tech-article-pipeline

name

tech-article-pipeline

description

Write publishable technical articles from a clear topic plus a small amount of background. Use when the user wants to turn a technical practice, architecture change, tool evaluation, workflow upgrade, or incident analysis into a WeChat-ready draft and an X thread. The workflow is collaborative: decide whether research is light or deep, draft 1-2 paragraphs at a time, and ask only one direction-changing question when needed.

Tech Article Pipeline

Overview

This skill is for technical content production, not generic copywriting.

Input should be:

• a clear topic
• a small amount of background or intent
• optional source links, notes, screenshots, or local files

Default outputs:

• a publishable WeChat draft
• an X thread version
• a Markdown file written into the Obsidian draft directory

Workflow

Read references/workflow.md before starting.

Use this workflow:

1. Clarify the topic and desired article direction from the user's short background.
2. Decide whether the task needs light research or deep research.
3. Draft only 1-2 paragraphs at a time.
4. Ask at most one direction-changing question at a time, only when the answer would materially change the article.
5. Continue drafting from the approved direction instead of restarting from scratch.
6. Converge into:
   • WeChat long-form draft
   • X thread version
   • optional title candidates and cover copy

Research Mode Selection

Use agent-reach when:

• the topic needs current web information
• the user provides links to web pages, videos, posts, or platform content
• you need lightweight fact collection fast

Use deep-research when:

• the topic is disputed, high-stakes, or strongly time-sensitive
• the article depends on multi-source comparison
• source quality will materially affect the argument

If the topic is mainly a personal technical practice recap, default to light research.

Drafting Rules

Do not force a fixed article template.

Instead, enforce these writing constraints:

• open with a concrete entry point, not generic framing
• surface the author's core judgment early
• spend most of the article on architecture, implementation, tradeoffs, and concrete details
• preserve technical facts, names, paths, commands, versions, and dates
• prefer specific evidence over broad claims
• avoid standard AI listicles unless the material is inherently list-shaped
• keep the collaboration iterative: each round should move the draft forward, not re-summarize the whole article

If the user dislikes the current direction, revise from the last accepted section instead of regenerating the whole draft blindly.

Output Contract

Read references/output-spec.md before finalizing outputs.

Default article file target:

• root: /Users/zhangyufan/Workspace/Obsidian/我的大脑/1-Process/01-项目/公众号运营/钢之AI术师
• draft directory: /Users/zhangyufan/Workspace/Obsidian/我的大脑/1-Process/01-项目/公众号运营/钢之AI术师/草稿

When writing the final result:

• write the WeChat draft to the Obsidian draft directory
• keep Markdown frontmatter if the user wants a ready-to-publish note
• include an X thread version in the conversation output
• optionally include:
  • 3-5 title candidates
  • a short summary/deck
  • cover image copy suggestions

What Not To Do

• do not turn every article into a tutorial
• do not force a six-part outline when the topic does not need it
• do not ask a question after every paragraph
• do not overwrite an existing draft unless the user explicitly asks
• do not fabricate citations or claim verification you did not perform