Prime Agent Coding Skill

A reusable Agent Skill for Prime Agent.

Scope

This skill gives host agents and in-session Prime Agent users one secure process for bounded coding and repository-research work.

It covers:

• permission-bounded short and multi-step tasks;
• practical trust boundaries for repository content;
• safe CLI defaults that disable unreviewed workspace resources;
• read-only child-agent reviews and worktree isolation;
• command review, diff inspection, and evidence-based completion; and
• explicit stop conditions for external or destructive actions.

Runtime installation, package management, uploads, telemetry, releases, and deployments are intentionally outside the skill.

Local Check

Run this command from the repository root:

npx skills add ./ --list

The command must find one skill named prime-agent.

Install

Install the current public release with:

npx skills add oddradiocircle/prime-agent --skill prime-agent

Author and Attribution

Author: Daniel Gómez.

This skill is an independent guide. It is not an official Prime Intellect product. Prime Agent is developed by Prime Intellect.

License

MIT. See LICENSE.