What problem does it solve? Writing a persona file by hand so AI-generated content sounds like a specific person is slow, and skipping it leaves every article, tweet, and digest in generic AI voice. This Skill reads a person's public footprint and drafts the identity-and-voice files automatically. ## Core Features & Use Cases - Multi-source identity extraction: Reads an X account via the X.AI Responses API (with x-mcp and WebSearch fallbacks), plus name-based web search and direct links like blogs, GitHub, and LinkedIn. - Structured soul files: Writes soul/SOUL.md (identity, worldview, opinions, tensions), soul/STYLE.md (tone, punctuation, anti-patterns), and soul/examples/good-outputs.md (10-20 real voice-calibration posts). - Provenance guardrails: Rejects biographical claims that cannot be traced to the actual subject, preventing conflation with same-named people or examples inside the subject's own repos. - Use Case: Point it at a founder's X handle before launching a content pipeline, then review and edit the drafted soul in the dashboard so every downstream skill writes in that person's voice. ## Quick Start Build a soul from the X handle karpathy and draft the SOUL.md, STYLE.md, and voice examples files for review.