creating-opencode-agents

This skill provides a comprehensive framework for developing specialized OpenCode AI agents, ensuring they are configured with correct metadata, operational modes, and security-conscious tool permissions. It is intended for software developers and AI engineers who need to build domain-specific assistants that integrate seamlessly into the OpenCode ecosystem. By utilizing standardized YAML frontmatter in agent configuration files located at .opencode/agent/, users can define consistent behavior, model overrides, and temperature settings tailored to specific development tasks like code review, testing, or backend architecture.

• Define agent behavior using required description fields and flexible configuration parameters such as mode (primary, subagent, or all) and custom system prompts.
• Implement granular tool access control with permission configurations, allowing fine-tuned management of read, write, edit, bash, grep, glob, and web-based tool usage.
• Utilize robust security patterns by setting specific shell command permissions (allow, deny, or ask) to prevent unauthorized file modifications or sensitive command execution.
• Follow best practices for common agent patterns, including dedicated configurations for code reviewers, backend developers, test writers, and documentation specialists.
• Integrate with the PRPM ecosystem to search, install, and publish agent templates, ensuring compatibility across various AI-assisted coding environments.
• Optimize agent performance by selecting appropriate temperature settings, ranging from low randomness for analytical tasks like debugging to higher values for creative brainstorming or documentation.
• Manage agent lifecycle and discoverability through standard file structures and manifest-driven package management, facilitating easier deployment and sharing of specialized assistant logic.
• Adhere to safety guidelines by restricting agent scopes through wildcard tool disabling and environment-aware permission logic, protecting sensitive files like .env or private keys.
• Leverage the interaction between subagents and primary assistants to build multi-agent hierarchies for complex development workflows, enhancing modularity and task-specific expertise.