and social sentiment to differentiate genuine KOL wallets from bot activity.
• The framework achieves over 70% precision in optimal trading agent selection, demonstrating practical efficacy in combating market manipulation.

Multi-Agent Chain-of-Thought Reasoning for Robust Memecoin Copy Trading

Introduction and Market Background

The memecoin market, notably catalyzed by the launch of $TRUMP coin on Solana, has seen rapid onboarding of retail users who employ copy trading as a low-barrier strategy. Copy trading automates position replication from wallets believed to be key opinion leaders (KOLs), under the presumption that their expertise or insider status yields above-average returns. However, the absence of robust regulatory scrutiny and the prevalence of bot-driven manipulation have rendered naive copy trading hazardous. Adversarial actors deploy a range of bots—including bundle bots, volume bots, bump bots, and comment bots—not only to fabricate the appearance of liquidity and community engagement, but also to obscure blatantly manipulative activities or precipitate rug pulls.

Memecoin lifecycle events on platforms like Pump.fun are determined by a bonding curve subscription mechanism and migration thresholds, which fundamentally govern the early-stage pricing dynamics and transition to secondary market liquidity.

Figure 1: The bonding curve establishes the nonlinear relationship between total SOL deposited and meme coins received, driving price escalation with increased demand.

Manipulation Tactics and On-Chain Detection

The adversarial sophistication in memecoin markets is evidenced through coordinated bundles, wash-trading via bump bots, and sentiment spoofing through comment bots. Bundle bots allow project creators to obfuscate ownership concentration by dispersing initial token purchases across multiple synchronized wallets.

Figure 3: Typical structures and interactions of pump.fun manipulation bots across the lifecycle of a memecoin.

A detailed forensic case study of the MAO memecoin demonstrates the interplay between these tactics: a launch bundle masks origins, a sniper bot executes fast front-running, bump bots repeatedly “bump” coins to top page visibility, and orchestrated fake comments induce FOMO among retail entrants. These stages culminate in a rapid price collapse as orchestrators offload their holdings.

Figure 2: Timeline of MAO demonstrates the relationship between key bot-driven manipulations, candlestick formation, and transactional signals.

The empirical distribution of memecoin outcomes—maximum return achieved and dump duration—shows strong dependence on robot involvement. Launch bundle presence correlates with higher short-term returns but rapid post-peak depreciation. Bump and comment bots, by contrast, significantly extend both performance metrics, underlining their impact on perception-driven market variables.

Multi-Agent LLM Framework with Chain-of-Thought Reasoning

The proposed architecture models post-hedge-fund asset management workflows as a modular multi-agent LLM system. Each agent (trader evaluation, wealth management, meme evaluation, DEX execution) is equipped with few-shot chain-of-thought (CoT) prompts, which synthesize multimodal evidence from on-chain indicators, candlestick chart dynamics, and unstructured social commentary.

Figure 6: LLM-driven multi-agent workflow for memecoin copy trading, with discrete expert agent modules specializing in meme selection, KOL wallet assessment, allocation, and execution.

Meme Evaluation Module

Bot-detection heuristics and algorithms are employed to isolate manipulative on-chain behaviors:

• Bundle bot detection aggregates same-block transactions and traces cross-wallet transfers from creators.
• Bump bot detection computes trader-specific wash trading scores based on repeated offsetting trades.

Combined with visual (candlestick) analysis and sentiment extraction from comments, the meme evaluation agent achieves robust binary classification of farming potential. Notably, the absence of creator-funded bundles and a protracted premigration phase with fluctuating price action are critical signals of organic market engagement.

Figure 8: Example of a pre-migration candlestick pattern correlating with strong farming opportunity—distributed participation and robust price discovery.

Wallet (KOL) Evaluation Module

Wallet selection is premised on high total, low-variance profit and limited trading frequency, serving as a proxy for authentic KOL identification. The agent leverages prior trading records across multiple migrated coins, quantifying experience, variance, and portfolio breadth.

Experimental Results

Evaluation on 1,000 meme coin projects reveals that:

• The multi-agent framework attains 73% precision in high-quality meme coin identification and 70% precision in KOL wallet selection.
• KOL wallets flagged by the system collectively generated $500,000 in aggregate profits across the study period.
• The system robustly outperforms both single-agent LLM baselines and traditional ML models, especially when leveraging multimodal inputs.

Figure 4: Performance metrics for meme selection and KOL identification across post-migration intervals, showing peak precision and F1 scores at the 30-minute horizon.

Profit analysis across wallet classes further substantiates the discriminative power of the approach, enabling systematic filtration of manipulators and underperformers.

Figure 11: Distribution of realized profits for KOLs, manipulators, and noise traders, highlighting the differentiated outcomes achieved by strategic agent selection.

Implications and Future Directions

This paper demonstrates that LLM-driven multi-agent architectures addressing task specialization, multimodal reasoning, and explicit prompt engineering provide meaningful economic utility in adversarial, information-rich markets like memecoin trading. By formalizing manipulatory signals and integrating them programmatically, the system mitigates several critical risks associated with naive copy trading.

The theoretical implications are substantial: the combination of explicit algorithmic manipulation modeling and chain-of-thought-primed LLM reasoning is shown to be necessary for robust portfolio construction in highly non-stationary, bot-dense environments—outperforming monolithic models and human intuition-driven heuristics.

Practically, this framework could generalize to other thinly regulated marketplaces, decentralized exchanges, or crypto assets characterized by rapid turnover and adversarial participant behavior.

Conclusion

The paper "Resisting Manipulative Bots in Memecoin Copy Trading: A Multi-Agent Approach with Chain-of-Thought Reasoning" (2601.08641) establishes a state-of-the-art blueprint for LLM-powered automated trading systems resilient to manipulative adversaries. By decomposing the trading workflow into specialized, explainable modules and incorporating explicit manipulator detection, the framework achieves both strong empirical performance—$>70\%$ precision in actionable predictions and substantial monetary gains—and theoretical robustness. The methodology and insights presented offer a promising direction for extending AI systems to complex, adversarial, and high-variance financial domains.