A Consultant Admitted Claude Does 80% of His Job. I Built That Skill on DeepSeek
Set up the DeepSeek harness from zero, connect it to Codex and Claude Code, then run a Strategy Consultant Skill that does 80% of the job. Full guide inside.
DeepSeek goes viral.
Because it is a lot cheaper yet effective.
Also, it is open source, meaning you can download it and use it for free.
(If you have a powerful computer.)
But there was one problem: they did not have a harness.
Meaning, you can use it like ChatGPT, but not Claude Code.
To learn more about harness, read this.
But recently, the DeepSeek harness landed on GitHub.
And no surprise, it became the fastest-growing GitHub repo, with 120K+ repos in 3 days.
OpenClaw used to hold this record, but now DeepSeek is becoming more popular.
So I installed this harness, so now I can open it with one click.
So I can start building with it immediately.
Plus, I injected DeepSeek inside the Codex app, so now I can use it inside Codex.
And also, I can now use it inside Claude Code too.
I’ll explain everything to you step by step.
But of course, by solving another problem.
I have too many options, but what should I build?
What Should I Build With the DeepSeek Harness?
So I searched for business problems to build something with the DeepSeek harness.
And I came across this Reddit post.
A strategy consultant admits that Claude Code and Cowork scare him.
In the full post, he admits that Claude does everything with 80-85% accuracy.
It’s an honest confession.
That got me thinking: can I turn this into a Claude Skill?
But I also know that not everyone has a Claude subscription, and some of you may find it a little expensive.
So I created the skill to run on the DeepSeek harness.
Let me first show you what I built.
How the Strategy Consultant Skill Works on the DeepSeek Harness
I installed this Claude skill into the DeepSeek harness.
I’ll show you how.
Then I pasted my newsletter statistics and said I wanted to increase them 10x.
Just after I pasted the prompt, look how DeepSeek chose the right skill from the catalog, loaded it, and started working on the task.
This skill also uses harness capabilities like parallel subagents, plan mode, workspace file outputs, and session resume inside DeepSeek Harness.
Next, it asked me 7 different questions:
I answered them one by one, and now 6 subagents are working in the background.
After 40 minutes of dense work, it started writing my growth-strategy.md.
And turned it into a dashboard.
How the Strategy Consultant Skill Works
In short, it has three phases.
In the first phase, this skill turns the DeepSeek into a junior consultant, actually multiple ones.
After collecting your agents, it creates 6 sub-agents. They run on v4-flash, the cheap version.
In the second phase, it stops and shows you a one-page summary.
In the third phase, it applies your corrections and writes the final files to your workspace: growth-strategy.md and growth-model.csv.
We are using harness features the whole way: plan mode for the kickoff, parallel subagents for research, session resume for the checkpoint, workspace files for the deliverable.
This is first issue of AI in Business. If you have a solid use case for AI in business, reach out to me via DM.
After the paywall, we will set up the DeepSeek harness from zero, connect it to Codex and Claude Code, and explore how to install the Strategy Consultant Skill.
We’re building together inside the AI Academy.
To get access, you need to join the Inner Circle, which is currently $200/year.
After September 1, the price will increase to $300/year.















