Skip to content

Security: Davincc77/vaultd

Security

SECURITY.md

Security Policy

Threat Model

.vaultd is a client-side encrypted file format. The following threats are in scope:

Threat Mitigation
File theft (device compromise) AES-256-GCM + Argon2id. Attacker must brute-force passphrase.
Passphrase brute-force Argon2id m=65536/t=3/p=1 makes GPU attacks expensive
File tampering AES-GCM auth tag detects any modification
AAD bypass 5-field JCS-canonicalized AAD covers all envelope metadata
Prompt injection via file agent_instructions is user-context only, not system authority
Private key exfiltration Spec forbids private keys. Agents MUST refuse to process them.

Out of scope

  • Passphrase compromise (social engineering, keylogger) — outside file format scope
  • OS-level memory attacks — outside file format scope
  • Side-channel attacks on Argon2id implementation — use a hardened library

Known limitations

  • Argon2id parameters (m=65536/t=3/p=1) are minimum recommended. Users with high-value portfolios should increase m to 131072 or higher.
  • The unencrypted variant (encrypted: false) offers zero confidentiality. For example/test use only.
  • agent_instructions is 4096-char max. Malicious content could attempt prompt injection — agents must treat it as untrusted user input.

Responsible disclosure

Report security vulnerabilities to: Luxlearn@pm.me

Please include:

  • Description of the vulnerability
  • Steps to reproduce
  • Potential impact assessment
  • Suggested fix (if any)

We commit to responding within 72 hours and to crediting researchers who report valid findings.

There aren't any published security advisories