Harness / SafetyCommunityOpen source
agents-best-practices
Provider-neutral Agent Skill for designing, auditing, and refactoring agent harnesses — installable for both Codex and Claude Code.
Visit sourceKey capabilities
Core principle: the model proposes actions; the harness validates, authorizes, executes, records, and returns observations.
Installs as a skill under ~/.codex/skills or ~/.claude/skills
Best for
Harness design principles
Validation and authorization gates
Project instructions
Not for
Stack-specific assumptions without adaptation
Risk notes
Adapt the rules to your own stack and permissions.
This is a community project. Verify maturity, licensing, permissions, and safety behavior before relying on it in production.