MobiusQuant/OpenMobius-skill

OpenMobius‑skill – Trading‑knowledge skill for AI coding agents

What it is – A skill (plug‑in) that can be installed into several AI‑coding agents (Claude Code, Codex, OpenClaw, Hermes, Cursor, WorkBuddy). Once installed, the agent can answer trading‑related questions, fetch live market data, compute over 60 technical indicators, and generate annotated charts. The knowledge base contains ~2 000 curated “knowledge cards” from ICT, SMC, ChanLun, Wyckoff, etc., together with 18 645 source‑level evidence records.

Key features

• Multi‑school knowledge – 14+ trading “Schools” (ICT, SMC, ChanLun, Price Action, Elliott Wave, …) with strict scoping so the agent only uses the selected methodology.
• Real‑time market data – Supports crypto (Binance, Bybit, OKX, Hyperliquid), US/China stocks, forex; data is pulled on‑demand for the current turn.
• Technical indicators – 60+ built‑in indicators; the agent can return a specific indicator on a given asset without extra fetches.
• Chart generation – Two paths: (1) annotate a user‑provided image using Pillow, (2) generate a fresh chart with lightweight‑charts rendered in headless Chromium via Playwright.
• Intent & analysis routing – The skill first classifies the user’s intent (Q&A, chart analysis, annotation, K‑line analysis) and then resolves a lens (methodology) and School before deciding whether market data can be fetched. Unsupported routes fail closed with an explicit explanation.
• Composition modes – strict (use only the selected School), augment (primary + supporting School), compare (side‑by‑side Q&A of two Schools).
• Self‑contained installation – Each platform gets its own virtual environment, index, and cached model weights (nomic‑embed‑text‑v1.5) and Chromium binary.
• Platform support – Claude Code, Codex, OpenClaw (Linux/macOS), Hermes (Linux/macOS), Cursor, and WorkBuddy (ZIP import).

How to install

# clone the repo somewhere temporary
OPENMOBIUS_SRC=$(mktemp -d)
git clone https://github.com/MobiusQuant/OpenMobius-skill.git $OPENMOBIUS_SRC
cd $OPENMOBIUS_SRC
# pick the target agent (example: Claude Code)
python3 install.py --platform claude-code   # or codex, openclaw, hermes, cursor, workbuddy

The installer:

1. Copies the skill files into the agent’s skill directory (e.g. ~/.claude/skills/openmobius-skill/).
2. Creates a dedicated .venv and installs Python dependencies.
3. Downloads Playwright Chromium (280 MiB) and the nomic embedding model (547 MiB).
4. Builds the local vector index using the released seed; only missing documents are embedded on first run.
5. Generates a platform‑specific SKILL.md and runs a health check.

Subsequent runs reuse the cached model and index, so only the first install needs the large downloads.

Typical usage (after install)

"What is Liquidity Sweep?"                     # knowledge Q&A, strict ICT/SMC
"How is ETH 4h looking?"                      # live market analysis, chart output
"Explain 中枢 using ChanLun only"               # Q&A limited to the ChanLun School
"Compare ICT and ChanLun definitions of market structure"  # compare mode, Q&A only
"Generate a chart with my entry/SL/target"    # chart generation via Playwright
"Paste a CSV of OHLCV"                         # local K‑line analysis, no live data

The skill will either return a grounded answer (with citations) and/or a PNG chart, or it will explicitly state that the requested capability is unavailable (e.g., trying to use a non‑native School for live analysis).

Roadmap highlights

• Finish ICT sub‑school coverage (Inner Circle Mentorship, Silver Bullet, Power‑of‑3 variants).
• Add native analyzers for additional Schools beyond ICT/SMC.
• Expand the knowledge base (more concepts, cases, and source evidence).
• Improve semantic search options and provide a pure lexical fallback.

License & community

• Apache 2.0 license.
• Discord and Telegram channels for user support.
• Documentation includes README_AGENT.md (install steps for agents), INSTALL.md, and detailed workflow specs.

---

OpenMobius‑skill is a genuine, installable software component that equips AI coding agents with a structured, multi‑school trading knowledge base and live market‑analysis capabilities.