@ericrisco/rsc 3.1.3 → 3.3.0

package/skills/design-dna/SKILL.md

@@ -1,162 +0,0 @@
----
-name: design-dna
-description: "Use when a design has won and its look must survive: codify it into a permanent style skill with measured ratios, bans and tests that can fail, then build new pieces in that exact look. Fires on 'turn this into a skill', 'keep the same style', 'my output drifted'. NOT finding the winner (that is `design-loop`), NOT the words (that is `marketing`)."
-tags: [design, style, dna, codify, skill-factory, consistency]
-recommends: [design-loop, design, nextjs, marketing]
-origin: risco
----
-
-# Design DNA - one beautiful design into a permanent skill
-
-*A design loop finds you one beautiful thing. This turns that one thing into a skill you keep forever, so you can make more of it, in any format, without the look drifting.*
-
-**Not only about "designs".** The subject is anything that got built: a website, a landing page, a poster, a motion graphic, a carousel, a deck. Same system, same file, same tests. The format is an output, not the subject.
-
-## The problem this solves
-
-Nobody wrote down **why** the winner was beautiful. Next time you ask for "the same style" you get something close, then something a bit off, then something generic. Every rule you do not write down, the model has to guess, and it does not guess randomly: it guesses the average of everything it has ever seen. That average is what people call slop. Slop is the sound of an unwritten rule.
-
-## Two modes
-
-| You say | Mode | What happens |
-|---|---|---|
-| "codify this", "turn this into a skill", here is the winner | **CAPTURE** | Run the seven steps below. Output: a new style skill folder. |
-| "build X in <style>", "another one like the last", `/<slug>` | **REUSE** | Load that style's `PROMPT.md` only, build, then run its self-check before returning. |
-
-If a `dna.json` already exists for the style named, you are in REUSE. Never re-derive a style that has a record.
-
-To find out which records exist — in either scope, without guessing at paths — run
-`npx @ericrisco/rsc doctor --json` and read `designStartingPoint`. A record it reports as `unreadable`
-is **not** a record you can reuse: say what is missing, do not use half of one.
-
-## Hard rules (these are the skill)
-
-1. **A specification that cannot fail is not a specification.** If every rule you wrote is one a bad output could still satisfy, you wrote a mood board.
-2. **Compile, do not paste.** Two files, always. `dna.json` is the record, any size, and it **never** enters a prompt. `PROMPT.md` is the payload, **hard cap 2KB**, and it is the only thing that does.
-3. **Always attach the reference image.** Highest-leverage move available and close to free: image conditioning runs on its own pathway, so it does not eat the attention your words are competing for.
-4. **Ratios, not values.** "The headline is 96px" is nearly worthless. "The headline is 8x the body, never under 6x" is the identity.
-5. **Coverage, not just colour.** The same three hex codes at 60/30/10 and at 90/8/2 are two unrelated designs.
-6. **Percentages, not pixels,** in everything spatial. One spec has to drive a 1080x1350 carousel and a 1920x1080 slide with no rewrite.
-7. **When output looks generic, add a ban, not a rule.** A positive instruction is one weak vote against a landslide; "never centre the hero" deletes the landslide from the options. Bans should outnumber positive style rules.
-8. **One skill per style.** Never merge two identities into one spec. The average of two good designs is a bad design.
-9. **Reproduce the system, never the marks.** Real logos, wordmarks, licensed photographs and proprietary typefaces go in `meta.not_copied` with a named substitute.
-10. **Step 5 is not optional.** A spec that has never been used to rebuild its own source has never been tested.
-
-## CAPTURE - the seven steps
-
-Run `PROMPT.md` as written. It is the operative artefact and it is self-contained; this section is the map, not a second copy of it.
-
-| Step | Name | Non-negotiable output |
-|---|---|---|
-| 1 | **Observe** | Flat inventory of literal measurements. Sampled hexes, coverage estimates, type counts, largest-to-smallest ratio, margins as percentages, texture, and **what is absent**. No judgements: interpreting here is how you end up specifying a design that is not the one in front of you. |
-| 2 | **Debate** | Loop A argues exhaustive, Loop B argues nine moves and a wall of refusals, then adjudicate each property with one question: *if I changed this value, would it stop looking like the reference?* Yes is load-bearing, no is trivia, and trivia dilutes attention. Resolve it by splitting documents, not by compromising. |
-| 3 | **Codify** | `dna.json` against `design-dna.schema.json`. Descriptive colour names ("dusty plum", never "accent-500": image and video models cannot read token names). Every family gets a fallback. |
-| 4 | **Tests that can fail** | 8 to 12, binary and measurable, each marked auto or human. |
-| 5 | **Reconstruct and diff** | Close the reference. Rebuild it from `dna.json` alone. Put them side by side and list **every** difference: each one is a field the spec forgot. Fold them in, record them in `reconstruction.gaps_found`, go again. Expect two or three passes, and expect the gaps to be things you were sure were obvious. |
-| 6 | **Emit the skill** | The folder below. |
-| 7 | **Declare uncertainty** | Every value inferred rather than measured, and every rule under 70% confidence, into `meta.confidence_notes`. That is where the style drifts first. |
-
-### What gets emitted
-
-    ~/.claude/skills/<slug>/
-      SKILL.md          how to use this style, and when not to
-      PROMPT.md         the 2KB payload. THIS is what enters a context window
-      dna.json          the full record. NEVER pasted into a prompt
-      design-dna.schema.json -> validated against the parent copy
-      reference/        the original, kept forever
-      example/          one worked output, the canonical proof
-      scripts/check.py  the auto tests, exits non-zero on failure
-
-### The record is the harness's design identity
-
-The folder above makes the style **reusable**. Writing the record into the harness is what makes it
-**this project's identity** rather than a loose folder in a global directory:
-
-- Write the record as an article under `02-DOCS/wiki/brand/`, beside the brand study, and link it
-  from `02-DOCS/wiki/index.md` and the root `CLAUDE.md` `## Brand & voice` section — the same
-  convention `design` follows, and the same place `marketing`, `presentations`, `brand-voice` and
-  `content-engine` already read.
-- **It does not replace the brand study.** The study is the judgement and the direction — why this
-  identity, what it refuses. The record is the measurable contract that can fail a test. They share
-  only a few fields; each carries what the other cannot.
-- **One harness, one design identity.** If a second appears, name the conflict. The average of two
-  good designs is a bad design.
-- A style captured before this convention still works from its folder. To bring it into the harness,
-  write its record as the article above; nothing needs re-deriving.
-
-`rsc doctor` reports whether this harness has an identity, so the rule is not just prose.
-
-Scaffold from `templates/` and validate before you claim it works:
-
-    python3 ~/.claude/skills/design-dna/scripts/emit.py <slug>            # scaffold + validate
-    python3 ~/.claude/skills/<slug>/scripts/check.py <output-file>        # run the auto tests
-
-### PROMPT.md ordering is fixed
-
-Attention is strongest at the two ends and weakest in the middle, so the order is a mechanism, not a style choice:
-
-1. The reference image, attached and named first
-2. `soul.one_line`
-3. The weird move, alone, unmissable
-4. The 3-9 signature moves, as ratios
-5. The bans, as absolutes
-6. Palette and type: roles and coverage only, never a full ramp
-7. Archetype names and when to use each
-8. The self-check, last
-
-Nothing else. If it does not change what the output looks like from three metres away, it belongs in `dna.json`. To add a tenth signature you must delete one.
-
-Last line of every `PROMPT.md`, verbatim:
-
-> Before returning any output, run every test in the self-check. Name each test and its result. If any fails, repair the output and run them again. Never return output with a failing test and a note explaining it away.
-
-## REUSE - building with a captured style
-
-1. Load **only** that style's `PROMPT.md`. Attach `reference/`. Do not read `dna.json` into context; read it only to answer a specific measured question or to regenerate the payload.
-2. Pick an archetype by name before writing anything. Structure first: unnamed layouts are why piece 8 does not sit beside piece 1 as a set.
-3. Build. For web, hand the build to `../nextjs/SKILL.md` and the words to `../marketing/SKILL.md`; the identity stays here.
-4. Run the self-check, name each test and its result, repair and re-run on any failure.
-5. If output drifts, do not add a rule. Consult the table below.
-
-## When it goes wrong
-
-| What you see | Cause | Fix |
-|---|---|---|
-| Matches every value, still looks generic | Signature moves missing, or too many | Cut to 3-9, rewrite each as a ratio |
-| Drifts back to the stock AI look | Not enough bans | Bans should outnumber positive style rules |
-| Piece 1 and piece 8 do not match | No named layouts | Add archetypes |
-| The accent reads as a theme, not an accent | No coverage percentages | Add them |
-| Works on a slide, breaks on a carousel | Pixel values in the spacing | Convert to percentages |
-| Obeys some rules, ignores others, differently each run | Payload over the cap | Cut to 2KB |
-| Your best rule keeps getting ignored | It is buried in the middle | Move it to the top or the bottom |
-| Fine, but forgettable | No weird move | Find the one break in the system |
-| Spec feels complete, output is wrong | You skipped Step 5 | Rebuild the original from the spec. Compare. |
-
-## Styles captured so far
-
-| Style | Reads as | Weird move |
-|---|---|---|
-| `../cold-press/SKILL.md` | newsprint editorial, serif at 8x, one vermilion | section numeral clipped by the right edge |
-| `../night-shift/SKILL.md` | near-black instrument panel, one phosphor green | status ribbon severed mid-word by the right edge |
-| `../soft-optics/SKILL.md` | warm paper, lowercase display at 6x, one olive | display line crossing a tinted plate's bottom edge |
-
-Each is a finished skill: reference, `dna.json`, a capped payload, a worked proof, and tests that exit non-zero. Never merge two of them: the average of two good designs is a bad design.
-
-## Hand-offs
-
-- **No winner yet, no reference to measure** -> `../design-loop/SKILL.md` first: it takes a real reference and converges on a piece that beats it. Come back with the winner. Capturing an average of several candidates produces the average, which is the thing you are trying to escape.
-- **The words on the page** -> `../marketing/SKILL.md`. `voice` in `dna.json` constrains them; it does not write them.
-- **The build** -> `../nextjs/SKILL.md` (App Router, React 19) or `../react/SKILL.md` (Vite SPA).
-- **SDD gate.** Capturing a style writes a skill, not product code, so it needs no spec. The moment its output becomes a new feature or a non-trivial behaviour change in a real app, `../specify/SKILL.md` runs first.
-
-## Files
-
-| File | What it is |
-|---|---|
-| `PROMPT.md` | The standalone prompt. Paste anywhere, attach the design, done. |
-| `design-dna.schema.json` | Every field of `dna.json`, explained. |
-| `scripts/emit.py` | Scaffolds and validates an emitted style skill. |
-| `templates/` | `SKILL.md`, `PROMPT.md` and `check.py` skeletons for the emitted skill. |
-| `references/method.md` | The full method and the research each decision traces to. Read once, not per run. |
-
-**Why any of this works** is not opinion: lost-in-the-middle recovery, instruction-compliance decay under constraint count, style being formally the correlation between features rather than the features, and decoupled image cross-attention. The citations are in `references/method.md`.

package/skills/design-dna/design-dna.schema.json

@@ -1,230 +0,0 @@
-{
-  "$schema": "https://json-schema.org/draft/2020-12/schema",
-  "title": "Design DNA",
-  "description": "The exhaustive record of one visual identity. Written for you and for build tools, never pasted into a prompt. The prompt payload is compiled from this file into PROMPT.md under a 2KB cap.",
-  "type": "object",
-  "required": ["meta", "soul", "palette", "type", "space", "signatures", "weird_move", "bans", "tests", "reconstruction"],
-  "additionalProperties": false,
-  "properties": {
-    "meta": {
-      "type": "object",
-      "description": "Provenance. What this is, where it came from, and what must never be reproduced.",
-      "required": ["name", "slug", "source", "captured", "not_copied", "medium_of_origin"],
-      "properties": {
-        "name": { "type": "string", "description": "Human name of the style, e.g. 'Cold Press Editorial'." },
-        "slug": { "type": "string", "pattern": "^[a-z0-9]+(-[a-z0-9]+)*$", "description": "kebab-case. Becomes the emitted skill folder name and its slash command." },
-        "source": { "type": "array", "items": { "type": "string" }, "description": "Paths or URLs of the reference artefacts this was measured from." },
-        "captured": { "type": "string", "description": "ISO date the capture was made. Styles age; a dated record tells you when to re-measure." },
-        "not_copied": { "type": "array", "items": { "type": "string" }, "description": "Real logos, wordmarks, licensed photographs, proprietary typefaces. Reproduce the SYSTEM, never the MARKS. Name the substitute used for each." },
-        "medium_of_origin": { "type": "string", "description": "What the reference actually was: website, poster, carousel, deck, motion graphic. The format is an output, not the subject." },
-        "confidence_notes": { "type": "array", "items": { "type": "string" }, "description": "Step 7 output: every value inferred rather than measured, and every rule under 70% confidence. This is where the style drifts first." }
-      }
-    },
-    "soul": {
-      "type": "object",
-      "description": "The compression key. Read this alone and you are most of the way there.",
-      "required": ["one_line", "adjectives", "lineage", "read_distance", "energy"],
-      "properties": {
-        "one_line": { "type": "string", "maxLength": 160, "description": "Max 160 chars. Concrete nouns only. No marketing words, no 'premium', no 'clean'." },
-        "adjectives": { "type": "array", "minItems": 3, "maxItems": 5, "items": { "type": "string" } },
-        "lineage": { "type": "string", "description": "The family tree: Swiss editorial, 90s rave flyer, brutalist web, Memphis, ECM sleeve." },
-        "read_distance": { "enum": ["thumb", "arms-length", "desk", "room", "street"], "description": "How far away it is read from. This sets the minimum type size, so it is a functional field, not a poetic one." },
-        "energy": {
-          "type": "object",
-          "description": "Each 1-10. A fingerprint that catches drift no single value does.",
-          "required": ["density", "variance", "contrast", "warmth"],
-          "properties": {
-            "density": { "type": "integer", "minimum": 1, "maximum": 10, "description": "How much of the canvas is occupied." },
-            "variance": { "type": "integer", "minimum": 1, "maximum": 10, "description": "How much elements differ from each other. Low = systematic, high = collaged." },
-            "contrast": { "type": "integer", "minimum": 1, "maximum": 10 },
-            "warmth": { "type": "integer", "minimum": 1, "maximum": 10 }
-          }
-        }
-      }
-    },
-    "palette": {
-      "type": "object",
-      "description": "Colour AND its proportions. The same three hex codes at 60/30/10 and at 90/8/2 are two unrelated designs.",
-      "required": ["colors", "coverage", "banned"],
-      "properties": {
-        "colors": {
-          "type": "array",
-          "minItems": 2,
-          "items": {
-            "type": "object",
-            "required": ["role", "hex", "name"],
-            "properties": {
-              "role": { "type": "string", "description": "ground, ink, accent, secondary, hairline." },
-              "hex": { "type": "string", "pattern": "^#[0-9a-fA-F]{6}$", "description": "Sampled, never guessed." },
-              "name": { "type": "string", "description": "Descriptive, not systematic. 'dusty plum' travels to an image model; 'accent-500' does not." }
-            }
-          }
-        },
-        "coverage": { "type": "object", "description": "Percentage of canvas per role, summing to roughly 100. The most-skipped and most-decisive field in the whole document.", "additionalProperties": { "type": "number" } },
-        "mode": { "type": "string", "description": "light, dark, or duotone-on-stock." },
-        "banned": { "type": "array", "items": { "type": "string" }, "description": "Colour behaviours forbidden: gradients, two accents in one frame, pure #000000." }
-      }
-    },
-    "type": {
-      "type": "object",
-      "description": "Families, and above all the ratio between them. 8-to-1 and 2-to-1 are different designs with identical fonts.",
-      "required": ["families", "scale", "treatment"],
-      "properties": {
-        "families": {
-          "type": "array",
-          "items": {
-            "type": "object",
-            "required": ["role", "family", "fallback"],
-            "properties": {
-              "role": { "enum": ["display", "body", "mono", "caption", "ui"] },
-              "family": { "type": "string" },
-              "fallback": { "type": "string", "description": "REQUIRED. Silent Arial substitution is the most common way a reproduction dies quietly." },
-              "weights": { "type": "array", "items": { "type": "integer" } },
-              "licence_risk": { "type": "string", "description": "free / webfont-ok / proprietary-substitute-needed." }
-            }
-          }
-        },
-        "scale": {
-          "type": "object",
-          "required": ["display_to_body_ratio", "max_sizes_per_frame"],
-          "properties": {
-            "display_to_body_ratio": { "type": "number", "description": "The typographic signature. Record it as a floor too ('8x, never under 6x')." },
-            "ratio_floor": { "type": "number" },
-            "steps": { "type": "array", "items": { "type": "number" }, "description": "Relative steps, not pixels." },
-            "max_sizes_per_frame": { "type": "integer", "description": "Usually 2 or 3. A testable ceiling." }
-          }
-        },
-        "treatment": {
-          "type": "object",
-          "properties": {
-            "display_tracking": { "type": "string", "description": "e.g. '-0.03em'." },
-            "body_leading": { "type": "string", "description": "As a multiple of size." },
-            "case": { "type": "string" },
-            "measure": { "type": "string", "description": "Characters per line. Under 65 for body is the usual bound." },
-            "numerals": { "type": "string", "description": "lining / oldstyle / tabular." }
-          }
-        }
-      }
-    },
-    "space": {
-      "type": "object",
-      "description": "USE PERCENTAGES. One spec must drive a 1080x1350 carousel and a 1920x1080 slide without a rewrite. Pixel values here are the reason a spec works on a slide and breaks on a carousel.",
-      "required": ["grid", "margin_pct", "alignment", "negative_space"],
-      "properties": {
-        "grid": { "type": "string" },
-        "margin_pct": { "type": "number" },
-        "gutter_pct": { "type": "number" },
-        "alignment": { "type": "string", "description": "Where things line up, and what is deliberately never centred." },
-        "rhythm": { "type": "string", "description": "The spacing progression, e.g. 'x1.5 geometric off a 8px base'." },
-        "negative_space": { "type": "string", "description": "Where the emptiness is and how big. Emptiness is a placed element, not leftover." },
-        "safe_area": { "type": "object", "description": "Per platform, in percentages.", "additionalProperties": { "type": "string" } }
-      }
-    },
-    "surface": {
-      "type": "object",
-      "description": "What the thing is physically made of.",
-      "properties": {
-        "texture": { "type": "object", "properties": { "kind": { "type": "string" }, "opacity": { "type": "number" }, "scale": { "type": "string" }, "stacking_order": { "type": "string" } } },
-        "edges": { "type": "string", "description": "Radii, hairline weights, whether anything bleeds off canvas." },
-        "elevation": { "type": "string", "description": "Often the honest answer is 'none'. Record the absence." },
-        "imagery": { "type": "string", "description": "Treatment: duotone, grain, crop discipline, subject distance." },
-        "iconography": { "type": "string", "description": "Or the absence of it." }
-      }
-    },
-    "signatures": {
-      "type": "array",
-      "minItems": 3,
-      "maxItems": 9,
-      "description": "The payload. Carry these and nothing else and it is still recognisable. To add a tenth you must delete one; the cap forces a decision instead of shipping indecision.",
-      "items": {
-        "type": "object",
-        "required": ["move", "how", "when", "never"],
-        "properties": {
-          "move": { "type": "string", "description": "Name it like a technique, so it can be referred to in one word." },
-          "how": { "type": "string", "description": "AS A RATIO OR RELATIONSHIP. Style is formally the correlation between features, not the features." },
-          "when": { "type": "string", "description": "Where in a piece or a sequence it appears." },
-          "never": { "type": "string", "description": "How it gets used wrong. The failure mode, written down." }
-        }
-      }
-    },
-    "weird_move": {
-      "type": "object",
-      "description": "The single deliberate break in the system. Its own key, deliberately, so it can never get buried in the middle of a document. Highest-information element present and the first thing a mechanical extraction loses.",
-      "required": ["what", "how", "why"],
-      "properties": {
-        "what": { "type": "string" },
-        "how": { "type": "string", "description": "Reproducible instruction, as a relationship." },
-        "why": { "type": "string" },
-        "frequency": { "type": "string", "default": "exactly once per piece" }
-      }
-    },
-    "archetypes": {
-      "type": "array",
-      "description": "Named layouts. Without these you get consistent styling and inconsistent structure, which reads as sloppiness: output #8 will not sit beside output #1 as a set.",
-      "items": {
-        "type": "object",
-        "required": ["id", "purpose", "anatomy"],
-        "properties": {
-          "id": { "type": "string" },
-          "purpose": { "type": "string", "description": "Its job in a sequence: opener, proof, quote, closer." },
-          "anatomy": { "type": "string", "description": "Slot-by-slot structure in percentages." },
-          "content_shape": { "type": "string", "description": "The content budget: max words per slot." }
-        }
-      }
-    },
-    "motion": {
-      "type": "object",
-      "description": "Omit entirely for static work. Restraint is a signature too, which is why never_moves is required when this key exists.",
-      "required": ["never_moves"],
-      "properties": {
-        "easing": { "type": "string" },
-        "durations": { "type": "object", "additionalProperties": { "type": "string" } },
-        "entrances": { "type": "string" },
-        "never_moves": { "type": "array", "items": { "type": "string" } },
-        "reduced_motion": { "type": "string", "description": "The prefers-reduced-motion fallback. Not optional in shipped work." }
-      }
-    },
-    "voice": {
-      "type": "object",
-      "description": "Copy is part of the design. Right layout, wrong sentence, broken look.",
-      "properties": {
-        "register": { "type": "string" },
-        "sentence_length": { "type": "string" },
-        "headline_shape": { "type": "string" },
-        "banned_words": { "type": "array", "items": { "type": "string" } }
-      }
-    },
-    "bans": {
-      "type": "array",
-      "minItems": 5,
-      "items": { "type": "string" },
-      "description": "Written as absolutes, starting with 'Never'. A positive instruction has to out-vote the training average; a ban deletes the landslide from the options. One ban can rule out an entire space in one line. Bans should outnumber positive style rules."
-    },
-    "tests": {
-      "type": "array",
-      "minItems": 8,
-      "description": "8 to 12. Binary and measurable. If two people could disagree about the answer, it is not a test. A spec with no failing test is a mood board with better formatting.",
-      "items": {
-        "type": "object",
-        "required": ["id", "check", "fail_looks_like", "auto"],
-        "properties": {
-          "id": { "type": "string" },
-          "check": { "type": "string", "description": "Stated so the answer is yes or no, or a number against a bound." },
-          "fail_looks_like": { "type": "string" },
-          "auto": { "type": "boolean", "description": "True if a script could decide it. Those go into tools/check.py." }
-        }
-      }
-    },
-    "reconstruction": {
-      "type": "object",
-      "description": "Step 5. A spec that has never been used to rebuild its own source has never been tested at all.",
-      "required": ["attempted", "gaps_found", "passes"],
-      "properties": {
-        "attempted": { "type": "boolean" },
-        "passes": { "type": "integer", "description": "Expect 2 or 3." },
-        "gaps_found": { "type": "array", "items": { "type": "string" }, "description": "Every difference found between rebuild and original, and the field added to close it. Expect these to be things you were sure were obvious." },
-        "stranger_test": { "type": "string", "description": "Whether a stranger could still pick the copy from the original, and how you checked." }
-      }
-    }
-  }
-}

package/skills/design-dna/evals/README.md

@@ -1,64 +0,0 @@
-# Eval harness — `design-dna` skill
-
-Agent-run eval, not a shell script. The cases in `cases.yaml` are fed to a Claude Code agent and
-graded by inspection. Unlike most skills here, part of this one **is** mechanically checkable —
-see §C.
-
-## What's under test
-
-1. **Triggering** — does it fire when a winner exists and its look must survive, and stay quiet when
-   there is no winner yet (`design`), when the winner is still being found (`design-loop`), or when
-   the user is exploring variants (`prototype`)?
-2. **Capability** — does it hold the rules that are the method, especially the two it is cheapest to
-   drop: Step 5, and the separation of record from payload?
-3. **Its own gate** — does the emitted skill's checker actually check?
-
-## A. Triggering accuracy
-
-The boundary that costs the most is `design-loop` ↔ `design-dna`, because both involve a reference
-image and both are about a look. The split is whether a winner exists: finding it is upstream,
-keeping it is here.
-
-## B. Capability
-
-Three scenarios, chosen because each targets a rule that a plausible-looking run will quietly skip:
-
-1. **CAPTURE.** The long list is deliberate — every line is a rule the method names as load-bearing.
-   The two that a good-looking run drops first are **Step 5** (rebuild the reference from the record
-   alone and diff) and **coverage percentages**. Step 5 is the only honest completeness test that
-   exists: a spec that has never been used to rebuild its own source has never been tested.
-2. **REUSE.** The failure is re-deriving a style that already has a record, which produces a second,
-   slightly different identity under the same name.
-3. **Still generic.** The correct answer is a ban, not a rule. This case exists because the instinct
-   is always to add a positive instruction, and a positive instruction is one weak vote against the
-   average of everything the model has seen.
-
-## C. The mechanical part — check both directions
-
-Most of this eval is judgement. The emitted checker is not, and it must be verified the way the
-constitution's P2 demands: prove it can fail **and** prove it can pass.
-
-    python3 scripts/emit.py <slug> --validate     # the capture is finished, or it is not
-    python3 <slug>/scripts/check.py <output>      # the output obeys the style, or it does not
-
-Both directions, on a real captured style:
-
-- **Can pass** — run the checker over the style's own worked example. Zero failures, with the
-  manual tests reported as manual rather than silently passing.
-- **Can fail** — run it over an output that breaks a known ban. It must exit non-zero and name
-  which test failed and what the failure looks like.
-- **Cannot lie either way** — run it over an artefact a probe cannot read (a PNG for a probe that
-  reads declarations). Every such probe must report SKIP with its reason. This is the regression
-  that was found and fixed at port time: the declaration-reading probes used to run against empty
-  text, which produced spurious failures on one probe kind and, worse, **vacuous passes** on
-  another — five greens on an image-only run without having checked anything.
-- **Fails readably when unfinished** — a scaffolded style with no record must exit non-zero with a
-  message naming the fix, not a traceback.
-
-## Known limits of this eval
-
-- The CAPTURE scenario can check that the agent *runs* Step 5 and reports gaps; it cannot check that
-  the reconstruction was honest. A rebuild done with the reference still on screen looks identical to
-  one done from the record alone.
-- The 2KB payload cap and the two-ends attention ordering are asserted by the method, not measured
-  here. They are plausible and cheap to honour; they are not proven by this eval.

package/skills/design-dna/evals/cases.yaml

@@ -1,68 +0,0 @@
-skill: design-dna
-
-should_trigger:
-  - prompt: "This landing page came out beautiful. Turn it into a skill so I can reuse the look."
-    why: "The canonical CAPTURE case: a winner exists and the ask is to make its identity permanent and reusable."
-  - prompt: "Make five more carousels like this one, same look."
-    why: "'More like this' with an existing reference is REUSE if a record exists and CAPTURE if it does not; either way this skill owns the decision."
-  - prompt: "Keep the same style as the deck you built me last week."
-    why: "Style continuity across pieces is the problem this skill exists for; without a written record the model reproduces the average instead."
-  - prompt: "Everything I generate now looks like generic AI slop. It didn't at the start."
-    why: "Named verbatim in the description as a trigger. Drift is the symptom of unwritten rules, and the fix is a record with bans, not another prompt."
-  - prompt: "Lock this look in so it stops changing on me."
-    why: "Explicit trigger phrase; the ask is permanence, which is the record plus tests that can fail."
-  - prompt: "Every carousel I post looks slightly different from the last one. I want them to read as a set."
-    why: "Non-obvious: no keyword overlap with 'skill' or 'style'. The stated symptom is missing named archetypes — inconsistent structure under consistent styling."
-  - prompt: "Esto ha salido bien. Quiero que las siguientes cinco se vean igual."
-    why: "Non-English, no keyword overlap; the semantic shape is a winner plus a demand for reproducible identity."
-
-should_not_trigger:
-  - prompt: "Design a landing page for my new analytics SaaS — hero, pricing, the works."
-    route_to: "design"
-    why: "No winner and no reference to measure. Capturing an average of candidates produces the average, which is the thing the method exists to escape."
-  - prompt: "Make our homepage as good as linear.app and don't stop until it wins."
-    route_to: "design-loop"
-    why: "Finding the winner is the upstream half of the pipe. There is nothing to codify until a piece has cleared the bar."
-  - prompt: "Write the hero headline and value prop for this page."
-    route_to: "marketing"
-    why: "The words. `voice` in the record constrains copy; it does not write it."
-  - prompt: "Build these pages with the App Router and server actions."
-    route_to: "nextjs"
-    why: "The build. The identity stays in the record; the framework work is the sibling's."
-  - prompt: "Give me three different directions for this poster and I'll choose."
-    route_to: "variant-explorer"
-    why: "Exploration among variants. One skill per style: merging two candidate identities into one spec produces the average of two good designs, which is a bad design."
-
-capability:
-  - scenario: "User attaches a screenshot of a landing page they love and says 'codify this'. (CAPTURE mode, nothing captured for this style yet.)"
-    must_include:
-      - "Runs the seven steps in order and shows work at each; does not jump straight to writing a style file."
-      - "Step 1 produces literal measurements with NO judgements — sampled colours, coverage estimates, type counts, largest-to-smallest ratio, margins as percentages, and what is ABSENT."
-      - "Records ratios and relationships rather than absolute values ('headline is 8x body, never under 6x', not 'headline is 96px')."
-      - "Records colour COVERAGE percentages, not just the colours — states that the same three colours at 60/30/10 and at 90/8/2 are two unrelated designs."
-      - "Uses descriptive colour names ('dusty plum'), never systematic ones ('accent-500'), because image and video models cannot read token names."
-      - "Gives every type family a fallback."
-      - "Identifies the ONE deliberate break in the system (the weird move) and gives it its own field."
-      - "Writes 8-12 tests that are binary and measurable, marks which are automatable, and rejects 'feels premium' / 'looks clean' as non-tests."
-      - "Produces MORE bans than positive style rules, written as absolutes."
-      - "Does NOT skip Step 5: rebuilds the reference from the record alone, diffs against the original, folds every difference back in as a forgotten field, and records the gaps found."
-      - "Keeps the record and the payload as two separate artefacts: the payload is capped at 2KB and the record never enters a prompt."
-      - "Puts any real logo, wordmark, licensed photograph or proprietary typeface into not_copied with a named substitute — reproduces the system, never the marks."
-      - "Declares every inferred-rather-than-measured value and every rule under 70% confidence."
-      - "Validates the emitted skill with the emit script rather than asserting it works."
-  - scenario: "A style called `cold-press` already has a full record. User says: 'build me an about page in cold-press'."
-    must_include:
-      - "Recognises this as REUSE and does NOT re-derive the style — a style with a record is never re-measured."
-      - "Loads that style's payload ONLY and attaches the kept reference; does not read the full record into context."
-      - "Picks a named archetype BEFORE writing anything, so the new piece sits beside the earlier ones as a set."
-      - "Hands the build to the nextjs sibling and the words to marketing, keeping the identity here."
-      - "Runs the style's own auto-check before returning, names each test and its result, and repairs then re-runs on any failure."
-      - "Does not return output with a failing test plus a note explaining it away."
-  - scenario: "A style is captured and its record is complete, but the pages built from it still come out looking generic. User asks what to add."
-    must_include:
-      - "Adds a BAN, not another positive rule — states that a positive instruction is one weak vote against the average while a prohibition deletes a whole space."
-      - "Checks whether the signature moves are missing or too numerous and cuts to 3-9, each rewritten as a ratio."
-      - "Checks whether the payload is over its cap, since an over-cap payload is obeyed unevenly and differently each run."
-      - "Checks whether the strongest rule is buried in the middle of the payload rather than at either end."
-      - "Checks whether a weird move exists at all, since a spec with no break in the system reads as forgettable rather than wrong."
-      - "Does not respond by merging in a second style or by re-capturing from scratch."