SkewX

Claude plugin skill for accurate, high-performance market data: trading platforms, research, and bots. Live today: orderbook imbalance (multi-DEX WebSockets, distance-weighted formula, 5-stage noise reduction). The plugin description notes that more modules may follow the same standards.

This site only summarizes what ships in the repository—.claude-plugin, skills/skewx/SKILL.md, and skills/skewx/reference.md—plus the repo README.

Quick start

Use SkewX inside Claude as a plugin skill (see README).

/plugin install skewx@claude-plugins-official

@skewx

Install

Install paths from the README:

/plugin install skewx@claude-plugins-official

/plugin install skewx@your-marketplace

What it provides

From the README (what the skill is for):

• WebSocket URLs for Pacifica, 01 Exchange, HotStuff, Paradex, Hibachi, Hyperliquid, Extended, Aster, Nado
• Subscribe message formats per DEX
• Symbol mapping (e.g. BTC/USD → per-DEX symbols)
• Distance-weighted imbalance formula
• 5-stage noise reduction pipeline
• Cross-DEX use cases for bots (spread comparison, arbitrage, liquidity mapping)

Venues

From SKILL.md—DEXes covered for WebSocket connection details:

Pacifica, 01 Exchange, HotStuff, Paradex, Hibachi, Hyperliquid, Extended, Aster, Nado.

URLs, subscribe payloads, symbol table, and level format are in skills/skewx/reference.md.

Order book & imbalance

The skill documents a distance-weighted imbalance: bid vs ask size near mid, with farther levels down-weighted via decay λ (typical value 10). Output range [-1, 1]—positive buy pressure, negative sell pressure (SKILL.md).

Primary formula

I = (Σ B_i·e^(-λ·d_i) - Σ A_i·e^(-λ·d_i)) / (Σ B_i·e^(-λ·d_i) + Σ A_i·e^(-λ·d_i))

Where d_i = fractional distance from mid |p_i - mid| / mid, mid = (best_bid + best_ask) / 2, and B_i, A_i are bid/ask size at level i.

5-stage noise reduction

From SKILL.md—apply in order:

1. Spike filter — If |value - prev| > 0.3, replace with median of last 5 samples.
2. EMA — Exponential moving average, 1s half-life.
3. Kalman filter — Process noise 0.01, measurement noise 0.1.
4. Robust mean — Median of last 15 Kalman outputs.
5. Regime — From smoothed history std: stable (σ<0.02), trending (σ<0.08), volatile (σ≥0.08). Confidence: stable=1, trending=0.8, volatile=0.5.

Full math write-ups (LaTeX-style) can live in the skill file or your own notes; the repo does not ship a separate formula markdown file.

WebSocket reference

Per-venue wss:// URLs, subscribe messages (or URL-embedded symbols), snapshot vs delta notes, symbol mapping table, and level format are maintained in the repo only:

skills/skewx/reference.md

Example from that file: Hyperliquid wss://api.hyperliquid.xyz/ws with l2Book subscription—see the file for the exact JSON.

Commands

The skill runs in your Claude environment. The README lists:

@skewx
/skewx:skewx

Install (official marketplace): /plugin install skewx@claude-plugins-official

Use cases

From SKILL.md (“Usefulness for Bots”):

• Cross-DEX spread: Compare imbalance across venues. Divergent imbalances = arbitrage opportunity.
• Directional signal: Sustained positive imbalance → short-term bullish bias; negative → bearish.
• Liquidity mapping: Identify support/resistance zones, thin vs thick levels.
• Order flow context: Distinguish passive resting liquidity from aggressive directional intent.

About these docs

Static HTML in docs/ (index.html, style.css, script.js). Authoritative technical detail for WebSockets and symbols remains in skills/skewx/reference.md; narrative and formulas in skills/skewx/SKILL.md.

SKILL.md also mentions a separate orderbook-imbalance web app for real-time comparison across DEXes; the skill itself documents methodology so you can implement connectors and calculations. See your README for how you link that project.

Plugin metadata: .claude-plugin/plugin.json (name skewx, version, author). Maintainer map for files: docs/README.md.