Run a live Kalshi prediction-market account autonomously
🔴 AI agent ran $20 Kalshi account; earned $20.30 in payouts vs. Task: The agent Mirabilis, built on Claude Code, operated a live Kalshi prediction-market account for nine days under human supervision. It executed trading strategies and managed liquidity incentives, with a toolkit released to model the reward structure based on snapshot size and price ticks. Human involvement was stated as supervision, but specific intervention points were not detailed in the source.; tools: Claude Code. Human role: The original post does not say how a person stayed involved.; limitations: Prerequisites include access to Kalshi and a funded account. The source does not state specific operational limitations or costs beyond the trading losses. Integration uncertainty exists regarding how human supervision was practically implemented. Source reliability is secondary, based on a social media post..
Tools
Published on Agent Work Atlas:
Overview
🔴 AI agent ran $20 Kalshi account; earned $20.30 in payouts vs.
Workflow
The agent Mirabilis, built on Claude Code, operated a live Kalshi prediction-market account for nine days under human supervision. It executed trading strategies and managed liquidity incentives, with a toolkit released to model the reward structure based on snapshot size and price ticks. Human involvement was stated as supervision, but specific intervention points were not detailed in the source.
Reported result
The agent earned $20.30 in Liquidity Incentive Program (LIP) payouts but sustained $36.06 in trading losses, resulting in a net loss of $15.76 on the $20 account. The source does not provide further measured outcomes or success metrics beyond these financial figures.
Human role
The original post does not say how a person stayed involved.
Limitations
Prerequisites include access to Kalshi and a funded account. The source does not state specific operational limitations or costs beyond the trading losses. Integration uncertainty exists regarding how human supervision was practically implemented. Source reliability is secondary, based on a social media post.
Try it
I want to simulate a low-risk prediction market trading workflow using Claude Code. My goal is to test a liquidity incentive strategy on a Kalshi account with a maximum budget of [replaceable: $20]. Steps: 1. Analyze the provided Kalshi LIP reward structure: rewards = size × d^(ticks below reference price), with daily payouts only if ≥$1.00. 2. Draft a Python script using the Kalshi API to monitor market snapshots and calculate potential LIP rewards. 3. Propose a trading strategy that maximizes LIP payouts while minimizing directional risk. 4. Run a dry-run simulation on historical data for [replaceable: 1 day] to estimate potential earnings vs. losses. 5. Present the simulation results and the proposed live trading plan to me for approval. Do not execute any live trades or spend real money without my explicit confirmation after reviewing the dry-run results. If API access is missing, ask for credentials or mock data. Output the script, simulation report, and risk assessment in a markdown file.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.