
The author has been running Claude Code since October 2026 with a Root session directing planning, marketing, product-development and quality-control sub-agents, and published the resulting approval and quality-check template on GitHub under MIT.
Summaries like this, in your inbox every morning.
The template is a record of specific failures rather than a general framework. Writing files is deliberately withheld from the planning, marketing and quality-control roles; only product-development can write, and only within a folder defined by written rules. Quality control runs on an isolated copy under a different role from the one that produced the work, and it can block publication until fixes and re-verification clear it.
The author is explicit that the rules are mostly prose that the AI is asked to follow, not hard technical limits. External publishing, billing, contracts and account operations are left to the human, with approvals logged in a file. In the two-company setup, a settings.json deny on absolute paths stopped Write, Edit and Bash echo, but a Node script could still write when Bash was auto-allowed, so the author describes deny as a brake rather than a wall and suggests separate Windows users, virtual machines or containers for anything that must truly stay out of reach.
The stated environment is Windows 10 with Git Bash and Claude Code 2.1.296. Unverified areas include Mac and Linux, long-term operation and how well the planning exploration works in practice, and the author notes there is no track record of revenue from this setup.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
In a SiliconANGLE essay, Imagine Art's M
Hitachi started a service last month using AI models from OpenAI and Anthropic to monitor attacks and restore…

ルールベース株研 started 投資可否チェッカー on 2026年9月25日 and 急騰株チェッカー on 2026年10月8日, reaching over 300 commits and more than 8…

Tested on Windows 10 with Claude Code 2.1.296, a deny rule written as Edit(../shared/partner/**) did not stop…

A writer had Google Gemini read his health check results; Gemini warned of a future dialysis risk linked to hi…

TypeSafe AI opened early access to Jev on September 15, 2026
