@tank/agent-creator

Multi-agent workflows for code review and security audits are designed by composing individual agent atoms into coordinated bundles, such as reviewer-fixer loops and quality gates, ensuring each agent has a specific identity-first role.

Least-privilege tool scoping involves restricting an agent's available tools to only what is necessary for its specific role, preventing unauthorized actions and ensuring safe operation across tasks like readonly auditing or documentation updates.

Portable agent definitions are created by keeping platform-specific behavior in extensions, allowing the same core agent schema with required fields like name, role, and model to function across Tank-compatible harnesses.

An agent atom should be set to readonly mode when its role involves observation or analysis tasks like code review, auditing, or security scanning, preventing unintended modifications to the codebase during execution.

A Tank agent schema definition requires a compliant name, role, tools, model, readonly setting, and extensions guidance to ensure the agent is properly configured for its designated task and permissions.

Yes, multiple agents can be bundled into coordinated workflows that orchestrate reviewer-fixer loops, quality gates, and security scans, allowing distinct single-agent roles to interact within a controlled pipeline.