The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to run a Python script (scripts/voice-stylometry.py) for stylometry analysis. The command includes a profile path constructed using a {name} variable (skills/voice-{name}/profile.json), which represents a potential command injection vector if the voice name is not strictly validated before interpolation.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process user-provided content (drafts) as its primary function, creating an attack surface for instructions embedded within the text.
Ingestion points: The draft content is loaded from a file and scanned for violations in Phase 2.
Boundary markers: No explicit delimiters are used to wrap the untrusted content or instruct the agent to ignore instructions within the draft.
Capability inventory: The skill possesses Bash, Read, Write, and Edit tools.
Sanitization: The instructions do not prescribe any sanitization or validation of the draft content prior to analysis.
[METADATA_POISONING]: The references/narrative-patterns.md file contains a reference to a likely fabricated academic citation ("Russell et al. (2026), arXiv:2604.03136v4"). The future date and invalid arXiv format indicate deceptive metadata designed to grant unearned authority to the skill's validation logic.

voice-validator