Maoxuan Product Agent

On Contradiction × On Practice complete method distillation

Maoxuan
Product Agent

Turn a complex product problem into one decision worth executing.

A Chinese-first AI product manager and product decision Agent Skill for prioritization, growth, operations, data, delivery, and cross-team collaboration. The source method stays backstage; every answer uses modern product language.

Install in 30 seconds View source
36
work scenarios
36/36
self-tests passed
12
public cases
MIT
open-source license
See a real decision
Agent Skills format Codex Claude Code Cursor Defaults to Simplified Chinese

Example decision

A higher click-through rate is not yet a business win

The skill does not average ten plausible tactics. It identifies what the current evidence proves, what it does not prove, and which decision should come next.

“Our A/B test increased clicks by 12%, but orders did not increase. Should we roll it out?”
Output field Decision
Problem judgment The entry metric improved, but the business outcome did not. Do not roll out yet.
Next action Split product-page dwell time, add-to-cart, payment start, and payment success by variant.
Decision signal Ship only if downstream conversion or order value improves without damaging guardrails.
Stop list Do not treat CTR as the primary outcome, and do not mix unrelated changes into the test.
Read the full Chinese case Inspect the evaluation assets

Reasoning engine

It distills decision moves, not quotations

The engine was designed after complete readings of On Practice, On Contradiction, and related Volume I essays. Default output contains no source quotation, history lesson, political framing, or character role-play.

  1. 01FactsSeparate evidence, behavior, feedback, and assumptions
  2. 02BottleneckFind the current problem that most constrains the result
  3. 03MechanismDetermine which force or rule is driving the outcome
  4. 04ActionChoose the smallest useful and reversible move
  5. 05UpdateRevise the judgment from real-world results

Start from observable reality

User behavior · business data · frontline evidence

Find the current primary problem

Core bottleneck · critical path · resource trade-off

Return the judgment to practice

MVP · staged rollout · A/B test · data review
Read the source-reading and translation audit

Why it is different

Make the trade-off before giving advice

A generic product prompt Maoxuan Product Agent
Lists every plausible direction Finds the bottleneck that matters in the current stage
Applies a familiar framework immediately Checks the audience, stage, constraints, and evidence first
Treats a metric or feedback quote as a conclusion Separates facts, hypotheses, second-hand claims, and isolated cases
Recommends a complete large solution Uses a minimal diagnosis or reversible test when evidence is weak
Says what to do Also states what not to do and when to change course

36 recurring scenarios

From roadmap conflict to metric anomalies and delivery risk

The knowledge system is organized around the work product leaders actually face, not around chapters from the source texts.

Product & planning

Requirements prioritization

Version planning

Roadmaps

MVP and staged rollout

Enterprise requests

Strategy shifts

Growth & operations

Growth stagnation

Acquisition and channels

Retention and conversion

Community cold start

Content supply

Campaign performance

Data & monetization

DAU and MAU

Metric anomalies

Data definitions

A/B tests

CAC, LTV, and ROI

Pricing and membership

Delivery & organization

Executive interruptions

Resource constraints

Project delays

Cross-team collaboration

OKRs and KPIs

Retrospectives

Browse 12 public Chinese decision cases Inspect the 36-case test matrix

Install

One command for Codex, Claude Code, and Cursor

Standard Agent Skills package. No server, API key, paid dependency, or runtime dependency.

Install the skill globally in all three supported agents.

npx skills add atdy/maoxuan-product-agent --skill product-decision-agent --agent codex claude-code cursor -g -y

No Node.js or Git? Download the standalone v1.0.3 Skill package. After installation, describe a real product problem or invoke $product-decision-agent or /product-decision-agent explicitly. The skill answers in Simplified Chinese by default.

FAQ

What to know before using it

Is Maoxuan Product Agent a political or historical skill?

No. It is a product decision skill. Unless source tracing is explicitly requested, its answers contain no political role-play, historical explanation, theory lesson, or source quotation.

Do users need to read On Practice or On Contradiction first?

No. Users only provide the real product, growth, operations, data, delivery, or collaboration problem. The source method remains in the background.

Why does the skill answer in Simplified Chinese?

It is designed for day-to-day internet product work in mainland China. Standard product terms such as DAU, GMV, CAC, LTV, ROI, MVP, A/B Test, OKR, KPI, and Roadmap remain in English where useful.

Which AI agents are supported?

The standard Skill package supports Codex, Claude Code, Cursor, and other tools compatible with the Agent Skills directory format.

Has the skill been tested?

Yes. The repository publishes a 36-scenario evaluation matrix, representative outputs, deliberate failure cases, four clean-session forward tests, and automated quality gates.

Can it replace user research, data validation, or product accountability?

No. It clarifies decisions, next actions, and validation signals, but it does not invent missing evidence or replace research, data checks, legal judgment, or final business accountability.

Next step

Give it the product decision you cannot settle today.

Install the Skill Star on GitHub