molu 墨炉

A Chinese-language content production engine packaged as AI Agent Skills — one package, eleven modules, from insight to review. Works with Claude Code, Codex, WorkBuddy and other Skills-compatible Agents.

Note

Distilled from 899 public WeChat Official Account articles (2015-09 ~ 2018-12, 2.13M characters) — 16,227 content units verified against source text with a 100% match rate.

English | 简体中文

---

Why

Producing Chinese content today is a sequence of isolated, re-invented decisions: who are you writing for, how does the trend connect to a timeless pain point, what structure carries the story, is it safe to publish, why did that piece break out. Writing tools help you type faster; almost none help you decide better.

molu 墨炉 is a decision pipeline, not a text generator. Depth over speed: it turns a content task into a checkable pipeline from insight to review.

What

molu is a Chinese-language content production engine packaged as 11 AI Agent Skills. It covers insight research, strategy positioning, trending-topic selection, writing, pre-publish review, and viral teardown — installable into any Skills-capable AI Agent.

Core belief: depth over speed. When you catch a trending topic, first ask what timeless pain point sits underneath it; before generating topics, confirm who you're writing for; after writing, the draft must pass review. It is not a traffic predictor — it delivers checkable methods, not promises.

Pipeline

  原料层  Corpus Source (pluggable — swap sources without touching the engine)
    方法论(4A·咨询框架) · 素材库(16,227 单元) · 实时热榜
                ↓
         kb.py · unified data layer
                ↓
  主管道  Pipeline (①→⑤ in order)
     ① 洞察 insight → ② 策略 strategy → ③ 雷达 radar
     → ④ 选题 topic/reuse → ⑤ 生产 writer/video
                ↓
  收束层  Closing
     ⑥ 审核 guard → ⑦ 拆解 teardown ⇄ back to ①
     ⑧ lens ◈ 视角 — runs throughout

Modules

#	Module	Purpose
①	/molu-console	Show pipeline map, diagnose stage, dispatch modules
②	/molu-insight	Separate observation from insight; analyze reader blockage in four layers
③	/molu-strategy	Define positioning, audience, cadence, format — including "what NOT to do"
④	/molu-radar	Collect real-time hot lists, filter compliance risks, output writable angles
⑤	/molu-topic	20 topics, each with target reader and rationale
⑤	/molu-reuse	Ten different angles, five angle-swapping methods
⑥	/molu-writer	Produce publishable drafts to platform specs
⑥	/molu-video	First-3-seconds hook, shot rhythm, caption splitting
⑦	/molu-guard	Legal / ethical / platform three-layer check, three verdict tiers, read-only
⑧	/molu-teardown	12 metrics + percentile positioning, based on a measured baseline
◈	/molu-lens	Stress-test writing decisions with practitioner standards, runs throughout

Install

Choose a method in your terminal (never send install commands to an Agent as chat content).

Skills CLI

npx -y skills add Guislabert/molu -g --all

-g installs globally; --all installs into every Agent the CLI recognises. To target one Agent, drop --all.

install.sh

bash install.sh agents           # source into ~/.agents/skills
bash install.sh all              # source + symlinks to pi / claude / codex

After installing, enter in the Agent chat:

/molu-console what does this do

Quick start

1. First run — learn the entry point:

   /molu-console what does this do
   

   The console maps the pipeline, lists all 11 modules, identifies where you're stuck, and suggests the smallest next step.

2. Already know what you need — invoke a module directly:

   /molu-topic  I write about money-saving for young professionals. Give me 20 ready-to-write topics, each with an angle and rationale.
   /molu-writer Here's my topic and material. Write a publishable piece to WeChat specs.
   /molu-guard  This piece is going to WeChat. Run the pre-publish review and flag risk spots.
   /molu-teardown Break down why this 100k+ article went viral and show my gaps.
   

3. Connect a corpus source (optional) — methodology works out of the box; with a source connected, topic / reuse / teardown can cite verified material:

   python3 molu/kb.py stats
   python3 molu/kb.py install <id>

4. Refine in the same conversation — provide new facts; any finished piece goes through /molu-guard before publishing.

Adopting / Contributing

Full engine documentation lives in molu/README.md. version discipline follows SemVer 2.0.0 and is guarded by tools/version-check.mjs.

Methodology provenance and framework acknowledgements (including dontbesilent2025/dbskill) are detailed under License.

Feedback

Raise an issue for a bug, a module idea, or a question about the method.

Acknowledgements

• Method: influenced by dbskill and the 4A / consulting-framework tradition; see License for provenance.
• Header/layout: readme-harness v0.1.0 (@8c65495).

License

This project's methodology is licensed under CC BY-NC 4.0 — non-commercial use for personal learning and research must credit the source; commercial use requires contacting the author.

Content boundary: this repo contains no verbatim corpus. When using this engine to create content, replace output with your own original phrasing; do not fabricate cases, and do not promise traffic numbers.

---

Layout: readme-harness v0.1.0 @8c65495 · baseline: docs/readme-harness.md