This document explains what the OrgX OpenClaw plugin stores, how authentication works, and how to avoid leaking sensitive data into prompts.

The plugin stores:

• An OrgX credential (oxk_...) in OpenClaw's credential store after browser pairing, or a manually entered key if you choose that path.
• Local configuration (plugin config + runtime state) required to run the dashboard and background sync.
• Optional MCP client autoconfig backups (if enabled) when patching local MCP config files.

The plugin does not intentionally store:

• Your OpenAI/Anthropic/provider API keys (BYOK keys stay in OpenClaw settings).
• Raw files from your repositories.

Supported auth modes:

1. Browser pairing: you authenticate on useorgx.com and approve the connection. The plugin stores a scoped OrgX key.
2. Manual API key entry: you provide an oxk_... key directly.

The plugin uses the stored OrgX key to call OrgX APIs and to expose the local MCP bridge endpoint at /orgx/mcp.

When using the MCP tools:

• Do not paste secrets (API keys, passwords, tokens) into prompts.
• Prefer referencing files by path and letting tools read only what is necessary.
• Avoid sharing customer PII in agent prompts. Use redacted examples.

If you believe you accidentally shared sensitive data:

• Rotate the affected secret immediately.
• Revoke the OrgX key and re-pair.

If you discover a security issue:

• Do not open a public issue with exploit details.
• Email the OrgX team with reproduction steps and impact.

When developing the plugin locally:

• Prefer using a test OrgX account and minimal-permission keys.
• Keep .bak.* backups of MCP config files out of git.