6 files that give AI coding agents a self-directed instruction chain
Most AI coding agents are only as reliable as the prompts you hand them at the start of a session. A piece by Jes Fink-Jensen on the Generative AI publication proposes a more durable approach: a reproducible test built around six files that get an agent to follow a full chain of instructions on its own.
How It Works
The test targets agents that read CLAUDE.md or AGENTS.md on startup, which covers Claude Code, OpenCode, and most other modern coding agents. The idea is that the right file structure can drive sequential, self-directed behavior without a human steering each step.
Fink-Jensen frames it as an ICM test, a reproducible benchmark you can drop into any project to verify that your agent is actually following a chain of instructions rather than improvising its way through a task.
Who This Is For
If you’re using Claude Code or OpenCode for real development work, the six-file structure is worth examining. Agents that read startup config files are increasingly common, and having a repeatable way to verify their instruction-following behavior has practical value for anyone who depends on them for production tasks.
The full walkthrough, including the files themselves, is on Generative AI on Medium.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
