Security

BrandKit is an unauthenticated, single-tenant image-processing service designed for deployment within trusted environments. Every security decision follows from that model. When exposing the application across untrusted networks, perimeter security controls must be implemented at the reverse proxy layer.

Reporting a vulnerability

Do not open a public GitHub issue for security disclosures.

Email fabrizio.salmi@gmail.com with:

• Vulnerability description and impact assessment
• Reproduction steps and proof of concept
• Remediation proposal (if available)
• Contact details

Machine-readable contact metadata is published at /.well-known/security.txt per RFC 9116.

Response timeline

Stage	Target window
Acknowledgment	Within 48 hours
Initial assessment	Within 5 business days
Status updates	Every 7 days until resolved
Remediation (Critical)	Within 30 days
Public disclosure	Within 90 days, coordinated with reporter

Researchers are credited upon publication unless anonymity is requested.

Severity classification

Level	Examples
Critical	Remote code execution, arbitrary file writes
High	Data exposure, path traversal
Medium	Cross-site scripting (XSS), cross-site request forgery (CSRF) bypass
Low	Verbose error disclosures, localized resource consumption

Supported versions

Security patches are published exclusively against the latest release branch (currently v1.1.5).

---

Built-in security controls

Request handling

Control	Mechanism	Description
CSRF defense	Flask-WTF CSRFProtect	Enforced on all state-changing HTTP POST endpoints; token minted on GET /
Rate limiting	Flask-Limiter	200/day, 50/hour globally; 5/min on /upload; in-memory tracking
Security headers	Flask-Talisman	CSP, X-Content-Type-Options, X-Frame-Options, strict referrer policy
Payload ceiling	MAX_CONTENT_LENGTH	16 MB default, configurable via BRANDKIT_MAX_UPLOAD_MB
Session secret	BRANDKIT_SECRET_KEY	Deterministic secret storage; displays startup warning when omitted

Content Security Policy

default-src 'self'
img-src     'self' data: blob:
script-src  'self' 'unsafe-inline' 'unsafe-eval'
style-src   'self' 'unsafe-inline'

'unsafe-inline' and 'unsafe-eval' are required by Alpine.js for declarative directive execution.

No external origins are permitted: Tailwind CSS and Alpine.js assets are vendored under static/vendor/ and served same-origin, eliminating external script injection vectors.

File system integrity

• Extension validation: Restricted to png, jpg, jpeg, gif, webp (matched case-insensitively).
• Name sanitization: secure_filename() applied to all incoming uploads alongside a UUID4 prefix (<uuid4>_<sanitized>).
• Buffer validation: Uploads are immediately parsed through Pillow. Corrupt or unparseable buffers are deleted and rejected with HTTP 400.
• Mandatory EXIF stripping: Ingested images are re-encoded through a clean Pillow memory buffer, stripping camera metadata, serials, and geolocation coordinates.
• Scheduled purge: Internal worker thread sweeps static/uploads/ and cache directories according to configured retention periods.

Network isolation

Loading application pages initiates zero external network requests. The only external request the container can trigger is downloading neural weights on initial background removal invocations, which can be pre-cached to support air-gapped environments.

---

Operational hardening considerations

Asset visibility

Files in static/uploads/ are served without authentication. On multi-tenant or internet-exposed instances, enforce access control at the reverse proxy layer (see Deployment).

Verbose error feedback

Selected endpoints (/analyze) return raw error strings for debugging convenience. In production, proxy layers should intercept 5xx errors to prevent revealing filesystem paths.

Antivirus scanning

Buffers are validated for image decodability. Deploy an external antivirus scanner if processing untrusted public submissions.

---

Dependency management

Dependencies are pinned in requirements.txt and monitored via Dependabot and OSV audits. Image processing libraries (particularly Pillow) require consistent maintenance:

# Audit installed packages against known CVEs
pip install pip-audit && pip-audit

Rebuild container images after applying updates: docker compose up -d --build.

---

Core deployment rules

1. Authenticate site-wide: Enforce reverse-proxy authentication over both application endpoints and /static/.
2. Configure BRANDKIT_SECRET_KEY: Required for multi-worker Gunicorn stability.
3. Terminate TLS externally: Configure reverse proxies to manage certificates and enforce HSTS.
4. Enforce FLASK_ENV=production: Never enable debug mode in exposed environments.
5. Bind container to localhost: Map "127.0.0.1:8000:8000" to avoid exposing internal sockets to public network interfaces.

---

Threat model boundaries

In scope: Image parser exploit mitigation, CSRF protection, resource limit enforcement, EXIF metadata sanitization.

Out of scope: Native multi-tenant isolation, user identity management, audit logging. Security boundaries must be enforced upstream.