The Codex Top-1% Playbook — the 9 moves on one page each: the 4-part brief, the AGENTS.md contract, the model-x-effort money matrix, the permissions profile, and the bug that shipped with 16 green tests

The companion to the video. Codex is now included in the free ChatGPT plan (July 9, 2026 unified app — honest fine print inside: OpenAI publishes no Free-tier limits, cloud features need paid), and OpenAI's own paper claims 97.9% of their employees run it. This playbook is the nine-move system, every claim verified against learn.chatgpt.com or our logged runs on August 13, 2026. INSIDE: (1) THE MAP — what Codex actually is since July: six doors into one brain, the Sol/Terra/Luna lineup, and the Aug 31 GPT-5.4 retirement. (2) THE 4-PART BRIEF — Goal/Context/Constraints/Done-when with the exact filled brief from our Run B, plus OpenAI's own anti-pattern (asking for upfront plans causes premature stopping) and the official 'interview me' trick. (3) THE AGENTS.MD OPERATING CONTRACT — copy-paste template, commands first, the hierarchy and override rules, the 32KiB cap, and the known global-file bug. (4) THE MONEY MATRIX — Sol $5/$30, Terra $2/$12, Luna $0.20/$1.20 per 1M; Plus per-5h ranges (10-100 Sol vs 250-2,000 Luna); the effort dial you can step mid-session. (5) THE PERMISSIONS PROFILE — sandbox x approval pairings and the config.toml with the one line that blocks Codex from ever reading your .env files. (6) STEER, DON'T RESTART — Enter steers the current turn NOW, Tab queues for the next, /fork before risky directions. (7) PARALLEL LANES — /worktree, the 3-5 concurrent ceiling power users report, cloud best-of-3 with --attempts, and the .codex/setup.sh gotcha. (8) SKILLS & AUTOMATIONS — $skill-creator, Record & Replay, and the automate-only-what-you-trust rule. (9) THE VERIFICATION LOOP — /review, cross-vendor review, /compact early. PLUS THE RECEIPTS: our pre-registered A-D runs, same repo, same task — including the honest finding (the lazy prompt did NOT produce worse code; the brief buys YOUR interface instead of the model's invention), the ~19-cent-vs-1.2-cent cost gap for the same shipped feature, and the Terra run that passed 16 of 16 of its own tests while scheduling a Jan-31 monthly task for March 3rd. Independent playbook from Hyperautomation Labs — not affiliated with OpenAI.

By submitting, you'll be subscribed to the Hyperautomation AI Report newsletter — one practical AI email per week. Unsubscribe anytime.

Free. No spam. Unsubscribe anytime.