profit-atlas

Profit Atlas

Find where an industry or company actually makes money — and show the evidence.

Profit Atlas is a small, evidence-first collection of Agent Skills for research that starts in one of two places:

The skills compose, but remain separate: industry profit distribution and company economics need different units of comparison, evidence, and outputs.

中文说明

See the work, not just the prompt

Both reports distinguish disclosed facts, calculations, and inference; link primary sources; and state the limits of public data.

Quick start

This repository is a portable Agent Plugin: its root plugin.json identifies the package and each reusable workflow lives under skills/. The layout follows the open Agent Skills convention.

1. Add the public marketplace: codex plugin marketplace add Cheerfay/profit-atlas.
2. Install the plugin: codex plugin add profit-atlas@profit-atlas.
3. Start a new chat and ask a natural-language research question. The installed plugin exposes both Skills for automatic routing between industry-first and company-first questions.

If your host does not support marketplaces, clone or download this repository and copy either required folder from skills/ into that host's configured Skills directory.

For current Codex desktop local-plugin installation, use an official local marketplace and restart the app after installing. The package layout and both Skill manifests were validated in a clean copied-folder check; full automatic activation has not yet been tested across every Agent Skills host.

Ask naturally

The descriptions are designed for automatic routing. For example:

• “Map the profit pools in China's game industry and identify defensible startup entry points.”
• “Analyze what BYD fundamentally lives on and how value becomes cash.”
• “Compare Tencent, NetEase, and 37 Interactive: how do their business models and profit engines differ?”

To force a route, name the Skill:

• “Use industry-profit-pool-research to study the automotive value chain.”
• “Use company-value-chain-review to review Meituan.”

How the two Skills work together

An industry study selects representative companies by role or business model to support a profit-pool conclusion. It does not replace a full company review.

When a company study needs industry context, run the industry Skill first and then use the company Skill to synthesize two layers:

1. Company economics: payer, reason to pay, revenue, core profit, costs, and risks.
2. Value-to-cash workflow: the few operating steps that determine revenue, cost, risk, and trust.

What the benchmark does—and does not—show

Benchmark v0.1 preserves two raw, fresh-context no-Skill baselines alongside the Skill-guided cases. It evaluates explicit value-chain coverage, source traceability, evidence labels, calculation consistency, and conclusion boundaries.

It is deliberately a two-prompt, author-run process evaluation—not a model leaderboard or a claim that unscripted research cannot be good. The observed advantage is more disciplined structure and evidence boundaries.

Evidence boundaries

• Company revenue is not automatically comparable with industry sales.
• Revenue share does not prove profit contribution.
• Gross profit, operating profit, net profit, and cash flow are kept distinct.
• Calculations and inference are labeled; undisclosed segment profit is not invented.
• Startup opportunities are tested for external pain, budget, entry rights, defensibility, and economics. AI may be a core capability, an enhancement, an ordinary feature, or unnecessary—it is not the default answer.
• Dedicated sector modules improve question coverage but are not required. An unlisted sector uses the general workflow and states missing evidence explicitly.

Profit Atlas is a research framework, not investment advice. Independently verify evidence before making financial or operating decisions.

Repository map

skills/       two installable research workflows
examples/     complete, sourced research outputs
evals/        raw baselines and evaluation rubric
assets/       social-preview artwork
plugin.json   portable plugin manifest

Contributing

Contributions should preserve the boundary between the two Skills and add an evaluation case for changed behavior. Add a sector module only after repeated real tests reveal a stable, sector-specific question the general workflow misses.

Before submitting a change, validate the Skill front matter, check every referenced resource, rerun relevant evaluation cases, and scan for private paths or credentials.

License

MIT