harness-agent-permissions

What problem does it solve? When delegating work to AI agents and subagents in OpenCode or Cursor, unclear permissions can lead to agents editing files they should not touch or running unsafe commands. This Skill provides a default permission matrix that maps each agent role to explicit allow, deny, ask, or scoped rules. ## Core Features & Use Cases - Role-based permission matrix: Defines read, edit, bash, task, and external_directory permissions for Planner, Generator, Evaluator, and Security reviewer roles. - Safe defaults: Generators get scoped edit access, evaluators can run targeted checks, and external directory access is denied by default for all roles. - Workflow integration: Designed to be integrated into /harness-standards and /prevc before delegation. - Use Case: Before dispatching a generator subagent to implement a feature, apply the matrix so it can edit only scoped files while bash commands require confirmation. ## Quick Start Apply the harness agent permissions matrix to assign safe role-based permissions to my planner, generator, evaluator, and security reviewer subagents before delegation.