The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/ataru_recall.py uses subprocess.run to invoke the local ataru or lovcode binary. The execution is performed by passing a list of arguments directly to the process, which prevents shell-based command injection. Additionally, the script enforces a version check (--version) to ensure the binary is at least version 0.41.3 before performing any operations.
[DATA_EXFILTRATION]: The skill is designed to retrieve historical AI conversation data from a local Ataru index. This involves accessing sensitive user data, but the implementation is local-only and does not include any network communication or data exfiltration mechanisms. The scripts also use projections and clipping to ensure that only relevant snippets are loaded into the agent's context.
[CREDENTIALS_UNSAFE]: The utility scripts/profile_store.py manages persistent user settings. It includes a security filter (SENSITIVE_PARTS) that prevents the storage of fields containing sensitive keywords like 'token', 'secret', or 'password' in the shared profile, mitigating the risk of inadvertent credential persistence.

lov-ataru-search