The Agent Skills Directory

[COMMAND_EXECUTION]: The automated test suite in scripts/test_voice_check.py uses subprocess.run to invoke the skill's own internal scripts (voice_check.py, stylometry.py, and pick_exemplars.py) for integration testing. These calls are restricted to specific local file paths and do not involve a shell, preventing arbitrary command injection.
[DYNAMIC_EXECUTION]: The scripts/voice_check.py script performs dynamic module loading by searching for a sibling stylometry.py file in paths determined by the PERSONAL_TONE_STYLOMETRY environment variable and relative directory structures. This is a standard modular architecture for the voice-profile tool suite to share stylometric analysis logic.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes external content from draft.md files provided by the user, creating a surface for potential instructions hidden in the processed data.
Ingestion points: The checker script reads the entire content of a user-specified draft file passed as a command-line argument.
Boundary markers: The script output is structured with clear VOICE CHECK: labels and severity markers. Instructions in SKILL.md direct the agent to report these specific measurements to the user, providing context for interpreting the results.
Capability inventory: The checker script is functionally restricted to text parsing and statistical analysis; it does not possess capabilities for network exfiltration or modifying the host system.
Sanitization: The script implements a CTRL_RE regular expression to strip control characters from its rendered output, mitigating risks associated with malicious escape sequences or output poisoning.

voice-check