The Agent Skills Directory

[DATA_EXFILTRATION]: The skill records interaction data (prompts, tool parameters, and tool outputs) to unencrypted local files in ~/.claude/conversation-logs/. This behavior is the primary intended function and is documented with associated privacy warnings.\n- [COMMAND_EXECUTION]: The skill includes shell scripts intended to be used as Claude Code hooks and suggests using system utilities like crontab for automated log maintenance.\n- [PROMPT_INJECTION]: Documentation suggests feeding previous log files back into the agent for context, which creates an attack surface for indirect prompt injection if a previous session contained untrusted or malicious content.

conversation-logging