The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Bash shell commands to manage local directory structures and file organization for voice profiles. These operations are limited to the user's home directory and involve standard file management tasks.
Evidence: modules/sample-intake.md contains shell scripts to create directories (mkdir -p) and copy samples (cp) to $HOME/.claude/voice-profiles/.
[PROMPT_INJECTION]: The skill processes untrusted user-supplied writing samples, which presents a surface for indirect prompt injection. The design incorporates several defensive practices to mitigate this risk.
Ingestion points: Samples are collected via directory scanning or interactive text pasting in modules/sample-intake.md.
Boundary markers: Samples are explicitly labeled (e.g., "Sample 01") and compared against generated baselines, which helps the model treat them as data rather than instructions.
Capability inventory: The skill has the ability to read and write local files and execute shell commands.
Sanitization: modules/sample-intake.md includes an anonymization step that removes identifying metadata, proper nouns, and context from samples to focus the model on stylistic patterns.
[DATA_EXFILTRATION]: The skill reads user-provided files for analysis but performs all processing locally. No network operations or external data transmission patterns were found.

voice-extract