proyecto26/seldon — reverse-engineered prompt

Build me a Claude Code plugin called Seldon that acts like an independent reviewer for implementation plans, specs, and RFCs.

I want to be able to point it at a plan file, have it read the workspace for evidence, then return a clear verdict of approve, approve_with_changes, or request_major_revision. The response should include a confidence score, a short summary, strengths, blocking findings, open questions, severity levels, and file plus line references when possible.

It should work inline with Claude Code without requiring API keys, but also support optional outside judges through Codex, OpenAI, or Anthropic if those are configured. Add simple focus modes like balanced, architecture, evaluation, product, operations, and safety, so I can ask for a specific kind of review.

Please include the plugin skill files, runner scripts, a schema for the output, a validation smoke test, and clear install and usage docs. Look up current Claude Code plugin docs online if you need to.