Adds three skills:
humanizer (Communication & Writing): voice-anchored AI text humanization. Statistical fingerprint built from the author's own writing corpus replaces generic "clean" output with rewrites that match the author's sentence rhythm, vocabulary, and quirks. Bilingual EN/ES rule libraries. --diff mode scores any draft 0-100 against the fingerprint. MIT, fork of blader/humanizer with v2.6+ enhancements.
ingest-youtube (Creative & Media): pulls YouTube transcripts into a queryable markdown vault using the same connector pattern as Slack/WhatsApp/Notion/Linear/GitHub/Gmail ingests. yt-dlp with manual-subs > auto-captions > Whisper priority, VTT timing-marker cleanup, idempotent file writes to External Inputs/YouTube/<channel>/<date>-<title>.md, capture-seed stubs on trigger keywords. MIT, EN+ES default.
ai-brain-starter (Productivity & Organization): verification harness around an AI agent so memory compounds instead of corrupts. Markdown vault as ground truth, deterministic pre-write hooks, bi-temporal rule lineage, and a session-end cascade that files decisions, captures, and to-dos automatically. Bundles journaling with an advisory panel, weekly + monthly insight cycles, knowledge graph extraction with community detection, drift detection, and rule-conflict detection. MIT. In production against a 10K-file vault for 8+ months. Inspired by Microsoft DELEGATE-52 (arxiv.org/abs/2604.15597) finding that frontier LLMs corrupt 25% of professional content over 20 edits.
Thank you for maintaining the list.