ai-topic-scout

# AI Topic Scout ## Overview Use this skill to scan a fixed set of essential AI information sources and turn current signals into creator-ready topic ideas. Prioritize freshness, source quality, and audience relevance over raw volume. ## Workflow 1. Load `references/sources.md` for the current source list, source roles, and topic filters. 2. Browse or otherwise verify current items from the sources before claiming anything is new, latest, released, ranked, or trending. 3. Collect signals across at least 6 of the 10 sources when the user asks for a daily run. If browsing is blocked, state the limitation and use only verifiable local/user-provided context. 4. Deduplicate overlapping news. Merge repeated stories into one stronger topic with multiple source angles. 5. Score candidates with the topic rubric below. 6. Return the best topics in a creator-facing format. ## Daily Output Format Default to 8-12 topic ideas unless the user requests another number. For each topic, include: - `选题`: A short punchy title in Chinese. - `为什么现在`: The freshness signal and source basis. - `内容角度`: The opinion, explainer, tutorial, comparison, or test angle. - `适合形式`: Short video, long video, thread, newsletter, live demo, or carousel. - `素材来源`: Source names and links. - `优先级`: High / Medium / Low with a one-line reason. End with a short `今日首选` section naming the top 1-3 topics and why. ## Topic Rubric Score each candidate from 1-5: - Freshness: Is it new or newly resurging today/this week? - Creator leverage: Can a creator add explanation, demo, opinion, or comparison beyond repeating news? - Audience value: Does it help viewers understand what changed, what to use, what to avoid, or what to learn? - Evidence strength: Is it backed by primary sources, papers, code, benchmarks, or reputable analysis? - Distinctiveness: Is it less likely to be a generic repost everyone will make? Prioritize topics with high creator leverage and evidence strength. Do not over-prioritize incremental product announcements unless they change user behavior or the market map. ## Angle Patterns Use these patterns to turn source signals into content: - New model/product release -> "What changed, who should switch, and what still fails?" - Paper breakthrough -> "Explain the core idea with one visual metaphor and one real use case." - Open-source repo spike -> "Hands-on test: can normal users actually use it?" - Leaderboard movement -> "Benchmark drama: what this ranking does and does not prove." - Safety/policy update -> "What builders and creators need to change now." - Big-company strategy -> "What this reveals about the next platform war." - Toolchain update -> "Workflow before/after demo for creators or developers." ## Quality Rules - Use primary sources first: official blogs, papers, repos, changelogs, leaderboards, and docs. - Treat X/Twitter, Reddit, and newsletters as discovery/context unless they link to stronger evidence. - Include concrete dates when discussing "today", "yesterday", "latest", or "this week". - Separate confirmed facts from inference. - Avoid hallucinating product details, benchmark ranks, pricing, availability, or release dates. - If sources disagree, mention the disagreement rather than smoothing it away. ## References - `references/sources.md`: The 10 core AI information sources, what to pull from each, and search patterns.