The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes local Python scripts (voice_analyzer.py and voice_validator.py) via the Bash tool to perform quantitative analysis and validation of writing patterns.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it requires ingesting 50-100+ writing samples from untrusted external sources such as Reddit or HackerNews. Maliciously crafted content within these samples could influence the generation of the voice skill's instructions. * Ingestion points: Writing samples stored in skills/voice-{name}/references/samples/. * Boundary markers: Uses XML tags (<context>) and horizontal rules to separate instructions from sample data. * Capability inventory: Access to Bash (script execution), Read/Write (file management), and the Skill tool (creation of new agent skills). * Sanitization: No explicit content sanitization or instructional filtering of the ingested samples is described.
[PROMPT_INJECTION]: The instructions explicitly direct the agent to override default AI safety and neutrality hedging (e.g., "It's worth noting", "One might argue") in favor of "unhedged strong opinions" to achieve human-like authenticity. This represents a deliberate behavioral override of standard model constraints for functional purposes.

voice-calibrator