shield_lock
JinDaGe
← Back to skills
extension
Category: Development & EngineeringNo API key required

codex-think-about

Peer debate between Claude Code and Codex on any technical question. Both sides think independently, challenge each other, and converge to consensus or explicit disagreement.

personAuthor: jakexiaohubgithub
open_in_newRepositorydownloadDownload package

Codex Think About

Purpose

Use this skill for peer reasoning, not code review. Claude and Codex are equal analytical peers; Claude orchestrates the debate loop and final synthesis.

Prerequisites

  • A clear question or decision topic from the user.
  • codex CLI installed and authenticated.
  • codex-review skill pack is installed (npx github:lploc94/codex_skill).

Runner

RUNNER="{{RUNNER_PATH}}"

Workflow

  1. Ask user to choose reasoning effort level: low, medium, high, or xhigh (default: high). Gather factual context only (no premature opinion). Set EFFORT.
  2. Build round-1 prompt from references/prompts.md.
  3. Start Codex + Claude Independent Analysis (parallel): a. Start Codex thread: node "$RUNNER" start --working-dir "$PWD" --effort "$EFFORT" --sandbox danger-full-access. b. Claude Independent Analysis (IMMEDIATELY, before polling): Analyze the question independently using own knowledge and optionally MCP tools. Follow the structured format in references/claude-analysis-template.md. Complete this BEFORE reading any Codex output. See references/workflow.md Step 2.5 for detailed instructions. c. INFORMATION BARRIER: Do NOT read $STATE_DIR/review.md or interpret Codex's conclusions until Step 5. Poll activity telemetry (file reads, URLs, topics) is allowed for progress reporting.
  4. Poll Codex with adaptive intervals (Round 1: 90s/60s/30s/15s..., Round 2+: 45s/30s/15s...). After each poll, report specific activities from poll output. See references/workflow.md for parsing guide. NEVER report generic "Codex is running".
  5. Cross-Analysis: After Codex completes, compare Claude's independent analysis with Codex output. Identify genuine agreements, genuine disagreements, and unique perspectives from each side. See references/workflow.md Step 4.
  6. Resume via --thread-id and loop until consensus, stalemate, or hard cap (5 rounds).
  7. Present user-facing synthesis with agreements, disagreements, cited sources, and confidence.

Effort Level Guide

| Level | Depth | Best for | |----------|-------------------|---------------------------------| | low | Surface check | Quick sanity check | | medium | Standard review | Most day-to-day work | | high | Deep analysis | Important features | | xhigh | Exhaustive | Critical/security-sensitive |

Required References

  • Execution loop: references/workflow.md
  • Prompt templates: references/prompts.md
  • Output contract: references/output-format.md
  • Claude analysis format: references/claude-analysis-template.md

Rules

  • Keep roles as peers; no reviewer/implementer framing.
  • Codex must NOT modify, create, or delete ANY project files. danger-full-access sandbox is used SOLELY for web search. Prompt contains strict guardrails.
  • Codex MUST cite sources (URL) for factual claims from web.
  • Separate researched facts (with sources) from opinions.
  • Detect stalemate when arguments repeat with no new evidence.
  • End with clear recommendations, source list, and open questions.
  • Information barrier: Claude MUST complete its independent analysis (Step 3b) before reading Codex output. This prevents anchoring bias.