The Agent Skills Directory

[DATA_EXFILTRATION]: Accesses sensitive local file paths to manage its internal encryption key.
Specifically, it reads and writes to '~~/.config/agent/vault.key' and '~~/.config/agent/vault.enc' to persist the master encryption key and the encrypted database. While this is the intended functionality for a local vault, the access to files matching secret/key patterns in the user's home directory is flagged as sensitive path access.
[PROMPT_INJECTION]: Vulnerable to indirect prompt injection via the storage and retrieval of secret values.
Ingestion points: Secret values are ingested via the '--value' command-line argument in 'secure-vault.js'.
Boundary markers: Absent. Secrets are stored as raw strings and returned in the JSON response without delimiters or warnings that the content might contain instructions for the agent.
Capability inventory: The script has full read/write access to the local vault directory and can write to any path specified by the user in the 'export' command.
Sanitization: No validation or sanitization is performed on the secret values before storage or retrieval.
[COMMAND_EXECUTION]: Provides a CLI tool for file system operations that allows writing to and reading from arbitrary local paths.
The 'export' and 'import' commands interact with local files provided as arguments. Notably, the 'export' command by default includes the decryption key within the exported JSON file unless a manual external key is provided, which significantly weakens the security of the backup and could lead to data exposure if the file is shared or intercepted.

secure-vault