Skip to content

Security: safesploitOrg/1time-view

Security

SECURITY.md

Security model

Supported scope

0.2.0-alpha.1 is an initial development implementation, not an audited or production-approved service. The project follows SFW v1.1.0 with the explicit development-only Playwright exception EX-001.

Trust boundaries

  • Browser-generated per-note AES-256-GCM keys remain in the fragment and current page memory. Honest clients upload only encrypted envelopes.
  • The server sees ciphertext size, selected duration, creation/expiry times, retrieval tokens while processing, and network metadata. It hashes tokens before storing them.
  • A database copy alone lacks the key. A malicious live server can retain ciphertext, deny service, or replace JavaScript to capture keys/plaintext. HTTPS and CSP do not eliminate trust in client delivery.
  • The token alone can consume the encrypted note; the complete link also permits decryption. The link does not authenticate its holder as the intended person.
  • The consume transaction finishes and commits before returning the envelope. Delivery, successful decryption and human reading are not part of the consumption decision.
  • GET/HEAD/prefetch are harmless; an automated client that performs the actual POST can consume a note.
  • No plaintext/key logging, request body logging, third-party browser assets or persistent browser message state is intentional. Operator-added proxies, telemetry and extensions can change that boundary.
  • Logical deletion is not guaranteed physical erasure. Do not restore old note stores or promise that messages cannot be copied.

The threat/control/test mapping and residual risks are in ARCHITECTURE.md.

Verification limits

Unit, real HTTP/SQLite and process-level tests do not replace browser verification. Playwright sources exist but checks are disabled by default. Container build/run and image-vulnerability scans require Docker and were not run in this environment. Local project SAST rules have a deliberately limited scope; passing them is not a security audit.

Finding policy

Block a public/production release for exposed credentials, known active exploitation, reachable Critical vulnerabilities or newly introduced reachable/exploitable High vulnerabilities. Triage other findings with reachability, exposure, asset impact, a named owner and a remediation/review date. No unverified inherited High/Critical baseline is accepted automatically.

Gitleaks initially identified two reference-fixture keys. The exact synthetic/public key line AND exact fixture path are allowlisted in .gitleaks.toml; other values and paths remain scanned. These keys are published test data, never service credentials.

Reporting

No public security contact/repository has been configured. Report privately to the instance operator/project owner through your established channel. Do not include live share links or user message plaintext in public issues. Add a verified security-reporting route before public release.

Release approval requires recorded human owner review of the exact implementation, threat model, test results and operational recovery plan. No such approval has been granted by this development task.

Admin boundary added in 0.2

The admin dashboard has its own listener, port, static root and origin. It is protected by a generated 256-bit bearer credential stored privately outside the database; the API checks a constant-length digest on every request. The default Docker host binding is loopback. Token files with unsafe permissions or symlinks fail startup. Dashboard authentication is distinct from per-message encryption and never supplies message decryption keys to the server.

Admin API responses contain only aggregate statistics. No note identifiers/hashes, ciphertext, full links, plaintext, credentials or private paths are returned. Dashboard reads do not consume, clean up or mutate notes. No destructive/admin-write endpoints exist. Statistics include private operational metadata, so authentication and network restrictions remain necessary.

No cookies, persistent browser token storage or cross-origin permissions are used. Host/Origin/fetch-site checks and request budgets add defence in depth; they do not authenticate arbitrary HTTP software. Separate ports in one process do not provide an OS-level sandbox. The admin client and host remain trusted. A leaked admin token grants read-only operational access until the credential is rotated and the service restarted. Page sign out clears local references; it does not revoke that credential.

See admin operations and verification evidence.

There aren't any published security advisories