panusbodee-owen/Power-Ai-Ranger-Skill

This Claude Code skill enables Claude to plan tasks and create work orders, which are then sent to DeepSeek, Nemotron, Qwen, gpt-oss, or any model with an OpenAI-compatible interface for execution. The reviewer sits in a separate seat; if the output is unsatisfactory, it can be returned for multiple rounds of revision. It also automatically checks JSON and regular expressions, and runs code in a sandbox. A web dashboard displays who is doing what, pass rates, and usage; outbound content filters out keys and ID numbers, while limiting length and calls per hour.

This Claude Code skill enables Claude to plan tasks and create work orders, which are then sent to DeepSeek, Nemotron, Qwen, gpt-oss, or any model with an OpenAI-compatible interface for execution. The reviewer sits in a separate seat; if the output is unsatisfactory, it can be returned for multiple rounds of revision. It also automatically checks JSON and regular expressions, and runs code in a sandbox. A web dashboard displays who is doing what, pass rates, and usage; outbound content filters out keys and ID numbers, while limiting length and calls per hour.

Why it matters Use cheaper models for execution and stronger models for oversight, creating an "execute-review" pipeline with call limits, key filtering, and audit logs.

Stars 16

Reasoning

  • The README states that screenshots and GIFs are from fictional demo data, without displaying real keys or gateways

Usage

  • Clone to ~/.claude/skills/ai-team, use scripts/team.py add to register the interface addresses and names of each model (keys are entered silently in the terminal), then describe tasks in Claude Code

Risks

  • The repository does not declare a license
  • You can use locally logged-in command-line subscriptions such as Codex and Claude Code as seats; compliance with each provider's subscription terms must be verified by the user

Open

GitHub

Explore the latest AI Radar updates