Login is now by email, not username. username still exists internally
(JWT sub claim, audit logs, Jira actor attribution, SSO provisioning all
still key off it) but is now always kept equal to email everywhere a user
is created or their email changes — never a separately-chosen value.
- User.email is now unique + NOT NULL (migration b067 backfills any
missing/blank email from username first, so existing rows — notably
the seeded admin, which historically had none — never violate it).
- /auth/login and the (unused but updated for consistency)
authenticate_user() now query by email.
- create_user (legacy, unreferenced but kept) and
create_user_without_password both derive username from email.
- update_user keeps username in sync when email changes, and rejects
duplicate emails.
- seed.py reads ADMIN_EMAIL (new env var, wired through install.sh and
docker-compose.prod.yml) for the initial admin; falls back to an
email-shaped ADMIN_USERNAME or a placeholder that's flagged for the
operator to fix.
- admin_config.py's import bundle now matches/creates users by email,
skipping (not crashing on) entries with no email.
- sso_service.py always sets username = email for SSO-provisioned users.
- LoginPage/auth.ts updated to email input/copy (wire field name stays
'username' — that's the OAuth2PasswordRequestForm spec, not the value).
Fresh installs failed at 'docker compose up' with 'required variable
PLATFORM_URL is missing a value' — the install script computes ORIGIN_URL
for CORS_ORIGINS already but never wrote PLATFORM_URL, which the backend
now requires with no default. Also backfills it into an existing .env
(from CORS_ORIGINS) when the user chooses to keep their current config,
so upgrading an older install in place doesn't hit the same error.
When the wizard reconfigures and generates a new DB_PASSWORD, the existing
Postgres volume retains the old password (Docker only initializes credentials
on a fresh empty volume). The backend then fails to connect because .env
has the new password but Postgres still uses the old one.
Fix: run 'docker compose down -v' before 'up --build' whenever the wizard
reconfigures (SKIP_CONFIG=false), so Postgres always initializes with the
current .env credentials. Also add a pre-confirmation warning when existing
volumes are detected.
Critical (1-3):
- Replace hardcoded admin credentials with secure auto-generation (seed.py)
- Enforce SECRET_KEY configuration, fail in production if missing (config.py)
- Add Zip Slip and Zip Bomb protection to all ZIP import services
High/Medium (4-9):
- Add 50MB file size limit and extension whitelist to evidence uploads
- Configure CORS origins via environment variable instead of hardcoded
- Migrate JWT storage from localStorage to HttpOnly cookies (frontend+backend)
- Add rate limiting (5/min) on login endpoint via slowapi
- Replace generic dict payloads with Pydantic schemas (mass assignment)
Medium (10-17):
- Check is_active on login to prevent disabled users from authenticating
- Sanitize exception messages in API responses (system, data_sources)
- Escape LIKE wildcards in all ilike search filters across 8 routers
- Run Docker container as non-root user (appuser)
- Make MINIO_SECURE configurable via environment variable
- Add password complexity policy (12+ chars, upper/lower/digit/special)
- Implement JWT token revocation via in-memory blacklist + reduce TTL to 15min
- Replace xml.etree with defusedxml to prevent Billion Laughs attacks
Low (18-20):
- Add security headers to Nginx (CSP, X-Frame-Options, HSTS-ready, etc.)
- Disable Swagger UI/ReDoc/OpenAPI in production
- Restrict /health endpoint to internal networks via Nginx ACL
Also: rewrite install.sh as interactive wizard for guided deployment,
fix test-from-template validation error (technique_id UUID vs MITRE ID)
- Auto-detect project root from script location so it works from any dir
- Fail properly if docker-compose build fails (was hidden by pipe)
- Use docker exec for backend health checks (port 8000 not exposed in prod)
- Add fallback API access via docker exec if nginx not reachable yet
- Show backend logs during wait for better debugging
- Increase timeouts for MITRE sync and data source sync
- Fix hardcoded localhost:8000 URLs in frontend to use relative /api/v1
path (works with Nginx proxy in prod and VITE_API_URL in dev)
- Create production entrypoint (entrypoint.prod.sh) that runs migrations,
seeds, and starts uvicorn with 4 workers (no --reload)
- Create comprehensive install.sh script for production deployment that
generates secure .env, builds containers, waits for health, and
optionally triggers initial MITRE sync
- Update docker-compose.prod.yml to use production entrypoint
- Update Dockerfile to make both entrypoints executable
- Remove init.ps1 (production will always be Linux)
- Update README with production deployment instructions
kitos
kitos