Build less. Measure what matters.

Reference frameworks and practical guides

Use it in any project

• Audit a codebase or architecture with cited evidence.
• Review a pull request for environmental and accessibility risks.
• Choose measurable, reversible improvements without sacrificing resilience.
• Keep official acronyms and references intact.

Rules that adapt to your stack

Green Codex routes universal NR practices and language-specific checks to the technologies it detects. The catalogue covers Java, Rust, C, C++, Go, Python, JavaScript and TypeScript, PHP, Ruby, Scala, Kotlin, Julia, Nim, Zig, Shell, SQL, PL/SQL, HTML5, CSS3, React, Vue, Angular, Svelte, Preact, Astro and Solid.

Each assessed rule returns PASS, FAIL or REVIEW_REQUIRED with code location, impact, severity and evidence.

Read the technical rules

Useful, used and usable

Start with a confirmed user need. Compare useful outcomes with the current approach, observe actual use when the need occurs, and test independent task completion, including keyboard access. Keep three separate decisions; missing evidence stays REVIEW_REQUIRED.

Technical AI changes use relevant efficiency rules. Consult the responsible AI charter for an explicit assessment or an identified risk, without imposing an organisational checklist on every task.

Evaluation reports format and semantic review separately. Automated scans find narrow patterns and cannot establish real adoption, accessibility compliance or environmental savings.

NR charter, architecture and stakeholder practices

The NR charter guide translates five axes into actions for programming languages and natural-language tasks. Additional references cover BP-AIR architecture practices, stakeholder maturity and RIA31 responsible AI questions.

Use only the rules relevant to the task. Missing evidence remains REVIEW_REQUIRED; generated documents cannot prove organisational adoption or replace human decisions. RIA31 is distinct from the EU AI regulation.

Read the source inventory and coverage limits and reuse conditions.

Understand the NR actions taken

The skill requests a closing summary of applied rules, concrete actions, user benefits, evidence and unresolved checks. A simple text task needs only a proportionate explanation.

For repeatable reports, the included command validates a task record and reruns three static checks: SQL projection, media autoplay and JavaScript polling. It cannot verify every rule or establish environmental savings. An optional native hook checks summary structure and can request one correction. It requires explicit installation and trust review in Codex; it does not validate the truth of evidence.

Use the closing summary command

Open and reusable

Code is licensed under Apache-2.0. Original rules and documentation use CC BY 4.0; the BP-AIR, stakeholder maturity and RIA31 adaptations use CC BY-SA 4.0. See the file-specific licence terms. Contributions and translations are welcome through pull requests.

Read the contribution guide · View releases

Maintained by experts

Green Codex is developed and maintained by the Institut du Numérique Responsable (INR) team:

Meet the full team