The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands to manage file snapshots in the user's home directory. Specifically, it uses cp to store different stages of text and Bash to locate profile directories within $HOME/.claude/voice-profiles/. These operations are consistent with the skill's stated purpose of managing local voice profiles.
[PROMPT_INJECTION]: An indirect prompt injection surface is present in modules/pattern-analysis.md. The skill interpolates user-edited text directly into an LLM analysis prompt ({post_edit_text}). While this could allow a malicious user to influence the voice learning logic via crafted edits, the risk is minimized because the skill requires explicit user approval in Step 5 before any changes are applied to the profile files.

voice-learn