README.md

# Altair skills

Agent skills from [Altair AI](https://altairworld.com). They read live data from Altair's free public MCP server, so an installed skill always returns the newest research without updates.

## altair-research

The latest AI research on about 70 US stocks and ETFs: the **SPY** (S&P 500), **QQQ** (Nasdaq-100), **SOXX** (semiconductors) and **XBI** (biotech) index ETFs, refreshed weekly, and stocks such as NVDA, AAPL, MSFT, TSLA, AVGO and META.

Claude, GPT and Gemini models each run the same multi-agent [TradingAgents](https://github.com/BichengWang/TradingAgents) pipeline on every ticker: technical, sentiment, news and fundamentals analysts, a bull/bear debate, a trader, a risk team and a portfolio manager. Reports are published at [altairworld.com/TradingAgents](https://altairworld.com/TradingAgents/). The skill pulls every model's final decision for the newest analysis date and shows where the models disagree:

```text
## SOXX · analysis date 2026-10-21 · 4 models: 2 Hold, 2 Underweight

| Model            | Rating      | Close at report | Target | Target vs close | Confidence | Horizon    |
| claude-fable-5-1 | Hold        | 510.36          | 522.51 | +2.4%           | Low        | 1-2 months |
| gpt-6-astra      | Underweight | 512.64          | 493.39 | -3.8%           | Medium     | 4-8 weeks  |
| gemini-4-argon   | Underweight | 510.36          | 470.12 | -7.9%           | High       | 1-3 months |
| claude-opus-5-5  | Hold        | 512.64          | 535.33 | +4.4%           | Medium     | 1-2 months |
```

Then ask your agent things like "What do the AI models think of Nvidia?", "Compare the research on AMD and AVGO", "What's the semiconductor ETF outlook this week?" or "Why is GPT underweight XBI?"

### Install

Any agent that supports skills (Claude Code, Codex, Cursor, Gemini CLI and others), using the [skills CLI](https://github.com/vercel-labs/skills):

```bash
npx skills add altair-world/altair-skills --skill altair-research
```

Claude Code plugin:

```bash
claude plugin marketplace add altair-world/altair-skills
claude plugin install altair@altair-skills
```

The plugin also adds the free `altair` MCP server (no key needed).

GitHub CLI: `gh skill install altair-world/altair-skills altair-research`. OpenClaw, from [ClawHub](https://clawhub.ai): `clawhub install altair-research`. Or copy `skills/altair-research` into your agent's skills folder.

Needs Python 3.8+ and internet access. No API key or sign-in.

### OpenAI directory package

The `.codex-plugin/plugin.json` manifest packages the skill and the public remote MCP server declared in `.mcp.json` for the directory shared by ChatGPT and Codex. It lists the plugin under **Finance** for reading Altair's published stock and ETF research, includes the [terms of service](https://altairworld.com/terms) URL in both plugin manifests, and includes the existing 512×512 icon, five positive and three negative review cases, and release notes. Version 1.0.0 is unchanged. Build the upload ZIP with the command in [AGENTS.md](AGENTS.md#release); this prepares a package without submitting it.

Use a **With MCP** submission. The [skill guide](https://developers.openai.com/plugins/build/skills) assigns live data access to MCP tools; it does not guarantee internet access for bundled Python scripts in ChatGPT skill sandboxes. The script remains available in environments with Python and internet access, and the skill already supports connected MCP tools. A [skills-only publication cannot later gain an MCP server](https://developers.openai.com/plugins/deploy/submission#create-the-draft).

Before portal submission, complete organization/developer identity verification, ensure the [terms of service](https://altairworld.com/terms) page is published, and ensure the [privacy policy](https://altairworld.com/privacy) describes MCP daily counts by tool and client name, retention, and the absence of stored arguments, responses and IP addresses. Host the portal's exact plain-text token at `https://altairworld.com/.well-known/openai-apps-challenge`, then connect the unauthenticated MCP endpoint and resolve metadata, skill and tool scan findings. Record the five positive and three negative cases in a walkthrough and supply an accessible video URL in Review details or `extensions.com.openai.review.demo_recording_url` before review. The ZIP deliberately omits that URL until a real recording exists. See the [submission requirements](https://developers.openai.com/plugins/deploy/submission) for review and publication steps.

The live server also advertises `get_research_signals`, a separate paid digital-data tool using x402 payments. The skill's three research tools do not require payment, but a directory scan sees the server's full tool inventory. Resolve that paid tool's exposure before submission: the [plugin guidelines](https://developers.openai.com/plugins/plugin-guidelines#commerce-and-monetization) prohibit selling digital products or services. The package makes no commerce declaration pending that server-side decision.

### Run it yourself

```bash
python3 skills/altair-research/scripts/altair_research.py tickers
python3 skills/altair-research/scripts/altair_research.py latest
python3 skills/altair-research/scripts/altair_research.py latest NVDA AMD --date 2026-10-14
python3 skills/altair-research/scripts/altair_research.py report <report-url> --section research
```

The script calls the MCP tools `get_research_decisions`, `list_research_reports` and `read_research_report` at `https://altairworld.com/mcp/free`, the endpoint with only Altair's free tools, over plain HTTPS; set `ALTAIR_MCP_URL` to point it at another deployment. The plugin's MCP server uses the same endpoint. To use those tools directly in an MCP client instead: `claude mcp add --transport http altair https://altairworld.com/mcp/free`. See [altairworld.com/mcp](https://altairworld.com/mcp) for the other free tools, such as market breadth.

## Disclaimer

The reports are AI-generated for AI testing and experimentation only. They are not financial or investment advice, may contain errors, and must not be used to make investment or trading decisions.

## Contributing

Changes reach `main` only through pull requests that pass the `check` workflow: manifest and `SKILL.md` frontmatter checks, then the script against the live MCP server on Python 3.8 and 3.13. Add the `auto-merge` label to a pull request and it squash-merges once `check` passes. If a server change is not deployed yet, `check` fails; re-run it after the deploy.

The [main ruleset](.github/main-ruleset.json) enforces this. To recreate it: `gh api --method POST repos/altair-world/altair-skills/rulesets --input .github/main-ruleset.json`.

## License

[MIT](LICENSE)