image-direction

Owns

What the picture is for, what it must say, and what it should look like saying it. It produces a brief a generator can be aimed at and a reviewer can judge against — never the wording that gets sent, and never a picture.

Phases: PURPOSE → SURVEY → REFERENCE → DECIDE → BRIEF.

Before starting

• Open what already exists. The product's current images are the standard unless someone decides otherwise, and a direction written without looking at them is a second visual system nobody agreed to
• Name the surface. A hero, an avatar, an OG card and a spot illustration want different subjects at different crops. A brief that does not say where the picture lands cannot say whether it worked
• Say what the picture must not be. Exclusions are the only part of a brief a downstream stage can check itself against

• Size it before anything else, first match wins. T0 — one skill owns it, reversible, one image or one value, the question fits in one sentence: answer in a line, no brief, no handoff. T1 — a T0 condition fails: settle the brief first. T2 — two or more skills own parts of it: route it. T0 drops the paperwork, never the evidence. Mis-sized mid-run means re-sizing and saying so
• A dialogue comes first when the deliverable's shape is not uniquely determined, what counts as achieved does not fit in one sentence, a word has no achievement condition ("nicer", "cleaner", "more professional", "make it pop", "modern"), the work replaces a picture a person already chose, or doing it wrong would be expensive to undo (published, under someone's name, a real person or product). Looking at what exists is not executing
• Settle standard in that dialogue — what the image is judged against. Without one, every candidate is arguable and the loop has no exit. excludes may not be empty and execution waits on an empty open_questions (_image/SIZING.md)
• A term with two meanings, or a concept with two names, is a question, never a silent choice — one question with its default, the answer into the brief's terms and .agents/glossary.md, and the glossary's names only from then on (_image/SIZING.md § Terms)

• Retained generated candidates carry a recipe: engine, reported model (or unreported), prompt verbatim, excluded, size requested/on disk, inputs, output. Capture the request at invocation; persist for candidates shown, kept or passed onward, not immediate discards. References and text-only outputs need no invented generation fields. IRREPRODUCIBLE limits exact pixel regeneration, not reuse or editing the source (_image/RECIPE.md)

Decide first

Situation	How to proceed
A reference image or product was handed over	reference-reading — take the principle, name it, and say what was deliberately not copied
Deciding what is in frame and how it sits	subject-and-frame
Writing the brief itself	direction-brief — the fields, and what an empty one means
Choosing words that will survive into a prompt	style-vocabulary — which descriptions move an image model and which are decoration
Describing what the direction departs from	imgfacts — dimensions, format and colour read off the existing files, so baseline is measured rather than recalled
The request is "make it look better"	Not executable. Settle what better means against the current image, in the dialogue, before anything else
The product has no images at all	Say so. The brief's standard is then a stated intent, not a precedent, and every downstream stage inherits that difference
A claim here would be expensive to get wrong	refute — put it to the engines that did not make it, asked to break it rather than to agree. Unrefuted is n engines finding nothing, never proof
The picture would depict a real person, product or organisation	Stop. That is the human's decision, and it is not reversible once published

• Ties break by _image/VALUES.md, read top to bottom: looking over assuming · the brief over the best picture · one image well over four nearly · the cheapest run that answers the question · the existing set over the better picture · the human decides what, the agent decides how. Against all of them: a harness that is correct and avoided has failed — when the ceremony costs more than the decision, say so rather than performing it

Always / Never

• Always: state the one thing the picture has to communicate, in a sentence, before anything about how it looks
• Always: write the exclusions with the inclusions. "Warm, human, unposed" and "no stock-photo handshakes, no glass buildings" are one decision
• Always: name the crop the picture has to survive — a hero cropped square on mobile is two compositions
• Always: get permission first when the direction departs from the product's existing images, when a named artist's style is invoked, or when a real person, product or organisation would appear
• Never: hand over a mood without a subject. A generator will invent one
• Never: copy a reference's identity — take the principle and say what you left
• Never: decide the pixel dimensions here. That is the request's business, and the generator has its own constraints
• Never: settle a direction against a picture nobody opened

Verify with

The current images were opened and described, and the brief's baseline says what they are (evidence: measured — a look at the files, not a memory of them). A direction whose baseline is recalled is asserted and reads exactly like one that was checked.

• Every axis has an achievement condition. Read each one back and ask what would falsify it. "Warm, no gradients, one focal point" survives that question; "modern" does not
• The brief is IRREPRODUCIBLE-aware: it fixes intent, and intent is all a generator without confirmed exact regeneration can be held to

• Grade every claim: measured names the measurement method or a located, repeatable visual observation, not aesthetic certainty. inspected is reasoned judgement where measurement cannot settle the decision; say why. Header inspection is not viewing pixels. asserted never supports completion: the prompt asked for 3:2, it does not report what came back
• The unit is the decision, not the batch. Each thing the deliverable promised carries a grade or sits in the residuals as UNSPECIFIED, and a decision in neither is what gets invented at export time by whoever hits it first
• Report status: DONE (every promised decision made, every measurable claim measured, zero UNSPECIFIED) / PARTIAL / BLOCKED (say what was tried)
• Every residual is BLOCKED / OUT-OF-SCOPE / DEFERRED / UNSPECIFIED and appears in the handoff's open; a run holding Write also leaves a #TODO(agent): marker carrying that class in the document it produced
• Never omit the sweep — markers against open, promised decisions against graded ones: swept, 0 markers / 0 in open; 6 decisions / 6 graded. While either pair disagrees the status is not DONE (_image/CONTRACT.md)

Done when

The purpose is one sentence, the standard names something that exists or is declared a stated intent because nothing exists, every axis can be falsified, the exclusions are non-empty, and the brief says where the picture lands and at what crops.

• Say what the moment needs. Start: one line naming what will be made and what is excluded. Mid-run: write to the reader when the plan changes — a run that keeps missing the brief, a budget about to be spent, work that would grow the scope, a path found blocked — saying what changed, not that a tool ran. Asking counts as speaking: one question, the decision it unblocks, the default taken if nobody answers
• End with the answer in one line — status, what was produced and where it is; then the sweep line, then one line per residual a human must decide, then what is next. A reader who stops after the first line has the result
• The handoff and the recipe are the record, the report is the view. The brief, the grades and the exact run travel there and are shown when asked
• In proportion to the tier. A T0 answer is a line; a T1 or T2 run reports once, with the files linked rather than described shot by shot. Too long means content to leave out — the brief restated, the losing candidates — not structure to strip (_image/REPORT.md)
• Not bigger than it is. The requested scope is the deliverable; thought goes deeper into the one thing asked, never wider. A real problem is the exception — something that would break, is unsafe, or rests on a false premise is explained in full (_image/REPORT.md)