Two critical issues and several high/medium issues were identified during
a security review of the backend API.
**Critical fixes:**
1. Path traversal (CWE-22): user-supplied `simulation_id`, `report_id`,
and `project_id` values were passed directly to `os.path.join()`
without validation, allowing `../` sequences to escape intended
directories.
- Added `backend/app/utils/id_validator.py` with `validate_safe_id()`
(rejects anything that isn't alphanumeric/underscore/hyphen) and
`safe_join()` (resolves realpath and verifies containment).
- Applied to all 3 path-construction sites in simulation.py, all 12
relevant handlers in report.py, and 6 sites in graph.py.
- Sanitized uploaded filenames with `os.path.basename()` in graph.py.
2. Missing authentication: all API endpoints were publicly accessible
with no auth mechanism.
- Added `backend/app/utils/auth.py` with an `X-Api-Key` middleware
registered as a `before_request` hook.
- Auth is opt-in: set `API_KEY` in `.env` to enforce it; if unset a
startup warning is logged. This preserves local dev workflows.
**High fixes:**
3. Hardcoded `SECRET_KEY` fallback replaced with `os.urandom(32).hex()`
so an unset key is never predictable.
4. `FLASK_DEBUG` now defaults to `False` instead of `True`.
5. Full Python tracebacks removed from all API error responses (51 total
across graph.py, report.py, simulation.py) — tracebacks still go to
the logger.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
29fdb64fa0