AlphaGBM Skills: 29 Options-Intelligence Skills for Claude Code and Cursor

The gap AlphaGBM Skills fills between an LLM and an options chain

Ask a general-purpose model for a view on NVDA options and you get plausible prose with fabricated Greeks. Ask a brokerage API and you get a JSON chain you still have to interpret. AlphaGBM Skills sits between those two positions. It ships 29 skills that an AI agent can invoke as tools, and each one returns computed options metrics: implied volatility, Greeks, variance risk premium, skew, flow and a 0 to 100 options score. The README frames the distinction directly, contrasting LLM roleplay tools that emit a confidence figure with a system where, in its words, every number has a source. That claim is the whole product. If the numbers are real, the skill set is a shortcut around building your own analytics layer. If they are not, it is a wrapper with a marketing page. The target user is a discretionary trader or analyst who already lives in Claude Code or Cursor and wants options context without leaving the editor. It is not aimed at execution, order routing or portfolio accounting.

How the skills, the CLI and the hosted API fit together

There are three moving parts. The first is the skills directory itself, a collection of markdown skill definitions that an agent reads to learn when and how to call AlphaGBM. The second is a Python CLI installed from the cli subdirectory. The third is a hosted API at alphagbm.zeabur.app, which the README lists as the default base URL and which the CLI authenticates against with an agbm_ prefixed key. Data flows from that API through the CLI or through direct skill invocation, and the agent formats the result. The API exposes a health endpoint at /api/health that returns status, available data fields, data source health and market coverage without authentication, which the README suggests agents call first to check what is available. Skills are grouped into four families: core analysis (stock analysis, options score, strategy builder, volatility surface, volatility smile, Greeks, P&L simulator), data intelligence (IV rank, earnings crush, unusual activity, market sentiment, VIX status, FearScore), workflow tools (compare, watchlist, alert, Polymarket) and risk and portfolio discipline (hedge advisor, BPS backtest and others). The stock side uses a model the README calls G = B + M, gain equals basics plus momentum. FearScore is described as a per-ticker composite of six indicators where a reading of 60 or above is treated as a bull put spread entry signal. Note the arithmetic inconsistency in the README: the badge and headline say 29 skills, the body text says 26, and the section tables list 7 + 6 + 4 + 3. That discrepancy is unresolved in the material.

Installing into Claude Code, Cursor and the CLI

For Claude Code, the README gives a single clone into the skills path: git clone https://github.com/AlphaGBM/skills.git .claude/skills/alphagbm, with a submodule variant using git submodule add against the same target. Cursor uses the same repository cloned into .cursor/skills/alphagbm. The CLI takes a different route: clone the repository, cd into skills/cli, then pip install -e . for an editable install. Authentication is a single command, alphagbm config set-key agbm_xxxxxxxxxxxxxxxx. From there the documented entry points are alphagbm stock analyze AAPL and alphagbm options score NVDA. Live data can also be configured through environment variables, ALPHAGBM_API_KEY and an optional ALPHAGBM_BASE_URL. The demo path requires no key at all: the README states that all skills include built-in demo data for AAPL, NVDA, SPY, TSLA and META, so prompts like "Score NVDA options" or "Show me TSLA's volatility surface" return something immediately. The README also notes that the assets/demo-screenshot.png reference is still a TODO placeholder, so the visual documentation is not finished.

Quota limits are the real constraint on daily use

The free tier allows 2 stock analyses per day, 1 options analysis per day, and unlimited quick quotes and snapshots. Plus and Pro raise the analysis ceilings to 1,000 and 5,000 per month respectively, with snapshots still unlimited. Read that table carefully before planning a workflow. A single options score consumes the entire free daily allowance. Anything resembling a scan across a watchlist, or a backtest loop, will exhaust a free key in one sitting. The BPS backtest skill, which the README describes as a walk-forward backtest of a bull put spread with a signal versus no-signal control in one call, is the clearest example of a skill whose natural use pattern collides with per-call metering. The material does not state how a backtest call is counted against the quota, or whether backtests draw on the same analysis units. That is an open question worth resolving before building anything scheduled on top of this.

Where the design breaks down: coverage, data provenance and the missing release history

The README claims coverage of US, Hong Kong, Chinese and commodity markets, but every worked example in the material is a US equity or ETF ticker. There is no HK, CN or commodity example anywhere in the supplied text, so the breadth claim is asserted rather than demonstrated. Data provenance is the second soft spot. The README says numbers come from real market data and that the API health endpoint reports data source health, but it does not name the upstream providers, the refresh interval, or whether quotes are delayed. The comparison table lists generic finance APIs as offering delayed or basic data and AlphaGBM as real-time, which is a claim about competitors as much as about itself. Third, the repository shows no retrieved releases, so there is no changelog to inspect for breaking changes in skill definitions or CLI flags. Anyone pinning this into a team workflow is pinning to a moving main branch. Fourth, the README's own user and track-record figures are self-reported marketing, and the demo screenshot is still a placeholder. Treat the analytics as the product and the badges as decoration.

How this differs from pulling a chain from a broker API

The obvious alternative is to skip the skills layer and call an options data provider directly, then write your own scoring. That path gives you raw chains, your own storage, and full control over refresh and history. The difference in approach is that AlphaGBM does the interpretation for you: it returns a 0 to 100 options score, a FearScore composite, an IV rank against 252 days of history, and pre-built strategy templates (the README cites 15 or more). You are trading control and provider transparency for opinionated, ready-made metrics and an agent-native interface. A second alternative is simply asking the model without any tool, which produces fluent nonsense. AlphaGBM's value over that is entirely the data plumbing. If you need tick-level history, custom backtesting over years of chains, or execution connectivity, a direct provider relationship is the better fit. AlphaGBM is a decision-support layer, not a data warehouse, and the README does not present it as one.

Licence, maintenance and the cost of staying current

The project is MIT licensed, which permits commercial use, modification and redistribution provided the copyright notice and permission notice are retained. That is permissive and low-friction for internal tooling. It is not legal advice; if you redistribute the skills inside a product, read the LICENSE file in the repository and take your own counsel on attribution. Maintenance cost has two components. The first is the hosted API, which the skills depend on for anything beyond the five built-in demo tickers. If the endpoint at alphagbm.zeabur.app changes or the key format changes, the skills need updating. The second is the skill definitions themselves: cloning into .claude/skills/alphagbm means pulling updates requires a git pull in that directory, or a submodule update if you took the submodule route. There is no versioned release to pin against in the retrieved material, so upgrades are effectively continuous. Budget time for reading the repository diff before pulling into a shared team checkout.

Editorial conclusion

AlphaGBM Skills is worth installing if you already work inside Claude Code or Cursor and want options scoring, IV rank and Greeks available as queries instead of a separate dashboard. Skip it if you need unattended batch analysis, since the free tier allows one options analysis per day and the CLI requires an API key. Before adopting, clone the repository, run the built-in demo prompts for AAPL or NVDA to confirm the skills load in your agent, and check the quota table against your own usage pattern.

Official sources

1. AlphaGBM/skills on GitHub
2. Issues
3. License: MIT
4. Project website
5. README