adaptive-harness

To optimize agent orchestration, scan your project directory structure to detect the technology stack and dynamically manage agent activation states. This eliminates manual configuration overhead by activating only relevant specialist agents and pruning unused ones.

Failure pattern learning analyzes session history to identify recurring errors in agent execution. It then suggests rule overrides or model escalations to improve reliability and resolve persistent configuration gaps automatically.

To ensure consistent agent behavior, generate portable project-specific profiles within the .claude directory. These profiles can be exported and imported across different environments or team members to maintain identical agent configurations.

Managing agent profiles requires read-only access to project manifest files to detect technology stacks. It also needs write access to the .claude directory to save and maintain project-specific profile configurations.

Prune unused agents when your workspace becomes inefficient or cluttered with irrelevant specialist agents. The system automatically handles this by scanning project files and deactivating agents that do not match your detected tech stack.

You can resolve configuration gaps automatically by operating within the hiddink-harness ecosystem. It analyzes execution patterns and project structures to identify missing settings and suggest appropriate rule overrides.