claude-skills

Enforce a requirement/specification phase before any substantial content work begins. Use this skill whenever the user asks Claude to write, expand, revise, translate, or build content that will be 500+ words, or any task involving document generation (docx, pdf, pptx), multi-section drafting, content expansion/revision passes, or translation work. Also trigger when the user says 'write a chapter', 'expand this section', 'revise the draft', 'translate this', 'build the doc', or any similar content production request. Claude MUST draft a Content Spec and get approval BEFORE executing. This skill also manages a Requirements Ledger that accumulates decisions across iterations. Trigger this skill even if the task seems straightforward, because the spec catches failure modes that only become visible after wasted work.

Captures learning moments from Claude sessions and persists them to a GitHub knowledge repo via the GitHub MCP Worker. Use when the user says "save this", "push this to my notes", "capture this", "extract my learnings", "checkpoint", or runs /learned. Also triggers at the end of sessions with significant learning content, or when Claude explains a root cause, makes an architectural decision, or breaks a debugging spiral. Works in Claude.ai Chat, Cowork (via Desktop Commander bash), and Claude Code (via /learned slash command).

Use when Han wants to start learning a new technical skill or topic from scratch and wants a structured, research-backed path instead of starting cold. Trigger phrases include "I want to learn X", "teach me X", "help me learn X", "I want to pick up X", "get me started on X", "build me a learning path for X", "/learn-skill X", "how do I get good at X". Runs state-of-the-art research on X, compares the leading tools/methodologies with a defended pick (via the tool-comparison rubric), and emits a structured hands-on learning track under `learning/<topic>/` following the repo's learning-track layout, anchored on the evidence-based Skill-Loop methodology. Topic-agnostic. NOT for processing a class transcript (use learning-day-process), NOT for a single concept question (use concept-explain), NOT for general non-learning research (write to research/ directly), NOT for downloading books (use annas-fetch).

Sharpens and improves prompts before executing. Triggers on any non-trivial prompt where better phrasing would produce deeper, more specific answers. Use when the user submits a vague, broad, or under-specified prompt. Skip for straightforward operational tasks (file edits, git commands, direct instructions).

Extracts transcripts from short-form video URLs and files using Supadata MCP (primary) or local whisper (fallback). Use when the user shares a video URL from Instagram Reels, Facebook videos/reels, Threads video posts, TikTok, Twitter/X, or uploads any video file (mp4, mov, webm). Trigger on "transcribe this reel", "extract transcript from reel", "what does this reel say", "transcribe this video", or any instagram.com/reel/*, facebook.com/watch/*, facebook.com/reel/*, threads.net/*/post/*, tiktok.com/*, x.com/*/status/* URL. Pairs with knowledge-capture to push transcripts to the GitHub repo.