writing

原文涵盖的主题

• Writing (per-channel, one skill)
• Step 1 — pick the channel, then load its file
• Step 2 — content before packaging (the core lesson of the method this adapts)
• Rewriting an existing piece: 重构 or 从零
• Step 3 — cross-channel hard rules (always apply)
• AI-smell checklist (adapted from dbskill dbs-ai-check)
• Iterating these styles

This SKILL.md is the shared layer: how to pick a channel + rules that apply to every piece. Each channel's detailed contract is a separate file under references/ — load only the one you need (keeps this skill iterable: improve one channel without touching the rest). Methodology note: the structure + the AI-smell checklist below are adapted from dbskill (dontbesilent), a public writing-method set whose current licence is all-rights-reserved: adapting its methods for your own work is allowed; redistribution, repackaging, publishing modified copies and AI training are not. So absorb the frameworks in our own words and never paste its skill text into a repo we publish — and don't copy its specific copy/voice either. Nothing here depends on the original being installed. | Destination | Load this | Lang default | One-line shape | |-------------|-----------|--------------|----------------| | Academic paper/section | references/academic.md | EN (or ZH) | claim → evidence → method → limitation | | Technical blog | references/blog.md | EN/ZH | series: one article = one decision; 表象（标题/开头/骨架）+ 内核（误判故事/可回溯事实） | | X / Twitter | references/x-thread.md | EN/ZH | one idea/post, hook first | | Redd

这里只展示节选，请以完整原文和当前要求为准。