How to use the Market Microstructure Analysis agent skill

What is an agent skill, and what does this one make the agent do?

An agent skill is a folder containing a SKILL.md file with instructions and reference knowledge that an AI agent loads when a task matches it. This skill, named market-microstructure, is a reference of formulas and rules of thumb rather than a live data feed. It does not itself pull market data.

Once installed, the agent can use these methods when you ask about trading costs or execution: quoted, effective and realized spreads; VPIN and Kyle's lambda for order-flow toxicity; Amihud and Roll measures for liquidity; a power-law and square-root price-impact model; limit-order-book depth and imbalance; and China A-share specifics such as call auctions and block trades. The aim is to replace a flat fee assumption with a more realistic cost estimate.

• Estimate transaction costs for a backtest instead of assuming a flat 0.1% fee
• Plan how to split a large order using TWAP, VWAP or implementation shortfall (IS)
• Flag time windows where order flow may be dominated by informed traders
• Measure liquidity risk and warning signs of a flash crash

Install the market-microstructure skill

Download the ZIP for this skill from this page. Unzip it so that the market-microstructure folder, containing SKILL.md, ends up in one of two places: ~/.claude/skills/ to make it available across your projects, or a project's .claude/skills/ folder to limit it to that project.

If you use the Claude apps instead, open Settings, go to Capabilities, and upload the skill under Skills.

• Personal install: place the folder in ~/.claude/skills/
• Project install: place the folder in the project's .claude/skills/ folder
• Claude apps: Settings → Capabilities → Skills, then upload

Trigger the skill with natural requests

You do not need a special command. Describe the task in trading terms and mention the metric or situation you care about. The skill's description covers bid-ask spreads, VPIN, Kyle lambda, Amihud and Roll measures, price impact, limit-order-book analysis, and China A-share call auction and block trade mechanics, so requests that use these terms are the most likely to match.

Give the agent the data you have. The calculations need trade, quote or daily volume data, which the skill does not supply.

• "Estimate the round-trip cost for my backtest on 000001.SZ using a spread-based model."
• "How much would selling 100,000 shares of a small-cap stock move its price, given 500,000 shares average daily volume and 3% daily volatility?"
• "Compute the Amihud illiquidity measure for these monthly daily returns and volumes."
• "Suggest a TWAP versus VWAP plan to sell this position over three days."
• "Explain what VPIN measures and how to calculate it from my tick data."

Worked example: price impact of a large order

The skill gives a power-law impact formula, impact = η × σ × (Q / V)^0.6, where η is an impact coefficient (about 0.5 to 1.5 for China A-shares), σ is daily volatility, Q is the traded quantity and V is average daily volume. The source's example sells 100,000 shares of a large-cap stock with 5,000,000 shares average daily volume and 1.8% daily volatility, giving roughly 1.5 basis points of impact. The same 100,000 shares in a small-cap stock with 500,000 shares average daily volume and 3.0% volatility gives about 23 basis points, which the source says should be executed in slices.

A square-root model, impact = σ × √(Q / (ADV × T)), is listed for large trades where Q/ADV is above 5%.

Requirements and what you must supply

The skill lists no required software or tools. It runs no scripts and uses no network access. What it does need is data from you: for example, trade and quote data for VPIN or spread measures, and daily prices and volumes for Amihud, Roll or the impact formulas. The source's impact section names a tested implementation, src.quantlib.impact.sqrt_impact, in the wider repository, but this skill file itself is formulas and text.

• Required tools or software: none listed
• Scripts: none run by the skill
• Network: none used by the skill
• Your input: trade, quote or daily volume data for the calculations

Safety notes before you rely on the output

The thresholds in the file are simplified heuristics. For example, it says a VPIN above 0.5 means major news is about to be released, and the skill itself gives no evidence those cutoffs are reliable. Do not use them to make real trading decisions on their own. The order-book imbalance signal is also noted as needing filtering, because large orders are often added and canceled in China A-shares.

The skill does not implement a full Almgren-Chriss optimal execution. It gives no trading trajectory and no risk-aversion parameter. Treat outputs as estimates to check against your own data and assumptions.

• Do not treat VPIN, OIR or ILLIQ cutoffs as validated predictions
• Check any execution plan against your own data before trading
• The skill is not designed for production high-frequency systems

Tips for better results

Name the instrument, market and time window, and say which data you have. Specify whether you need a quick estimate or a calculation you can reproduce. Ask the agent to state its assumptions, such as the impact coefficient or the volatility figure it used, so you can change them.

• Always give the stock code and the period covered by your data
• Provide the exact inputs (volume, volatility, quantity) rather than rough descriptions
• Ask the agent to show the formula it applied and the numbers it substituted
• Compare results across a few impact coefficients, since the source gives a range rather than one value