The rules tell an agent what to do. These files show what "done" looks like.
A placeholder prompt (SUBJECT: [main object]) gets filled in generically, and generic is exactly the AI look the rules exist to prevent. Every prompt in this folder is finished — no brackets, no [BRAND], nothing left for the model to invent.
For generated visual demonstrations, see the README showcase: two presentation boards with exact prompts and fictional concept labels.
| # | File | Brief | Mode | Teaches |
|---|---|---|---|---|
| 00 | 00-anti-examples.md |
— | — | the same brief written weakly vs. properly, and what comes back |
| 01 | 01-restaurant-real-food.md |
coastal taverna, has real dish photos | A — native text | real-food hero, photo+panel layout, in-render spelling |
| 02 | 02-hotel-editorial.md |
seafront hotel, direct bookings | B — deterministic | clean background + placed typography, logo fidelity |
| 03 | 03-services-problem-effect.md |
stove installer, wants autumn leads | B — deterministic | Problem→Effect angle, deadline offer, real install refs |
| 04 | 04-retail-product-in-use.md |
product brand, reference photo supplied | A — native text | reference-as-source-of-truth, product-in-use, series of 5 |
Additional prompt-only teaching cases: model-independent directions covers a typographic flyer, food ad and service ad. These are fictional briefs without rendered results or invented QA scores.
Generated showcase walks the full research → questions → prompt → fal.ai generation → analysis loop for the README images.
2026 style directions pairs a static format with a visual language from the style atlas: performance sticker callouts, direct-flash editorial, a notes-app confession, a colour-block stat drop and a seasonal offer, plus one Andromeda-ready campaign set. Also fictional, prompt-only and unscored.
October 2026 craft styles works the 6.6.0 refresh: proof of process for a bakery, cut-paper collage for a drink and a contact sheet for a sock range, each with its decayed-cue check (R53), remix note (R54) and what must come from real material (R55). Prompts in prompts/craft-collage-styles.txt, linted.
1. BRIEF — what the client actually said, including what they got wrong
2. CREATIVE WORK — R28 steps 1–10, worked, with the reasoning visible
3. HEADLINE — five drafts, three killed, one shipped, and why
4. THE PROMPT — the finished prompt, copy-pasteable, zero placeholders
5. WHAT CAME BACK — first-pass result and its QA score
6. THE FIX — what changed on the second pass and why
7. WHAT TO STEAL — the transferable part
The briefs, prompts, headlines and QA verdicts here are composites drawn from real production campaigns (coastal hospitality and local-trade clients), reconstructed so they can be published without exposing client assets. The rejection modes are the real ones — every "what came back" failure in these files is a failure that actually happened and got fixed. No rendered images are committed, because the client photography they were built on isn't ours to publish.
Use them as templates for structure and specificity, not as claims about any named business.