skills

Common agent skills used within atvari, published to the open skills ecosystem.

Skills

Skill	Description
atvari-copyright	Copyright and license notices for atvari GmbH projects: LICENSE files, README sections, and package/site metadata, detecting atvari repos via the @atvari.eu git email.
auto-system-one	Teaches agents to reach for jev decisions (Jev-compatible decision endpoints; default OpenCode Console System One) proactively — screen before reading, verify claims, rank by meaning, gate choices, score rubrics — with skip rules and a hard privacy line.
ci-agents	Run Claude Code and OpenCode agents in CI pipelines authenticated with flat-rate subscriptions instead of pay-per-token API keys: workflow YAML for GitHub Actions, GitLab CI, and Forgejo Actions, secrets wiring, and comment-triggered bots.
conventional-commits	Default commit messages to Conventional Commits (type(scope): description), including breaking-change footers and repo-precedent awareness.
system-one	How to call a Jev decision endpoint from the terminal — default OpenCode Console System One, others via env overrides — with request shapes for noul / choice / score, the jev CLI, thresholds, and privacy guardrails.

On OpenCode, use the opencode-system-one plugin instead of auto-system-one and system-one: it enforces the same judgment layer natively (jev_decide / jev_judge_many tools, screening, scrubbing, routing). The two skills remain for agents without that plugin.

Installation

Install individual skills via the skills CLI:

$ skills add atvari-eu/skills@<skill-name> -g

Or via home-manager/NixOS:

programs.opencode.skills."<skill-name>" = "${inputs.skills}/skills/<skill-name>";

Layout

• skills/<name>/SKILL.md — one skill per directory
• skills/<name>/evals/evals.json — eval cases with fixtures in evals/files/

Skill directory paths are load-bearing — don't rename or restructure existing skill dirs without intent.

License

MIT