← Files ohmyho.stARCHIVED FILE
README.md
3.23 KB · Nov 4, 2026 · 12:09 UTC
# ohmyho.st plugin package This local package combines the ohmyho.st Skills with its remote MCP connection. It contains no API keys, device tokens, runtime credentials, hooks or local server dependency. The client connects through managed OAuth; project editing/preview and production publishing require the user's own explicit grants. The remote server advertises WorkOS Connect authorization metadata and standard OIDC scopes (`openid`, `profile`, `email`, `offline_access`). The client completes OAuth through that metadata; project selection and preview/publish rights are then approved in the signed-in ohmyho.st portal. A successful OAuth connection alone grants no application access. The hosted client must support the server's authorization flow and be approved for it. Use the included get-started and deploy Skills to continue the same project across chats. Keep the existing local CLI/device-login workflow for a local checkout, and the GitHub Skill for a project that already uses GitHub or a customer who explicitly selects it. ## Prepare and check From the platform repository, run: ```sh node scripts/prepare-agent-plugin.mjs node scripts/prepare-agent-plugin.mjs --check ``` The script copies the canonical customer Skills into this contained package. It refuses symbolic links, unexpected package files, invalid manifest paths and any credential-bearing MCP configuration. Rerun it after canonical Skills or generated surface references change. The folder contains an OpenAI compatibility manifest at `.codex-plugin/plugin.json` and a Claude manifest at `.claude-plugin/plugin.json`; both use the same `.mcp.json`. All referenced files stay inside the package. A platform-specific upload archive includes only that platform's manifest directory. An upload archive must contain the folder or these files without symbolic links, other repositories, credential files or unrelated workspace contents. ## Native testing and publication This repository package is prepared for testing, not evidence of a public directory listing or a completed native-client acceptance run. First verify the deployed HTTPS server and OAuth discovery, then install the package in the chosen client's private/local testing source. In Claude, add/connect the bundled connector from the plugin's Connectors tab; installing the Skills alone does not authenticate the user. In ChatGPT, test the custom remote MCP connection and the complete installed package through the supported plugin testing flow. Exercise a new app, a resumed app, a competing edit, a denied project and a private secret-input handoff. Check a verified preview and an explicitly approved publication of the exact commit. Public review, reviewer access, listings and publication remain separate authorized steps; never bundle reviewer credentials or fabricate a registered connector ID. Official packaging references: - [OpenAI plugin packaging](https://developers.openai.com/plugins/build/plugins) - [OpenAI connection testing](https://developers.openai.com/plugins/deploy/connect-chatgpt) - [OpenAI submission](https://developers.openai.com/plugins/deploy/submission) - [Claude manifest reference](https://code.claude.com/docs/en/plugins-reference) - [Claude plugins and bundled connectors](https://claude.com/docs/plugins/overview)
SHA-256: 8621ff349ecb2f8aa6b7bbec51dbe72b9b144a647663c22190a961a4a34581d9