article-forge

Article Forge — Opinions + Materials → Publishable Article

Input

Two types of materials are needed:

1. Insight summary: structured insights output by insight-extractor (workspace/insights/*.md）
2. Raw content: transcripts, notes, reference materials (provide context and details)

Article generation process

Step 1: Determine the article positioning

Based on user instructions or automatic judgment:

Dimension	Options
Platform	Blog / Zhihu / WeChat official account / Other
Style	Opinion piece / Tutorial / Retrospective / Dialogue style / Narrative
Length	Short (1-2K) / Medium (3-5K) / Long (5K+)
Audience	Tech community / General internet / Public
Tone	Laoguai style (casual + sharp) / Serious technical / Science-friendly

Step 2: Select insights and materials

Choose from the insight summary:

• 1 main thesis: the article's core argument, must be clear and powerful
• 2-3 supporting points: argue the main thesis from different angles
• 1-2 golden lines: serve as the article's "memorable points"
• 1 controversy/twist: create reading tension

Supplement from the raw content:

• Specific cases, data, dialogue
• Scenes and details (increase authenticity)

Step 3: Build the article structure

Do not use AI boilerplate structures("First... second... finally...", "not X but Y").

Recommended structure patterns:

Pattern A: Story introduction → Insight explosion

1. Start with a specific scene/story (taken from the materials)
2. Lead into the core insight
3. Expand the argument (use cases, not reasoning)
4. Ending (leave an aftertaste, do not summarize)

Pattern B: Controversy entry → Two-sided development

1. Throw out a controversial judgment
2. Evidence for the affirmative side
3. Evidence for the opposing side
4. Your stance (clear and unambiguous)

Pattern C: Problem-driven → Layer-by-layer progression

1. A specific problem
2. The common but wrong answer
3. The real answer (counterintuitive)
4. Why everyone thought wrong

Step 4: Writing

Follow these rules:

Core writing principles

• Speak like a human: write for real people to read, not for search engines
• Have a stance: dare to say "this thing doesn't work," dare to say "I just like it"
• Give details: abstract opinions are worthless, concrete scenes have value
• Leave breathing room: short paragraphs, fast pace, not tiring to read

Anti-AI check (must check after writing)

Read the anti-AI rules for the corresponding platform:

• Technical article → skills/zhiforge/references/anti-ai.md
• Story/Novel → skills/salt-story/references/anti-ai.md

General taboos:

• ❌ "Not… but…" sentence pattern
• ❌ Triple parenthetical explanations xxx（即yyy）
• ❌ Overuse of bold (more than 2 bold instances in a paragraph)
• ❌ "In conclusion," "To sum up," "It is worth noting that"
• ❌ Neat three-part/five-part structure
• ❌ Every paragraph is about the same length

How to cite source material

• Directly quote the speaker's original words (keep the colloquial feel): "He said something really harsh at the time—'xxx'"
• Don't turn colloquial speech into formal written language—formal written language actually feels fake
• Cite the source but don't make it academic: "This was mentioned in that livestream the other day"

Step 5: Self-check

After writing, check item by item:

• Is the title engaging? (Within 20 characters, with suspense or attitude)
• Can the first paragraph hook people? (3-second rule)
• Does it have an AI flavor? (Run through the anti-AI checklist)
• Are the punchy lines used?
• Does the ending summarize again? (Delete the summary paragraph)

Step 6: Output and archiving

Save to the corresponding directory:

• Blog → write directly into the blog repository's _posts/
• Zhihu → workspace/stories/zhihu/
• Other → workspace/articles/

File format: Markdown, with YAML frontmatter (if it's a blog).

Publish

After the article is finished, choose the publishing method according to the target platform:

• Zhihu: call the zhiforge skill's publishing process (markdown import → fill in title → submit to question → publish)
• Blog: git commit + push to GitHub Pages
• WeChat Official Account: export markdown, publish via the WeChat Official Account backend or API

Integration with other Skills

• Upstream:insight-extractor(provides structured viewpoints),video-to-text(provides raw material)
• Downstream:zhiforge(Zhihu publishing),salt-story(if it needs to be adapted into a story)
• Complete chain: video → transcript → viewpoints → article → publish