Daily workflow

By the time you reach this chapter you have, in your hands, every move Codex makes on budgetcli. You’ve set the approval and sandbox dials, written the rules that keep the money bug from coming back, dialled effort up and down, pushed noisy work into subagents, packaged the per-bank CSV import as a skill, reached an external API over MCP, gated the dangerous edits with hooks, and let the whole thing run headless once a month with nobody watching. The primitives are all learned. Nothing below is new.

What hasn’t happened yet is running them together, cold, on the clock. That’s still the same villain from chapter 1, in a new shape: Codex starts this session, like every session before it, knowing nothing about budgetcli that isn’t sitting on disk waiting to be read. What’s different by now is how much is sitting on disk - an AGENTS.md it reads before your first prompt, a profile that names a trust posture instead of making you re-derive one, a skill that fires the moment it recognises a bank export. This chapter doesn’t add a fifth thing to that pile. It runs the first four as reflexes instead of decisions, on the exact job the course opened with, and reports honestly on what happens.

The target, stated as a number

Chapter 1 made a promise before you’d installed anything: onboarding a fourth bank’s CSV export costs 11 turns on a cold Monday, and somewhere in those turns Codex reintroduces the float-money bug three separate times. By Friday, the same job costs 2 turns, and the bug doesn’t come back once. That’s this chapter’s actual target, stated as a number so it can be checked rather than taken on faith:

Run the fourth-bank job again - on a repo Codex still has no memory of opening - and land at 2 turns, 0 reintroductions, or explain honestly where it falls short.

(Same two estimators as the rest of the course - 10 tokens per line of code, 4 characters per token of prose - covered once in the opening chapter and not repeated here.)

The stack, cited back to where it was built

Four things get stacked, and every one already exists. This chapter builds nothing new; it just runs what’s already there without you having to assemble it by hand:

• Make the profile switch a reflex - a saved build profile from chapter 3, plus the AGENTS.md from chapter 4, take the job from 11 turns to the 6 chapter 7 already measured. The two-profile habit drilled here is a wider discipline than that one measurement.
• Brief it once, sharply - the skill from chapter 7 and the effort dial from chapter 5 take it the rest of the way, from 6 to 2.
• Recover when the stack doesn’t hold - the honest half: what a bank the skill never saw actually costs, and a pass/fail reading against the target above.
• Close the course - the full ledger replayed in one table, and the spoiler reconciled with the arithmetic shown.

None of it is hard, and none of it is new information - you’ve done every one of these moves at least once already. The only thing this chapter teaches is the difference between knowing a move and reaching for it without being asked.

Start with the cheapest rung to build and the one that pays the most: make the profile switch a reflex.