This repository is a Claude Code design skill that ships supply-chain gates as load-bearing components — SVG sanitizer, PNG / JPG chunk scanner, codex-image-import gate, codename-pattern catalog, and a GitHub Actions workflow that runs all of them on every push and pull request.

If you find a vulnerability in any of these components, or in the documentation that describes how they should be used, please report it privately rather than opening a public issue.

Preferred channel: GitHub's private security advisories for this repository (Security tab → Report a vulnerability). The advisory form will start a private thread between the reporter and the maintainer.

If GitHub advisories are not available to you, send an email to the maintainer through the contact information on https://github.com/0xmhha. Use a subject line that begins with [claude-design-skill SECURITY].

A useful report names:

1. The component (sanitizer / scanner / import gate / codename catalog / workflow / docs).
2. The exact regression: which input bypasses which gate, or which policy in the docs is wrong / out of date.
3. A minimal reproduction (preferably as a unit-test case in the format of scripts/test_*.py or scripts/test_*.js).
4. The repository SHA you tested against.

In scope:

• The sanitizer and scanner scripts under scripts/ (svg-sanitize.py, scan_assets.py, codex-image-import.py, init-brand.py, easing.js, and their regression suites).
• The GitHub Actions workflow at .github/workflows/sanitizers.yml.
• The codename pattern catalog and the host allowlist in references/security-config.md.
• The strip-then-scan import flow described in references/codex-design-workflow.md and references/svg-sanitize.md.
• The examples/dot-claude-settings.json template.

Out of scope:

• Third-party services referenced via the host allowlist (Lucide, Phosphor, MDN, Apple HIG, Wikimedia, Met Museum, Artic, Unsplash, Pexels, Google Fonts, unpkg / jsdelivr): report directly to those vendors.
• Vulnerabilities in huashu-design or alchaincyf/huashu-design: this repository does not derive from those projects (see LICENSE and NOTICE for the clean-room rewrite statement); report there separately.
• Fork-specific brand customization (team-brand-spec.json real values, internal codenames, internal asset hosts): owned by the fork operator; report through that fork's channel.

The maintainer is solo and contributes to this project as internal R&D. No SLA is offered. As guidance:

• The maintainer aims to acknowledge a report within seven days.
• Fix timing depends on severity. The strip-then-scan gate, the codename pattern catalog, and the GitHub Actions workflow have hard-fail regression suites that any fix must pass before merging.
• Coordinated public disclosure is welcome; the reporter and the maintainer agree the timing in the private advisory thread before it goes public.

• A reduced-strength regex in DEFAULT_CODENAME_PATTERNS. The catalog is conservative-pairing by design (see references/security-config.md §1.5); a "bypass" using a bare keyword without the paired noun is the documented trade-off, not a defect.
• The prefers-reduced-motion fallback in assets/animations.jsx. The Stage / Sprite engine respects the setting; if a deliverable that uses the engine still violates it, that is the deliverable's bug, not the engine's.
• Anything in the upstream alchaincyf/huashu-design skill. This repository does not derive from it.