What problem does it solve? Writing LinkedIn comments that sound like a real person instead of generic AI output is hard, and most generated comments rely on empty praise, banned buzzwords, or invented experience. This Skill produces 2-3 candidate comments grounded in the user's real background and a growing dictionary of previously approved comments. ## Core Features & Use Cases - Three-step comment engine: picks distinct angles (extension, lived experience, contrarian, domain transfer, tactical), builds substance with at least two levers, then humanizes format and rhythm. - Voice dictionary with few-shot learning: approved comments are appended to voice-dictionary.md and reused as style anchors, while rejected patterns are logged as anti-patterns to avoid repeating mistakes. - Strict guardrails: no invented facts or numbers (uses [NÚMERO?] placeholders), no em-dashes, banned AI-sounding words in PT and EN, PT-BR tells like gerundism and officialese are rewritten, and output language always matches the post. - Use Case: Paste a LinkedIn post about spec-driven development and receive three labeled comment candidates of 40-100 words each, in the post's language, ready to copy after selecting the one that sounds most like you. ## Quick Start Paste a LinkedIn post's text or screenshot and ask for a comment in your voice, then pick which of the generated candidates sound like you so the dictionary learns.
