Nebius Token Factoryv1.3.1

What to watch for before installing: - Naming mismatch: The docs and SETUP.md show model IDs and examples using the 'nebius/' prefix, but the plugin's internal provider ID and manifest model prefix are 'tokenfactory'. This is likely to cause 'Unknown model' or 'plugin not found' errors. Before trusting this plugin, confirm which provider/model prefix is actually registered after installation (try: openclaw models list --provider tokenfactory and openclaw models list --provider nebius). Ask the author or check the GitHub source to confirm the intended prefix. - Secret handling: The plugin requires you to put NEBIUS_API_KEY into macOS launchctl and into a plaintext JSON auth file (~/.openclaw/agents/main/agent/auth-profiles.json). That is necessary for the gateway to authenticate, but it increases the number of locations where your key is stored. If you must use this plugin, ensure the filesystem permissions for that file are strict and rotate the key if it may have been exposed. - Platform assumptions: The setup docs are macOS-specific (launchctl, LaunchAgent). If you run OpenClaw on Linux or Windows, the instructions will not apply; seek platform-appropriate guidance in the repo or from the author. - Verification steps: After installing, verify the plugin communicates only with the stated host (api.tokenfactory.nebius.com) and that the models listed match your expectations. Run openclaw plugins inspect nebius (or tokenfactory) and openclaw models list and confirm 'Status: loaded'. Check the GitHub repo (package.json repository URL) for the source and open issues/PRs that might explain the naming mismatch. Why this is 'suspicious' not 'malicious': The package and instructions appear to implement the advertised functionality and request only the Nebius API key. The primary red flag is the inconsistent naming/namespace between 'nebius' and 'tokenfactory' across docs, manifest, and code — this is evidence of sloppy packaging and can lead to misconfiguration or inadvertent credential exposure, but it does not in itself prove intentional misdirection or exfiltration. If the author clarifies the naming or you confirm that the provider registers under the advertised model prefix, the assessment would likely move to 'benign'.