brand-identity-skillCAUTION

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: brand-identity
description: Brand identity sets from one brief - logo, typography and colour palette built together, measured (contrast, colour blindness, glyph coverage, font licences, small-size logo renders) and compared side by side as identity cards, shown on the user's own site, then a brand guidelines kit (PDF, logo files, tokens) for the chosen set. Use for a new brand or any part of one (logo, fonts, colours, or a mix), for refreshing an identity, or for critiquing one, even if the user only says "branding".
license: MIT
---

# Brand Identity

You are the designer. Scripts measure what is broken; you decide what is good. Every claim you make about a
colour, typeface or mark is either measured by a script or labelled as judgement.

## 1. Intake: one message, only what blocks you

Ask in one message (skip anything the user or their site already answered):
- For **logo, type, palette** each: `new` · `refresh` (they have one, evolve it) · `keep` (use as-is) · `none`.
- How many sets: **3** (default), 2, 1 or 4.
- Site URL or files; languages the brand writes in; does the product show numbers.
- Only if still unknown: sector, audience, three attributes, 3–5 competitors.

With a site, run `site extract` first and ask only about what it could not tell you. Details:
`references/discovery-brief.md`.

## 2. Flow and turn plan (~16–20 turns to the checkpoint)

Scripts: `python3 <skill>/scripts/brand.py <command>` (`--help` on each). Put independent commands in one turn.
With uv the scripts fetch their own packages; if `check` lists missing ones, give the user its install line.

1. `check` + `init NAME --sets 3 --langs en,tr [--numbers] [--site URL] [--state logo=keep ...] --sector ... --audience ...
   --attributes a,b,c --competitors x,y --tagline ... --sentence en='...' --sentence tr='...' --nav A,B,C --cta ...
   [--source logo=file.svg --source palette=#hex,#hex]` (one turn). Sentence (one per brand language), nav and CTA are
   real brand copy for specimens and mocks, in the brand's voice. `--source` gives the asset of a keep/refresh part.
2. Read `references/cohesion.md`, `type.md`, `logo.md`, `ai-defaults.md`, and `color.md` when a palette is
   wanted (one turn, ≤ 24 KB). Other cards only when their topic comes up.
3. `fonts search --class ... --exclude-defaults` and `site competitors WORK url...` when there are competitors.
4. Write **all sets in one `sets.json`** (format: `templates/sets.example.json`): per set the identity fields,
   a partial palette (`brand[]` seeds, name, direction) and the symbol SVG inline.
5. `logos WORK` → look at the contact sheet → fix one defect per mark (name it in `logo.defect`) → `logos WORK`
   → look. Two rounds; a third only to clear a gate or a cliché reading. Letter positions for details are in
   `sets/X/logo/build/wordmark-glyphs.json`.
6. `build WORK` → look at `review.png` (not `board.png`, the user's large copy) → fix gates (re-run only
   the failing set with `--sets X`).
7. Present the checkpoint (§5) and **stop**. Never pick for the user.
8. After the pick: add 1–2 misuses of this mark (`references/kit.md`), then `kit WORK/sets/X` (`--full` for the
   long kit). Say which files are where.

`mix WORK A:logo A:type B:palette --as D` when the user asks for parts of different sets. `critique` for an
existing identity (`references/critique.md`).

## 3. Making the sets

- **One mechanism per set**: a single visual idea that produces the mark, the type choice and the palette
  together. Write it as one sentence before choosing anything. One **expression move**: the only loud thing.
- Pick 2–4 decision axes for this brief and place each set on them (−2..+2). Sets should stand apart on at least
  two axes; inside a set the three parts sit at the same position.
- **Differentiate against the user's real competitors**, not a generic list. Say in `differs_by` where
  the set stands apart.
- **Reflex test**: for each font, colour and mark idea ask "would I propose this for another brand in this
  sector?" If yes, justify it from the brief in `defaults_used[]` (id = the flag's id or its logo tag) or change it. At least one set uses nothing
  from `ai-defaults.md`. "Feels warm" or "feels premium" is not a reason.
- `keep` parts are constraints: build the new parts to fit them. If a kept part clashes, say so as a suggestion;
  never change it.
- Mark `recommended: true` on the set you would defend, and know why.
- Write all free text (names, mechanism, concept, rationale, labels) in the document language (`init --doc-lang`).

## 4. Component craft

**Logo** (`references/logo.md`): a mark identifies, it does not explain. Write 4–6 one-sentence ideas per set,
build one. Draw the symbol on a 100-unit grid from filled primitives; use `data-op="subtract|union|intersect"`
for cut-outs and `data-color="<palette role>"` for colour; never `<text>`. Wordmarks come from the set's display
face automatically; set `case`, `tracking` (1/1000 em) and `tags` (structure vocabulary in `ai-defaults.md`).
Look at 16 and 32 px every round. Add `symbol_small_svg` when the mark clogs at small size.
A **wordmark-only** set still needs (1) one ownable detail drawn on the outlined type with `wordmark_detail_svg`
(wordmark coordinates: cap top y=0, baseline y=100, x in `data-advance` units; each shape is a `data-op` step,
`data-glyph="n"` limits it to the n-th character) and (2) a designed small mark (`symbol_small_svg` or
`logo.monogram`) for 16–48 px. Plain type with a first-letter favicon reads as stock. A detail shape with its own
`data-color` stays a separate colour; `data-glyphs` elements take `data-wght`/`data-opsz` for a sturdier small cut.
Role names resolve to UI tones; brand ids (`brand-1`) keep the exact brand hex; a main part under 3:1 on a ground is
drawn in that ground's ink. `app_icon_svg` (100-unit
grid) lets the tile be part of the icon. When the idea is a dark or coloured surface, set `card.ground` (`dark` or
a role / brand id) so the card shows it.

*

CAUTIONgrade D · trust 65/100 Install with care. The audit found things worth knowing before you trust its output.

What the source does

Filesystem

declared (9 observation(s))

Network

declared (6 observation(s))

Shell

declared (4 observation(s))

Dependencies

not all pinned

Secrets in source

none-found

Findings (25)

languages.json.gz

return importlib.import_module(name)

return __import__(name)

with local_urlopen(f"http://127.0.0.1:{port}/json/list", min(timeout, 10)) as r:

f"{b.get('name', b['hex'])} {b['hex']} is ΔE2000 {d:.1f} from {host}'s {role} {hx}",

suggested_fix="move lightness or hue until ΔE2000 >= 10, or say why sharing it is "

_JS_WS = "\t\n\v\f\r              

   "

lines = text.lstrip("").replace("\r\n", "\n").split("\n")

const L = eval(LIB);

const found = LOGO ? eval(LOGO)() : null;

const L = eval(LIB); const map = L.makeMapper(oldRoles, newRoles, { scales });

const L = eval(LIB); const T = eval(GUARD)(L);

const L = eval(LIB); const T = eval(GUARD)(L);

self.assertFalse(a(page.rsplit("/", 1)[0] + "/../.ssh/id", page)["ok"])

ok[:3] + [{"svg": "/etc/x.svg", "label": "x"}], ok[:3] + [{"svg": "../x.svg", "label": "x"}],

cp -R "$here/../../skills/brand-identity/templates/demo/." "brand-identity/ferrow/"

cp "$here/../../skills/brand-identity/templates/sets.example.json" "brand-identity/ferrow/sets.json"

cp -R "$here/../../skills/brand-identity/templates/demo/." "brand-identity/ferrow/"

cp "$here/../../skills/brand-identity/templates/sets.example.json" "brand-identity/ferrow/sets.json"

env.update(HTTP_PROXY="http://127.0.0.1:9", http_proxy="http://127.0.0.1:9")

self.base = f"http://127.0.0.1:{self.srv.server_address[1]}"

@font-face{font-family:'BI Instrument Sans';font-style:normal;font-display:block;font-weight:400 700;font-stretch:75% 100%;src:url(data:font/woff2;base64,d09GMgABAAAAAN/0ABQAAAABzggAAN+AAAEAAAAAAAAAAA

@font-face{font-family:'BI Instrument Sans';font-style:normal;font-display:block;font-weight:400 700;font-stretch:75% 100%;src:url(data:font/woff2;base64,d09GMgABAAAAAEmoABQAAAAAmIwAAEk4AAEAAAAAAAAAAA

@font-face{font-family:'BI Instrument Sans';font-style:normal;font-display:block;font-weight:400 700;font-stretch:75% 100%;src:url(data:font/woff2;base64,d09GMgABAAAAAN/0ABQAAAABzggAAN+AAAEAAAAAAAAAAA

@font-face{font-family:'BI Instrument Sans';font-style:normal;font-display:block;font-weight:400 700;font-stretch:75% 100%;src:url(data:font/woff2;base64,d09GMgABAAAAAEmoABQAAAAAmIwAAEk4AAEAAAAAAAAAAA

Gates applied: no_behavioural_pass.