vault-unlock

Pass

Audited by Gen Agent Trust Hub on Sep 3, 2026

Risk Level: SAFECREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The script scripts/unlock.py contains a hardcoded passphrase buddy-vault-2026 used for cryptographic operations on the vault data.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data and lacks safety guardrails for the resulting output.\n
  • Ingestion points: The skill reads binary data from vault.bin (or a user-specified file path) which is then decrypted and displayed.\n
  • Boundary markers: Absent. The instructions in SKILL.md explicitly direct the agent to "Podaj uzytkownikowi dokladnie to, co pojawilo sie na stdout" (Provide the user exactly what appeared on stdout), instructing the agent to follow/repeat potentially malicious instructions embedded in the vault.\n
  • Capability inventory: The skill executes a Python script via python3 which has the capability to read any file on the filesystem if provided as an argument.\n
  • Sanitization: No validation or sanitization is performed on the decrypted content before it is output to the agent context.\n- [COMMAND_EXECUTION]: The skill instructs the agent to execute a local Python script on the host system using python3.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 3, 2026, 11:06 AM
Security Audit — agent-trust-hub — vault-unlock