# Plan & Critique (Turn 3+) > Transcribed from upstream [nexu-io/open-design `discovery.ts:155-195`](https://github.com/nexu-io/open-design) (Apache 2.0). See [../ATTRIBUTION.md](../ATTRIBUTION.md). Once the design-system / direction / brand-spec is locked, **the first tool call is `TodoWrite`**. The chat renders this as a live "Todos" card — it is the user's primary way to see the plan and redirect cheaply. The plan has 5–10 short imperative items in the order they'll execute. Mark items `in_progress` and `completed` live as you go — do NOT batch updates at the end of the turn. ## Standard plan template > **HARD RULE — Step 5 is per-section, not per-page.** A "landing page" is **4-6 separate `od_generate_design` calls** (hero, feature grid, how-it-works, callout, footer-cta…), not one monolithic call. A pitch deck is one call per slide-group of 3-5 slides, not one call for all 12 slides. **Split by *scope* (what the prompt asks for), not by `maxTokens` (the output ceiling).** Keep `maxTokens` at the default 64,000 — lowering it risks truncating the model's output mid-section, which is far worse than a "small" prompt. The reason to split is **request size**, not **response size**: asking for one section at a time lets the model focus, lets you critique each piece independently, makes failures recoverable (retry one section, not a whole page), and lets the user redirect mid-flight via the live todo list. Each focused call typically takes ~10-60s and well under 8k tokens of actual output even with the 64k ceiling. Adapt the middle steps to the brief, but keep steps 1, 7, 8, 9 — they are non-negotiable: ``` 1. Read active DESIGN.md + skill assets (template.html, layouts.md, checklist.md) 2. Bind tokens to :root - (Branch A) Confirm brand-spec.md + bind to :root - (Active DESIGN.md) Bind active design-system tokens/rules to :root - (Branch B) Bind the chosen direction's palette + type stack to :root 3. Plan section/slide/screen list with platform variants and rhythm (state list aloud before writing — "We'll build hero, features×3, pricing×3, FAQ×5, footer") 4. od_create_project { id: "", name, customInstructions: } (if no projectId yet — `id` is REQUIRED) OR identify the existing projectId from context 5. For each section/screen: od_generate_design { projectId, prompt: "" } The projectId ensures every section auto-inherits brand rules (issue #37) 6. Replace any [REPLACE] placeholders with real, specific copy from the brief Honest placeholder beats invented stat 7. Self-check: P0 anti-AI-slop checklist (full list in design-philosophy.md §C) Every P0 must pass. Fix before moving on. 8. 5-dim critique: score yourself silently 1–5 on philosophy / hierarchy / execution / specificity / restraint Fix any dimension < 3/5. Two passes is normal. 9. od_lint_artifact { html } + od_save_artifact { identifier, title, html } for each section Catch malformed HTML before persisting. ``` ### Step 5b — Merge sections locally with `Write` / `Edit` After Step 5 produces N section HTML fragments, **assemble them locally** — do NOT spend an `od_generate_design` call to glue sections together. The orchestrator (or a thin sub-task) builds the final page in three deterministic moves: 1. **Template the shell once** with `Write`: ``, `` with `:root` tokens bound, the consistent `