Tell the agent when to ask.

for skill in jev jev-task-router jev-model-router \
  jev-tool-guard jev-research-guard jev-completion-review; do
  mkdir -p ".agents/skills/$skill"
  curl -fsSL "https://jevai.ai/skills/$skill/SKILL.md" \
    -o ".agents/skills/$skill/SKILL.md"
done

When to install which

You do not need the whole pack in every repo. Install the ones that match a branch the agent already has.

• jev — the agent sometimes needs a decision that is not one of the presets. It should build a choice, score, or noul instead of asking the main model for JSON.
• jev-tool-guard — anything that sends, deletes, pays, writes to a customer, or changes production. Put it immediately before the tool runs.
• jev-task-router — the agent is about to choose between a fast edit and a long investigation.
• jev-model-router — you have two or more models configured and the choice should depend on the task, not on a fixed default.
• jev-research-guard — the agent cites a passage, a doc, or a tool result and is about to treat it as proof.
• jev-completion-review — the agent is about to tell a person the task is finished.

What a skill is not

A skill does not call Jev by itself. It does not widen the agent’s permissions. It does not hold an API key. If a skill file in a repo contains a key, treat that key as public and rotate it. The running agent reads JEV_API_KEY or TYPESAFE_API_KEY from the environment and passes it as a Bearer token to the MCP server or to TypeSafe directly.

The official TypeSafe skill (npx skills add typesafe-ai/skills --skill typesafe-ai) is a different file. It teaches a coding agent how to write a correct systemone integration. Use it while you are building. Use the files on this page while an agent is deciding.