Codex 102 for Academic Researchers (Bilal)

  • Author/Source: Mushtaq Bilal, X thread

  • Original: https://x.com/MushtaqBilalPhD/status/2055223177001726349

  • Key Ideas

  • Sequel to Codex 101. Covers five topics for scaling Codex from short projects to multi-year work: (1) structuring a long project; (2) Plan Mode and Skills; (3) custom agents for parallel research tasks; (4) connecting Codex to other apps; (5) hooks and automations.
  • Same folder taxonomy as the Claude Code 102 article (Literature / Chapters / Data / Notes / Correspondence), with AGENTS.md replacing CLAUDE.md as the per-folder instruction file.
  • Introduces Codex's Plan Mode and Skills (Codex equivalents of Claude Code's plan mode and skill bundles), custom agents for parallelizing different research tasks, MCP connections to external apps, and hooks for automation.

  • Maintains the deliberately non-technical voice: "if you can write sentences in English, you can use Codex."

  • Summary

A direct Codex translation of the Claude Code 102 organization pattern, plus a survey of Codex-specific features (Plan Mode, Skills, custom agents, MCP, hooks). The organization advice is portable across both ecosystems — the same folder layout, the same nested-instruction-file principle — and the comparison is useful for researchers choosing between Codex and Claude Code: most of the workflow conventions transfer.

The introduction of hooks and automations in a beginner-friendly tutorial is notable; Codex's hook system enables time-based and event-based automation (e.g., periodic checks for new bugs, automatic responses to incoming files) that Claude Code currently exposes only to more advanced users. For long-running research projects, this could be the differentiating Codex feature for academics willing to set up scheduled tasks (literature alerts, citation re-verification, draft auto-review).

  • Relevance to Economics Research

Useful for any economist running a multi-year empirical project on the Codex stack. The hooks/automations section in particular is worth a closer read — automating "every Friday, re-run the literature search and flag new papers citing X" or "after each data refresh, re-run integrity checks" is the kind of cadence-based maintenance most empirical projects skip because it would otherwise need a cron job and a Python wrapper. Codex hooks remove the wrapper.