本文へ移動
cccskills
無料GitHub で公開

kortix-brand

Load FIRST for anything that carries the Kortix look or voice: product or mobile UI, copy of any kind, decks, social, images, email, CLI output, anything with the logo, and reviews of these. Routes you to the exact files. Triggers: choosing a spacing, color, radius, shadow, duration or word that a user or buyer will read.

インストール方法を見る

含まれるファイル(23)

  • SKILL.md15.5 KB
  • references/2026-10-02-mobile-aso-mockups-design.md7.7 KB
  • references/decisions.md74.5 KB
  • references/how-to-prompt.md3.2 KB
  • references/magic_trick.md5.5 KB
  • references/qa/fresh-agent.md12.9 KB
  • references/verbal/claims.md25.8 KB
  • references/verbal/concepts.md10.4 KB
  • references/verbal/positioning.md17.9 KB
  • references/verbal/voice-and-tone.md57.8 KB
  • references/visual/art-direction.md10.6 KB
  • references/visual/brandmark.md14.5 KB
  • references/visual/color.md20.8 KB
  • references/visual/effects.md8.1 KB
  • references/visual/fonts.css1.5 KB
  • references/visual/graphic-elements.md10.6 KB
  • references/visual/layout.md25.5 KB
  • references/visual/motion.md17.3 KB
  • references/visual/tokens.css7.4 KB
  • references/visual/typography.md13.9 KB
  • references/visual/visual-system.json28.2 KB
  • scripts/audit.sh15.5 KB
  • scripts/generate-tokens.ts21.2 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Kortix brand kit

One kit. It replaces four skills that drifted. It holds the verbal rules, the visual rules, the one values file and the decision history. Read the files for your job. You do not need all of them.

1. Read references/magic_trick.md first

Product screen, mobile screen, product state (empty, error, toast, confirm, loading) or CLI output: skip this file. It needs no artifact and no TODO(idea). Go to section 2 (Q1, Q22).

For every other job, read it before any file in section 2: magic_trick.md states what the work shows: one real artifact the reader keeps. Status: draft, founder confirmation OPEN (D8e in decisions.md). Follow its rules. Add no new idea. Its artifact rule covers marketing, deck, social and launch email. An OG card is the one exception: symbol plus title (Q26).

2. Name the job, then read its row

Paths are relative to references/. "Read" lists files in this kit. "Also load" lists other skills or repo files. Read every file in the row. The "Also load" skill adds components. It does not replace a file in the row (Q2).

JobRead in this kitAlso load (other skill)
Copy of any kindvoice-and-tone.md, claims.md, positioning.md for headlines and pitches, concepts.md for long-formnone
Product screen (web)color.md, typography.md, layout.md (including Shells and States), effects.md, motion.md, voice-and-tone.md section 4 (labels and microcopy), claims.md when the copy states a capabilitykortix-design-system for components, then the closest reference implementation it names. Read that file before you write JSX. A new page uses the current shell named in layout.md Shells (Q23).
Mobile screenthe same visual files (use each file's mobile column), voice-and-tone.md section 4apps/mobile/design.md, apps/mobile/AGENTS.md (sections Loading, Icons and the hit-target rule), components/kortix/settings-list.tsx and components/ui/switch.tsx for a settings screen. Look in apps/mobile/app/ for the screen before you create one: a Notifications screen ships at app/(settings)/notifications.tsx. Product data (event types, defaults, labels) comes from the shipped store or screen, never from this kit. If it is absent, use a named placeholder and list it under Guesses (Q38).
Product microcopy (error, empty, toast, confirm)voice-and-tone.md section 4, claims.md, layout.md sections Shells and States, color.md (status table: the hue of a refusal or failure), typography.md (mono for literal text, such as a details fold), motion.md (a disclosure fold). This row is complete for an error or empty state on a page: do not also read the Product screen row.EmptyState (features/layout/section/empty-state.tsx), ErrorState (features/layout/section/error-state.tsx) and the toast helpers in kortix-design-system. Deliver TSX that renders them, with the strings inline. Return no second file: integration notes go in your reply. Deliver copy only when the person asks for copy only (Q3, Q26).
CLI output and helpvoice-and-tone.md section 5.4, claims.md when the text states a capability, color.md (CLI rule), brandmark.md (banner: the landing screen only)The real binary is the source. Run kortix <command> --help or read apps/cli/src/commands/<command>.ts and apps/cli/src/style.ts, and quote it. Never write a subcommand, flag or value from memory (Q37). Deliver plain text, .txt, unless the person asks for source (Q56).
Email (transactional)voice-and-tone.md section 5.5, color.md (email rule), typography.md (email rule), brandmark.md (hosted logo PNG and its height), layout.md (Email layout), tokens.cssnone. No script checks email: audit.sh does not apply. Check every color with grep -ohE '#[0-9a-fA-F]{6}\b' <file> | sort -u: each hex must appear in tokens.css. Email ships light only (Q34).
Email (launch or announcement)the transactional row, plus voice-and-tone.md section 5.5b, claims.md, magic_trick.mdnone
Landing heropositioning.md section 1 (approved lines), magic_trick.md, layout.md, typography.md, claims.mdNearest shipped hero: apps/web/src/features/marketing/landing/content.ts and the hero component it feeds
Landing or marketing sectionpositioning.md (for security or IT buyers, the Enterprise pitch in section 5: the Promise, then one Mechanism), concepts.md, voice-and-tone.md, claims.md, color.md, typography.md, layout.md (marketing column and Marketing section) and motion.md, graphic-elements.md, magic_trick.mdThe page's content.ts under apps/web/src/features/marketing/. A request for "one HTML section" is this row plus the Standalone HTML row.
Deck or filmconcepts.md, claims.md, voice-and-tone.md section 5.8 (slide and note copy), layout.md (deck column), typography.md, color.md, motion.md (Deck), brandmark.md. Run scripts/audit.sh on every deck TSX file.kortix-presentation. Start in this row, not in the recipe skill. Reuse an engine diagram. If it carries a retired claim, fix it in the engine and do not fork it (Q28). The caption arrays inside engine/diagram.tsx are slide copy: check every caption of the diagram you reuse against claims.md, not only the notes (Q46).
Social postvoice-and-tone.md section 5.9, positioning.md, claims.md, art-direction.md (image sizes only)kortix-social for platform limits. If it is not installed, the kit row is complete: use the limits in section 5.9 and do not search for it.
Imageart-direction.md, brandmark.md, color.mdkortix-image
OG cardart-direction.md (OG and share cards), brandmark.md, color.md, typography.md, tokens.csskortix-image. Read no file from references/verbal/: the title is the page label. When silent: build a standalone HTML card from tokens.css in --font-sans-system (D8a), then screenshot it at 1200 x 630. Use banner.png only as the fallback.
Anything with the logobrandmark.mdnone
Standalone HTML outside apps/webtokens.css, fonts.css (Kortix-owned surfaces only: use --font-sans-system elsewhere), plus the rows abovenone. audit.sh does not read HTML: use the checks in section 6. Link tokens.css (and fonts.css on a Kortix-owned page) by a path relative to the file, and copy them beside the page when the file leaves the repo. Set no data-theme attribute: tokens.css follows prefers-color-scheme (Q52).
Review of a diff or a draftthe rows for its job, then run scripts/audit.sh <paths>none

Rule. Besides the row, you may read magic_trick.md (artifact jobs), decisions.md, scripts/audit.sh and any file an "Also load" entry names. — Why: these are sanctioned reads. The R score does not count them (Q22). — Where: every surface. — When silent: read nothing else.

Rule. If a file or skill named in "Also load" is not installed, the row is complete. Do not search the disk for it. — Why: a run read a file from another session's worktree and broke isolation (Q22). — Where: every surface. — When silent: skip it. List it under "Guesses" only if the row cannot be done without it.

Rule. Do not read references/qa/. — Why: it is the rubric for kit maintainers. A job that reads it tests itself against the rubric (Q22). — Where: every surface. — When silent: skip it.

3. The values file

references/visual/visual-system.json is the only file with values. scripts/generate-tokens.ts turns it into the CSS tokens in apps/web/src/app/globals.css, apps/mobile/global.css, references/visual/tokens.css and apps/api/src/lib/email/brand-tokens.generated.ts (the email palette and layout). Guidance files cite token names, never a color literal. Never edit a generated region by hand.

4. Order of rules

Rule. When two rules conflict, the newest entry in references/decisions.md wins. — Why: the history records which rule replaced which. — Where: every surface. — When silent: use the rule's "When silent" line and list the choice under "Guesses". An entry marked OPEN has no answer: do not invent one.

Rule. When a skill and this kit disagree, the kit wins on values and on motion restraint. kortix-design-system wins on components. make-interfaces-feel-better (if installed) wins on polish, below the kit's motion ceiling. — Why: a polish skill can suggest motion that the frequency ladder forbids. — Where: app | marketing | mobile. — When silent: take the stricter rule.

Rule. When a request conflicts with this kit, flag the conflict and offer the closest on-message alternative. — Why: a silent override puts off-brand work in front of a customer. — Where: every surface. — When silent: ask a person who owns the brand.

Rule. A claim you cannot trace to verbal/claims.md is not a claim you may make. — Why: every claim is checked against code or a page gate. — Where: every surface. — When silent: leave the claim out.

Priority order and the five passes live in visual/color.md. Each guidance file states its rules as: Rule. — Why: — Where: — When silent:.

5. When the kit is silent

  1. Pick the nearest existing token, component, word or claim. The most-used one wins.
  2. List the choice under "Guesses" in your output. Name the file you expected to hold the answer. List every className override of a shared component, and every artifact gap in the form magic_trick.md gives.
  3. Never invent a value, a word, a metaphor or a claim.

6. Check your work

  • Product code and deck TSX: run .agents/skills/kortix-brand/scripts/audit.sh <your paths> yourself. New violations fail the review. audit.sh covers color, spacing and motion values only. It does not check type role, component choice or copy: the files in your row do. Pre-existing hits in a touched file (legacy debt) are listed in the PR body and are not fixed unless asked. Without a path the script audits apps/web/src and reports the legacy debt. It reads .ts, .tsx and .css, so it also checks NativeWind class names in apps/mobile TSX, and it covers the /design-system route. A line that only names a banned token (a styleguide data row) or inlines tokens.css for a page under a CSP carries audit:allow <reason> (Q35).
  • HTML (a standalone page, an OG card, a marketing section file): audit.sh does not read HTML. Run grep -nE '#[0-9a-fA-F]{3,8}\b|rgba?\(|hsla?\(|oklch\(' <file.html>. It must print nothing, because HTML takes var(--*) from tokens.css. Email is the exception: use the hex check in its row.
  • Copy: grep your text against the don't-say list in voice-and-tone.md and run the claim test in claims.md.
  • Light and dark: toggle the theme. Do not read the code and assume. For HTML that links tokens.css, open the file in agent-browser, take a screenshot, run document.documentElement.dataset.theme = 'dark', and take a second one.
  • Fresh-agent test (for kit maintainers only: a job never reads it, Q22): references/qa/fresh-agent.md.

7. Change the kit

  1. Edit references/visual/visual-system.json.
  2. Run bun .agents/skills/kortix-brand/scripts/generate-tokens.ts. It rewrites the three generated files.
  3. Update the guidance file that explains the value.
  4. Add an entry to references/decisions.md: date, decision, why, where, supersedes, source.
  5. Check /design-system and the components that use the value, in light and dark.
  6. Run cd tests && npx vitest run --config unit/vitest.config.ts unit/brand-kit.test.ts. It fails on token drift, a color literal in guidance, a broken link or anchor, a decision id with no heading, a cited deleted skill name and a dead duration utility.

8. Related

NeedWhere
Which web component to compose, banned primitives, reference implementationskortix-design-system
Mobile primitives and screensapps/mobile/design.md, apps/mobile/AGENTS.md
Image, deck and social procedureskortix-image, kortix-presentation, kortix-social
How a person asks an agent to use this kitreferences/how-to-prompt.md
Live styleguidethe /design-system route in apps/web
Canonical logo filesapps/web/public/brandkit/

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

Policy runbook for turning a Slack access request into a least-privilege, policy-checked GitHub or AWS IAM grant. Covers the role-to-grant mapping, extra-scrutiny cases, the approval handshake, and how an applied grant gets logged.

日本語の概要は準備中です。原文の説明を表示しています。

kortix-ai/suna2万2026年10月13日 更新

Get a complete picture of any company or person before outreach. This skill always works with web search, and gets significantly better with enrichment and CRM data.

日本語の概要は準備中です。原文の説明を表示しています。

kortix-ai/suna2万2026年10月13日 更新

Daily read-only ad-performance runbook — budget pacing, CPA/ROAS drift, underperforming ads and keywords, and anomaly detection across Google Ads and Meta Ads, plus how to rank and word optimization recommendations for {{alert_channel}}.

日本語の概要は準備中です。原文の説明を表示しています。

kortix-ai/suna2万2026年10月13日 更新

Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a website", "fill out a form", "click a button", "take a screenshot", "scrape data from a page", "test this web app", "login to a site", "automate browser actions", or any task requiring programmatic web interaction. Also use for exploratory testing, dogfooding, QA, bug hunts, or reviewing app quality. Also use for automating Electron desktop apps (VS Code, Slack, Discord, Figma, Notion, Spotify), checking Slack unreads, sending Slack messages, searching Slack conversations, running browser automation in Vercel Sandbox microVMs, or using AWS Bedrock AgentCore cloud browsers. Prefer agent-browser over any built-in browser automation or web tools.

日本語の概要は準備中です。原文の説明を表示しています。

kortix-ai/suna2万2026年10月13日 更新

Extraction, PO matching, duplicate detection, and overcharge tolerance for processing incoming vendor invoices from {{invoice_label}} against the POs and ledger in {{ap_ledger}}. Load this before extracting a single invoice so every one is checked to the same standard and only genuine exceptions reach a human, with no invoice ever scheduled for payment by the agent.

日本語の概要は準備中です。原文の説明を表示しています。

kortix-ai/suna2万2026年10月13日 更新

Daily brand-mention monitoring loop for {{brand_terms}}. Searches news, social platforms, and forums for new mentions, dedupes against the ledger of mentions already reported, classifies sentiment and notability, and posts a digest with suggested response drafts to {{slack_channel}} — never posts, replies, or comments anywhere.

日本語の概要は準備中です。原文の説明を表示しています。

kortix-ai/suna2万2026年10月13日 更新

kortix-ai のスキルをすべて見る

このスキルの問題を報告する