The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/find_ai_file.py uses subprocess.run to call the ripgrep (rg) utility for high-performance searching of transcript files. The search pattern is constructed from user-supplied terms which are escaped using re.escape() to mitigate command injection risks. The execution is limited to local directories and does not involve the system shell.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from local AI chat logs (.jsonl files) which are considered untrusted external data. It processes this data to extract file paths and short evidence snippets for the user. The risk of the agent following instructions embedded in historical logs is mitigated by the skill's design as a passive retrieval tool that provides specific metadata rather than executing the conversation content.
[DATA_EXPOSURE]: The skill reads private conversation histories to identify file locations. To protect this data, the script scripts/find_ai_file.py ensures that its local search cache is stored with restricted filesystem permissions (0600), preventing access by other users on the system.
[CREDENTIALS_SAFE]: The profile management script scripts/profile_store.py includes a validation check that explicitly prevents the storage of sensitive keys (such as token, secret, password, api_key) in the shared user profile, ensuring that authentication credentials are not accidentally persisted.

lov-search-file