Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -28,6 +28,8 @@ Get-FileHash .\NoMorePwn-<version>-Setup.exe -Algorithm SHA256

The checksum file is published in the same release as the binaries, so it detects corrupted downloads but does not protect against a compromised GitHub account or release. macOS and Linux have no packaged builds; run from source (below).

See [SECURITY.md](SECURITY.md) for reporting vulnerabilities and [docs/THREAT_MODEL.md](docs/THREAT_MODEL.md) for what the app does and does not defend against.

## Run from source

Use Python 3.10+ in a dedicated virtual environment:
Expand Down
27 changes: 27 additions & 0 deletions SECURITY.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,27 @@
# Security policy

NoMorePwn is a **pre-release**, single-maintainer project. It has not had an independent security audit. Please read [docs/THREAT_MODEL.md](docs/THREAT_MODEL.md) for what it does and does not try to protect.

## Supported versions

Only the latest `main` / most recent release is considered for fixes. There is no long-term-support branch. Note that the release marked "Latest" on GitHub may lag behind the newest pre-releases.

## Reporting a vulnerability

Please report privately; do **not** open a public issue for a suspected vulnerability.

Use GitHub's private vulnerability reporting: go to the repository's **Security** tab and choose **Report a vulnerability** (https://github.com/grloper/NoMorePwn-Password-Manager/security/advisories/new).

Helpful details: affected version or commit, platform, steps to reproduce, impact, and whether you only used disposable test vaults (please never send real vault data).

This is a volunteer project, so no response-time SLA is promised. The intent is to acknowledge a report within about a week and to aim for coordinated disclosure within 90 days, sooner if a fix ships earlier. There is no bug bounty.

## Scope

In scope: the vault and crypto code (`nomorepwn/`), the desktop app including local IPC and the updater (`nomorepwn_app/`), the browser extension and native messaging host (`extension/`), and the build/release workflows.

Out of scope: attacks that require an already-compromised, unlocked, same-user session (malware running as you, keyloggers, memory dumps), physical access to an unlocked machine, social engineering, and weaknesses in third-party dependencies that are not exploitable through this project (report those upstream).

## Known limitations

Windows installers are currently unsigned, and the update checksum is published in the same release as the installer. See the threat model for details.
57 changes: 57 additions & 0 deletions docs/THREAT_MODEL.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,57 @@
# Threat model

Status: informal, written from a code review (2026-10) and the source. It is **not** an independent audit. Statements below describe the intended design and what the code currently does; they are not guarantees.

## What is being protected

| Asset | Where it lives |
|---|---|
| Passwords and notes | Local SQLite vault (`vault.db` under the per-user data dir), encrypted per field |
| Master password / derived key | User's head; derived key in process memory while unlocked |
| Passwords captured from the browser | Briefly in the extension (in memory), native host, local IPC, then the unlocked vault |
| Recovery Kit (optional) | A user-held file that escrows the master key, wrapped under a 256-bit recovery code (optionally plus an authenticator second factor) |
| Update path | Installer downloaded from GitHub Releases |

## Cryptography (as implemented)

- Key derivation: Argon2id (t=3, m=64 MiB, p=4, random 16-byte salt), PBKDF2-SHA256 (600k iterations) as fallback. KDF name and parameters are stored in the vault and bounds-checked when read.
- Encryption: AES-256-GCM, fresh random 96-bit nonce per encryption, additional authenticated data binds a ciphertext to its row and field.
- Password generator uses Python's `secrets`.
- Minimum master password length is 10 characters. There is no unlock rate-limit; offline guessing is the realistic attack and depends on master password strength.
- KDF parameters in the vault metadata are not themselves authenticated (a writer could cause unlock failure; not a key disclosure).
- Master password is not Unicode-normalized, so the same visible password typed on different systems could in principle fail to unlock.

## Adversaries

| Adversary | Defended? | Notes |
|---|---|---|
| Thief with a copy of the vault file or a backup (stolen disk, cloud sync leak) | Largely | Secrets are encrypted; strength depends on the master password and Argon2id cost. **Service names, usernames and group names are plaintext** in SQLite. Files are created 0600 on POSIX; on Windows protection relies on the per-user `%APPDATA%` ACL. |
| Other local user on the same machine | Partly | Data dir 0700, vault and backups 0600, IPC socket restricted to the owning user (POSIX). Windows relies on directory ACLs. |
| Malware running as the same user | **No** | It can read process memory, keylog the master password, read the clipboard, read the IPC token file, or drive the UI. Out of scope. |
| Memory forensics / swap / hibernation | **No** | The key is held in immutable Python `bytes` and passwords are Python `str`; locking drops references but cannot zeroize memory. |
| Local process injecting captures via IPC | Mitigated | The IPC socket is user-only and every message must carry a per-install random token stored in a 0600 file (`ipc.token`). Same-user malware can read that token, so this stops other users and stray processes, not same-user malware. |
| Malicious web page | Partly | A page can fake or trigger a login submit and cause a bogus entry to be offered for capture. Capture success is heuristic (same-origin navigation, final HTTP status); cross-origin results go through an unverified confirmation path. The extension cannot read the vault; autofill is not implemented. |
| Network attacker | Mostly | Update and breach-check traffic uses HTTPS with certificate verification. Breach checks send only a 5-character SHA-1 prefix (k-anonymity). |
| Compromised GitHub account, Actions token or release | **No** | Installers are unsigned. The SHA256SUMS file comes from the same release as the installer, so it detects corruption, not a malicious publisher. The updater refuses releases with a missing or malformed checksum, refuses downgrades, only follows the GitHub "Latest" (non-prerelease) release, requires user confirmation, and locks the vault first. A compromised release could still run code on users' machines. Code signing and a signed checksum with a pinned offline key are not implemented. |
| Compromised dependency / build | Partly | `pip-audit` on `requirements.txt` found no known issues at review time. Runtime and build dependencies are version ranges, not hash-locked; builds are not reproducible. |

## Component notes

**Browser extension and native host.** The extension requests broad host access (content scripts on all URLs and frames) because it must observe logins on any site. Pending captures are held in memory, not in `chrome.storage`. The native messaging host is a courier only: it forwards captures to the desktop app over local IPC and never opens the vault. Locked vaults reject capture rather than queue plaintext.

**Local IPC.** Used by the native host and by a second app launch. Mitigated as above; messages are read in a single read, so very large messages may be truncated (reliability, not security).

**Clipboard.** Copied passwords are cleared after a timeout only if the clipboard still holds the copied value. The app does not currently ask Windows clipboard history or cloud clipboard to exclude the content, so those features may retain copied passwords.

**Auto-lock and sessions.** The desktop app auto-locks after inactivity. Re-keying revokes the old session.

**Recovery Kit.** Opt-in. Theft of both the kit and its recovery code (and authenticator, if enabled) yields the vault.

## Explicit non-goals and known gaps

- Protection against a compromised, unlocked or malware-infected device.
- Secure memory erasure.
- Metadata confidentiality (service names, usernames, groups are plaintext).
- Signed binaries and a signed update channel (planned, not done).
- Independent audit, reproducible builds, SBOM/provenance.
- Cross-platform packaged builds: only Windows builds are published.
Loading