What problem does it solve? High-reasoning models waste expensive tokens on mechanical work like reading repositories, gathering evidence, and running review passes. This Skill reserves the session model for senior judgment—deciding what to investigate, writing delegation prompts, and judging tradeoffs—while routing the legwork to cheaper delegates like Codex CLI, DeepSeek via OpenRouter, or Opus subagents. ## Core Features & Use Cases - Companion-script delegation: All Codex calls go through a dependency-free Node.js script that runs tasks in foreground or background, tracks job state on disk, and supports wait, watch, status, result, and cancel commands. - Tiered delegate routing: Choose Heavy (sol, high effort), Light (luna, max effort), or Light+wide (DeepSeek via OpenRouter with 1M context) based on whether the work feeds hard-to-reverse decisions or bulk evidence gathering. - Multiple delegation paths: Pin sessions to Codex, Luna, DeepSeek, an Opus single agent, or an Anthropic agent team via slash commands like /senior-mode:codex and /senior-mode:team. - Use Case: Facing a multi-file migration, you invoke /senior-mode, and Claude writes a bounded investigation prompt, delegates the repo scan to Codex in the background, reads the returned evidence report, and produces a decision document—without ever reading the codebase itself. ## Quick Start Type /senior-mode in Claude Code with a description of the architectural or debugging question you need investigated, and Claude will delegate the evidence gathering to Codex and return a senior judgment.
