The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/ataru_index.py executes a local binary named ataru (or lovcode) using subprocess.run. The command is built using a list format which prevents shell injection. The binary resolution logic searches through environment variables (ATARU_BIN), the system PATH, and standard application bundles on macOS (e.g., /Applications/Ataru.app).
[DATA_EXPOSURE]: The skill manages a local user profile via scripts/profile_store.py to persist the resolved binary path across sessions. This script includes a blocklist (SENSITIVE_PARTS) that explicitly prevents keys containing sensitive terms like 'token', 'secret', or 'password' from being written to the persistent profile, adhering to safe configuration management practices.
[SAFE]: The skill's behavior is entirely consistent with its documented purpose of managing a local indexing service. It includes robust version gating (requiring Ataru 0.41.3+) to prevent unintended side effects, such as older versions opening GUI windows during background tasks.
[INDIRECT_PROMPT_INJECTION]: The skill parses JSON output from the local ataru binary. While this represents a data ingestion surface, the risk is minimal as it processes local tool metadata (index size, session counts) rather than untrusted remote content. The analysis found no evidence of unsafe interpolation of this data into instructions.

lov-ataru-indexing