Maoxuan Product Agent

Maoxuan Product Agent is a Chinese-first decision skill for product managers, product operations, growth teams, founders, and delivery leads working in mainland China internet-product contexts. It converts a real work problem into a prioritized judgment instead of producing a long list of equally weighted tactics.

The reasoning engine was designed from complete readings of On Contradiction, On Practice, and related Volume I essays. The default user experience does not quote those texts, teach philosophy, discuss history, or perform political role-play. Source reasoning is translated into modern product language such as bottlenecks, constraints, product stage, MVP, gray release, A/B testing, evidence quality, and strategy updates.

Knowledge Freshness

This listing was verified on 2026-07-10 against release v1.0.3, the public skill package, reasoning reference, 36-scenario evaluation report, source-reading audit, case library, installation instructions, and independent Skillstore listing.

The repository is actively versioned. Re-check the latest release and product-decision-agent/SKILL.md before distributing a pinned copy inside a team.

Retrieval Sources

• https://github.com/atdy/maoxuan-product-agent
• https://github.com/atdy/maoxuan-product-agent/blob/main/product-decision-agent/SKILL.md
• https://github.com/atdy/maoxuan-product-agent/blob/main/product-decision-agent/references/reasoning-engine.md
• https://github.com/atdy/maoxuan-product-agent/blob/main/product-decision-agent/references/product-playbooks.md
• https://github.com/atdy/maoxuan-product-agent/blob/main/evaluation/source_reading_audit.md
• https://github.com/atdy/maoxuan-product-agent/blob/main/evaluation/self_test_report.md
• https://atdy.github.io/maoxuan-product-agent/cases/
• https://skillstore.io/skills/atdy-product-decision-agent

Core Workflow

1. Separate observed facts, user behavior, and business data from assumptions or second-hand opinions.
2. Identify the current bottleneck whose removal is most likely to change the outcome.
3. Diagnose the product or business stage, the mechanism currently dominating the result, and the constraints that limit available actions.
4. Choose one to three minimal, reversible actions with an owner, time window, metric, or decision rule.
5. State what not to do yet, then use actual results to continue, stop, roll back, or update the strategy.

When decisive facts are missing, the skill asks no more than three questions and explains why each answer can change the decision. If waiting is more expensive than acting, it provides a provisional action and names the assumption being tested.

Capability Scope

Installation

Install the skill globally for Codex, Claude Code, and Cursor:

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

The repository also documents a shell installer, manual installation paths, and a standalone release archive. Review the source and destination directory before using an installer on a managed or shared workstation.

Usage

Ask a real work question in Simplified Chinese or explicitly invoke the skill:

/product-decision-agent We have 20 requests competing for two engineers. What should the next version include?

Typical outputs contain:

1. A one-sentence problem judgment.
2. The mechanism and evidence behind that judgment.
3. One to three ordered actions with an owner, time window, metric, or decision rule.
4. A risk warning and stop list.
5. Up to three missing facts only when they can change the decision.

Evaluation Evidence

The public repository includes:

• A 36-scenario regression matrix covering prioritization, growth, retention, operations, metrics, experiments, delays, executive requests, and team conflicts.
• Four independent forward sessions that test behavior outside the original authoring conversation.
• A deliberately poor answer that must fail the quality gate.
• Python tests for source-language leakage, vague advice, missing decisions, missing metrics, missing stop guidance, and overly thin output.
• Twelve standalone decision cases with visible judgments, actions, stop lists, and switching conditions.
• Continuous validation through GitHub Actions and an MIT license.

Production Rules

• Treat model output as a decision proposal, not an accountable decision.
• Verify the evidence and metric definition before acting on an anomaly.
• Use a reversible test when the underlying mechanism is still uncertain.
• Keep product strategy, customer identities, incidents, and internal metrics within an approved model and retention boundary.
• Preserve the stop list and switching condition when turning an answer into a project plan.
• Do not ask the skill to manufacture certainty from missing business context.

Troubleshooting

Issue: The installed skill does not trigger automatically

Fix: invoke /product-decision-agent explicitly and provide a concrete product, growth, operations, data, or collaboration question.

Issue: The answer quotes source material or explains theory

Fix: confirm the current product-decision-agent/SKILL.md is installed and ask for a product decision rather than source-text interpretation.

Issue: npx is unavailable or user-level installation is not allowed

Fix: use the documented manual installation path or extract the standalone release archive into a project-scoped Skills directory after reviewing it.

Duplicate Review

The directory already contains product-management-ai-agent in the agents category. That entry is a generic code-heavy template for user stories, analytics, roadmap prioritization, and A/B testing. Maoxuan Product Agent is an installable SKILL.md package focused on Chinese product-work diagnosis, current-bottleneck selection, stage-aware tradeoffs, minimal validation, and explicit stop and switching conditions. A search of content/skills found no existing Chinese product-decision skill or equivalent methodology-backed diagnosis workflow.

Editorial Disclosure

Submitted by the project owner as a source-backed community skill. The project is MIT-licensed and free to use. It has no hosted service, paid dependency, telemetry, or affiliate link. Its source methodology is visible for audit, but default outputs intentionally use modern product language rather than source quotations or political framing.

• Source repositorygithub.com 2026-07-10
• Documentationgithub.com
• Websiteatdy.github.io
• Package downloadgithub.com
• Submitted by atdy2026-07-10