Build a diet plan from your targets

You finished your last bulk and the new cut starts in two weeks. You know roughly the calories and protein you want to land on. Instead of opening a builder, you sketch the plan with the agent in one or two sentences and let it stand up the structure.

What the agent needs to hear

Behind your sentence, the agent works backwards to four pieces: the plan name, the macro targets (or a goal that triggers auto-suggest), the cycle length, and the meal structure per day. Only the first two are mandatory — cycle defaults to 7 days, meal structure to three meals (breakfast, lunch, dinner) unless you override.

Naming carries no friction: « cut Q2 », « lean bulk », « maintenance — May » all work. The agent uses the name later when you have several plans saved. Overrides go in the same sentence: « 5-day cycle, 4 meals per day ». The agent shows a preview card with targets, cycle laid out as Day 1 through Day N, and the meal slots — and waits for a nod before saving.

Targets, structure, and what auto-suggest fills in

Two ways to land the macro targets. Explicit, when you know the numbers: « 2,400 kcal, 180 g protein, 240 g carbs, 80 g fat » goes in as written. Auto-suggest, when you don’t: « build me a cut plan, suggest the macros » tells the agent to read your goal, body stats, and training experience, then propose a starting point you can accept or nudge.

Cycle length is the rotation the plan repeats. Seven days is the default. A 5-day work-week cycle (weekends off-plan) is common for shift workers; a 14-day rotation makes sense if you alternate heavy and recovery weeks with different splits.

Meal structure is the slots inside each day. Three is the default. « 3 meals plus a post-workout snack » creates four slots with the snack tagged by timing; « 5 small meals » spreads things out; « intermittent — 2 meals only » compresses the window. The agent names slots in time order on the preview.

What the plan becomes

A plan is a template, not active state. Creating it doesn’t affect anything on your daily view — no adherence scoring, no meal queue, no targets in live progress. The plan sits in your library until you activate it.

Once active, the plan does four things at once: it populates the daily meal queue so log-from-plan can pull a slot instead of starting from scratch; it enables adherence scoring that compares every meal log against the plan; it feeds the weekly review summarizing how close your actuals tracked the cycle; and it sets the override behavior for off-plan days, which still log normally but get tagged out of the adherence percentage.

The plan is forward-looking. Editing it later — bumping protein, adding a slot, swapping a day — changes the template from that point forward; it doesn’t rewrite past logs. So the targets and structure should reflect how you’ll actually eat, not how you’d eat in an idealized week.

When the agent gets it wrong

Always check the preview card before nodding through. Three failure modes worth catching: a target macro mistyped (« calories were 1,800, not 18,000 »), the wrong cycle length (« 7-day cycle, not 5 »), or a meal structure that doesn’t match what you said (« three meals plus one snack, not four meals »). Each one is a single sentence to fix.

If you want to add a slot you forgot to mention, say so before activating: « also add a mid-morning snack between breakfast and lunch ». The agent inserts it into the time-ordered structure and re-renders. Same mechanic for removing or renaming a slot.

What makes the plan worth keeping

Three things decide whether this plan is useful when you activate it: targets are realistic for the goal (a cut plan that’s too aggressive becomes a streak-breaker after a week), structure matches how you actually eat (three meals if you eat three, five if you eat five — not the program book’s prescription), and cycle length reflects your real rotation (7 days for most, 5 + 2 for work-week eaters, 14 for longer rotations). The system reads the plan as the source of truth for adherence and live progress; an unrealistic plan corrupts both. Sketch it the way you’ll actually live it, then activate when you’re ready.