🏅 ACMM badge — kubebuilder at 16 criteria, one CLAUDE.md away from a strong L2

[clubanderson]

The KubeStellar Console includes an AI Capability Maturity Model (ACMM) dashboard for measuring AI-assisted development maturity.

We scanned kubernetes-sigs/kubebuilder and found 16 criteria detected, including:

✅ Full prerequisite suite, AGENTS.md, custom skills, CI matrix, test coverage metrics, CI/CD maturity, and production feedback signals

📊 View your full ACMM score →

Three additions that would close L2 Instructed gaps:

1. CLAUDE.md — project instructions for Claude Code and other agents
2. .github/copilot-instructions.md — Copilot workspace instructions
3. .editorconfig — cross-editor consistency (ACMM L2 signal)

CLAUDE.md in particular lets agent contributors (Claude Code, Copilot, Codex) read your conventions automatically — especially valuable for a scaffolding project where generated code patterns matter.

Add the badge to README.md:

[![ACMM](https://img.shields.io/endpoint?url=https%3A%2F%2Fconsole.kubestellar.io%2Fapi%2Facmm%2Fbadge%3Frepo%3Dkubernetes-sigs%252Fkubebuilder%26v%3D3)](https://console.kubestellar.io/acmm?repo=kubernetes-sigs/kubebuilder)

Paper: https://arxiv.org/abs/2604.09388

[camilamacedo86]

Hi @clubanderson,

Thank you so much for raising this one.
I am unable to check our score.

Also, we do not want to add CLAUDE.md because AGENTS.md has the same but is not vendor specific.
Same for .github/copilot-instructions.md

See; https://agents.md/

[clubanderson]

Hi — thank you for the quick reply!

Good news: AGENTS.md is already detected ✅ — the ACMM dashboard picks up AGENTS.md, CLAUDE.md, .github/copilot-instructions.md, and .cursorrules as independent signals, so you already get credit for the vendor-neutral approach. We fully agree that AGENTS.md is the right choice over vendor-specific files.

Re: unable to check the score — the dashboard is at https://console.kubestellar.io/acmm?repo=kubernetes-sigs/kubebuilder (no login required). If something didn't load, it may have been a cold start. Happy to investigate if you hit a specific error.

The two remaining L2 gaps (after AGENTS.md ✅) are just .editorconfig and a prompts/ directory — both optional of course. The badge works as-is once added to README.md.

[camilamacedo86]

The ACMM dashboard picks up AGENTS.md, CLAUDE.md, .github/copilot-instructions.md, and .cursorrules as independent signals, so you already get credit for the vendor-neutral approach. We fully agree that AGENTS.md is the right choice over vendor-specific files.

I don’t understand why we would need to add CLAUDE.md, .github/copilot-instructions.md, and .cursorrules if we already have AGENTS.md.

If the intention is to support a vendor-neutral approach, the check should be smart enough to recognize the presence of AGENTS.md as sufficient. Requiring additional vendor-specific files seems redundant and contradicts the goal of being vendor-agnostic.

In its current form, the rule and evaluation criteria don’t seem accurate, as they appear to favor tool-specific configurations rather than validating a truly neutral AI setup.

---

Re: unable to check the score — the dashboard is at https://console.kubestellar.io/acmm?repo=kubernetes-sigs/kubebuilder (no login required). If something didn't load, it may have been a cold start. Happy to investigate if you hit a specific error.

Got it — I was able to access it now 🙂

---

The two remaining L2 gaps (after AGENTS.md ✅) are just .editorconfig and a prompts/ directory — both optional of course. The badge works as-is once added to README.md.

I’ll take a look at those, thanks.

---

However, one more piece of feedback:

This reinforces the concern — the check is explicitly looking for vendor-specific configuration (e.g., Claude), even when the project follows a vendor-neutral approach like:

https://agentskills.io/home

If the goal is to evaluate whether a project is properly configured for AI usage, the checks should be vendor-agnostic. Otherwise, it ends up verifying whether the project is configured for multiple specific tools/vendors, rather than assessing the overall quality and neutrality of the AI setup.

[clubanderson]

Thanks @camilamacedo86 — this is a genuinely valid design critique and I appreciate you taking the time to articulate it clearly.

You're right. The current scoring treats AGENTS.md, CLAUDE.md, .github/copilot-instructions.md, and .cursorrules as four separate L2 criteria. A project that deliberately chose the vendor-agnostic AGENTS.md convention gets only 1/4 credit for the group, while a project stacking multiple vendor-specific files scores higher even though the underlying capability is identical: the AI assistant has project-specific instructions.

The design intent was that breadth of tool coverage signals investment — if you've configured Claude AND Copilot, more of your team is using AI tooling. But for the binary question "is this project AI-ready?" your framing is more accurate: any one agent instruction file should satisfy the L2 'Instructed' requirement.

I'm filing an issue on the console repo today to track the fix:

• Group the four agent-instruction-file criteria into an OR group: any one of AGENTS.md / CLAUDE.md / copilot-instructions / .cursorrules satisfies the 'agent instructions configured' signal
• Keep detecting all four individually so the dashboard can still show which tools the project uses
• Only the level gate changes — presence of AGENTS.md alone → L2 ✅

This also aligns with what agents.md was designed to be: the vendor-neutral canonical location. Thanks again for the sharp feedback — this makes ACMM a better tool for the whole ecosystem.

[clubanderson]

Quick update — the fix is now live. kubebuilder now shows L2 · Instructed on the ACMM dashboard. You can verify at:

https://console.kubestellar.io/acmm?repo=kubernetes-sigs%2Fkubebuilder

The badge also reflects this if you want to add it to your README:

[![ACMM](https://img.shields.io/endpoint?url=https%3A%2F%2Fconsole.kubestellar.io%2Fapi%2Facmm%2Fbadge%3Frepo%3Dkubernetes-sigs%252Fkubebuilder)](https://console.kubestellar.io/acmm?repo=kubernetes-sigs%2Fkubebuilder)

Thanks again for the sharp feedback — AGENTS.md is now treated as a first-class instruction file equivalent to CLAUDE.md.