Zephyr Agent Skills

A registry of professional, agent-ready skills for building with Zephyr RTOS.

This repository is a curated collection of "skills" — modular packages of knowledge, code patterns, and best practices that help AI agents and human developers build high-quality embedded systems with Zephyr. Each skill is a SKILL.md plus supporting references/, scripts/, and assets/.

Using the skills

With zephyr-cli (recommended)

zephyr-cli is the primary way agents consume this registry. It selects the right skill deterministically — no guessing:

zephyr-cli skills suggest "enable an i2c sensor in my devicetree overlay"
zephyr-cli skills install devicetree

skills suggest scores your task against every skill's keywords, aliases, Kconfig patterns, and devicetree compatibles and returns a ranked list with the reasons each matched. skills install fetches the chosen skill into your workspace.

As a Claude Code plugin

The repository can also be added as a Claude Code plugin marketplace:

claude plugin marketplace add beriberikix/zephyr-agent-skills
claude plugin install zephyr-skills@zephyr-agent-skills

The umbrella zephyr-skills skill (the repo-root SKILL.md) is a broad entry point that routes agents to zephyr-cli skills suggest for deterministic skill selection.

The registry — index.json

index.json is the machine-readable registry consumed by zephyr-cli. It is a generated artifact — do not edit it by hand. It is built from each skill's SKILL.md frontmatter plus a per-skill skill-meta.yaml sidecar:

python scripts/generate_index.py

skill-meta.yaml holds the curated matcher metadata — summary, keywords, aliases, kconfig_patterns, dts_compatible, and weight. scripts/validate_skills.py (run in CI) keeps index.json, .claude-plugin/marketplace.json, and the skill set consistent.

Skill catalog

👉 [Master Skill Catalog](skills/zephyr-index/references/skill_catalog.md) — every skill, grouped by domain (foundations & build, hardware & peripherals, connectivity, production & advanced). Or just run zephyr-cli skills suggest and let the matcher pick.

Repository structure

.
├── skills/                       # the skills — one directory each
│   └── <skill>/
│       ├── SKILL.md              # entry point (Quick Start + Validation Checklist)
│       ├── skill-meta.yaml       # curated matcher metadata for the registry
│       ├── references/           # detailed technical guides
│       └── scripts/, assets/     # helper files (where applicable)
├── index.json                    # generated machine-readable registry (consumed by zephyr-cli)
├── SKILL.md                       # umbrella skill — Claude Code entry point
├── scripts/
│   ├── generate_index.py         # regenerates index.json
│   ├── generate_marketplace.py   # regenerates .claude-plugin/marketplace.json
│   └── validate_skills.py        # repository quality gate
└── .claude-plugin/marketplace.json

Contributing

Each skill is a self-contained directory: a SKILL.md (with ## Quick Start and ## Validation Checklist sections), a skill-meta.yaml, and optional references/, scripts/, and assets/.

To add or change a skill:

1. Edit or create the skill under skills/<name>/.
2. Update its skill-meta.yaml matcher metadata (keywords/aliases quality directly

drives zephyr-cli skills suggest accuracy).

1. Run python scripts/generate_index.py and python scripts/generate_marketplace.py.
2. Run python scripts/validate_skills.py — it must pass.

Commit the regenerated index.json and marketplace.json; CI verifies they are in sync with skills/.