# Model compatibility

MarketPilot AI follows the open Agent Skills folder shape: each capability is a directory containing `SKILL.md`, optional support files and, where applicable, host UI metadata. The same 17 instructions can be used across several assistants, but each host has its own discovery paths, approval rules and execution tools.

| Environment | Practical integration | Setup / limitation |
|---|---|---|
| ChatGPT desktop | Standalone Agent Skills (`SKILL.md`, `agents/openai.yaml`, explicit `@` invocation) | The 17 folders include OpenAI UI metadata and disable implicit invocation by default. |
| ChatGPT Chat / Work (web, desktop, mobile) | Skills-only plugin package | Root `plugin.json` provides package metadata for supported plugin workflows. Host availability, import options and runtime permissions depend on the ChatGPT environment. |
| OpenAI Codex CLI / IDE | Native local skills, `$skill` invocation | `python scripts/install_skills.py codex` copies to `~/.agents/skills/`; use `codex-project` for a repository-local install. |
| Claude Code | Native local skills, `/skill` invocation | `python scripts/install_skills.py claude` copies to `~/.claude/skills/`; use `claude-project` for `.claude/skills/`. A Claude Code plugin manifest is also included under `.claude-plugin/`. |
| Grok CLI | Native local skills, `/skill-name` invocation | `python scripts/install_skills.py grok` copies to `~/.grok/skills/`; use `grok-project` for `.grok/skills/`. Grok also supports the shared `~/.agents/skills/` path and Claude Code skill/plugin layouts. |
| Grok apps | Built-in/custom Skills feature | xAI documents custom skills across Grok surfaces; the in-app builder and account-specific workflow may differ from the CLI folder workflow. |
| Gemini CLI | Native Agent Skills | `python scripts/install_skills.py gemini` copies to `~/.gemini/skills/`; use `gemini-project` for `.gemini/skills/`. Gemini CLI also recognizes the shared `~/.agents/skills/` alias. |
| Gemini Apps | Skills page upload | Upload an individual `SKILL.md`, a folder or a ZIP containing `SKILL.md` at its root. Internet-dependent scripts are not supported in this upload flow, so run MarketPilot AI's Python data tools separately in a local/approved environment. Availability and account requirements can change. |

## Install locally

```bash
# Install Claude Code + the shared Agent Skills path.
# The shared ~/.agents/skills path is discovered by Codex,
# and is an alias supported by Grok and Gemini CLI.
python scripts/install_skills.py all

# Or choose a dedicated user-level path:
python scripts/install_skills.py claude
python scripts/install_skills.py codex
python scripts/install_skills.py grok
python scripts/install_skills.py gemini

# Or a repository-local path:
python scripts/install_skills.py claude-project
python scripts/install_skills.py codex-project
python scripts/install_skills.py grok-project
python scripts/install_skills.py gemini-project
```

Existing MarketPilot AI skill directories are skipped by default. `--force` replaces only the same-named MarketPilot AI skill directories; inspect local modifications before using it.

## OpenAI package metadata

The portable root `plugin.json` describes a skills-only package. It contains no MCP server, external account connector or credentials, and it does not grant shell, filesystem, network or brokerage access. Standalone skills are available in supported ChatGPT desktop and Codex CLI/IDE workflows; plugin capabilities depend on the host environment. See the official [Build skills guide](https://learn.chatgpt.com/docs/build-skills) and [Plugins overview](https://learn.chatgpt.com/docs/plugins).

## Running the Python engine

A model reading a `SKILL.md` file does not by itself gain local shell, filesystem, Python or network access. Install the CLI separately with `pip install -e .`, run the requested `marketpilot` command in a trusted local environment, and provide its JSON output to the assistant—or connect an explicitly approved execution tool. Gemini Apps' current skill-file upload feature does not run internet-dependent scripts.

The skill files are workflow instructions; the deterministic CLI performs calculations. External research requires the relevant data provider and network access. No brokerage key is needed and no live order endpoint is included.

## Shared workflow contract

- Ask for missing inputs rather than guessing.
- Preserve source URL, retrieval timestamp, data interval and provider limitations.
- Separate observed data, derived calculations, simulations and assumptions.
- Research and alert features are read-only or local. The paper ledger never routes orders.
- Never request or paste brokerage keys, passwords, account identifiers, private tax records or private journals into model prompts.

## What “cross-model” means

The skills use portable Markdown and host metadata rather than proprietary model API calls. Native skill discovery is supported in OpenAI ChatGPT/Codex, Claude Code, Grok CLI and Gemini CLI; ChatGPT and Gemini Apps also provide their own skill distribution surfaces. That does **not** give every host identical browsing, local filesystem access, code execution, persistent state or market-data connectivity.

## Official references

- OpenAI, [Build skills for ChatGPT and Codex](https://learn.chatgpt.com/docs/build-skills).
- OpenAI, [Package your plugin](https://developers.openai.com/plugins/build/plugins).
- Anthropic, [Agent Skills overview](https://platform.claude.com/docs/en/agents-and-tools/agent-skills/overview).
- xAI, [Skills, Plugins & Marketplaces](https://docs.x.ai/build/features/skills-plugins-marketplaces).
- Google, [Gemini Apps Skills help](https://support.google.com/gemini/answer/17094296) and [Gemini CLI Agent Skills](https://github.com/google-gemini/gemini-cli/blob/main/docs/cli/skills.md).

Recheck vendor documentation before distributing updates; feature availability and paths can change.