Start

You are starting a review-by-opp session. This activates strict review tracking.

What to do

1. Check prerequisites:

   • Verify this is a git repository. If not, tell the user: "review-by-opp requires a git repo. Run git init first."
   • Check if a session already exists by looking for reviews/current.json. If one exists, ask: "A session is already active. Use /review-by-opp:resume to continue or /review-by-opp:finalize to close it first."

2. Initialize session:

   • Create the reviews/ directory structure: reviews/current.json, reviews/rounds/, reviews/summaries/
   • Load config from .review-by-opp.json if it exists, otherwise use defaults
   • Write initial reviews/current.json with:

     {
       "session_id": "<uuid>",
       "started_at": "<iso-timestamp>",
       "updated_at": "<iso-timestamp>",
       "current_round": 0,
       "max_rounds": 4,
       "findings": [],
       "rounds": [],
       "final_verdict": null,
       "config": { ... }
     }
     

3. Check for Codex:

   • Run npx @openai/codex --help to verify Codex CLI is available
   • If not installed, tell user: "Install Codex: npm install -g @openai/codex then npx @openai/codex login"

4. Report status:

   • Show session ID, max rounds, review scope, and blocking severities
   • Show Codex detection status
   • Show Claude model/effort: "user-controlled (via Claude Code UI)"
   • Show Codex config: model strategy, reasoning effort

5. Set the rules: Tell the user:

   Review ledger session started. Rules:

   • After implementing changes, run /review-by-opp:review to get Codex feedback
   • Use /review-by-opp:fix to address findings
   • Use /review-by-opp:status to check progress
   • You cannot finalize until all blocking findings are resolved
   • Use /review-by-opp:finalize when ready to close

CRITICAL: You are the builder. Codex is the auditor. You must not claim work is complete while blocking findings remain open. The stop gate enforces this.

$ARGUMENTS