| name | subagent-task |
| description | Guidance and protocols for spawning sub-agents via the task tool to handle complex, multi-step, or parallelizable tasks. You MUST load this skill when the task tool for invoking sub-agents is available. |
| license | MIT |
subagent-task
Provides policies and examples for using the task tool to spawn sub-agents for specialized, parallel, or complex tasks.
Core Principles
- Specialization: Delegate tasks to specialized project agents rather than attempting to handle broad contexts monolithically.
- Parallelization: Spawn multiple agents concurrently for independent sub-tasks (e.g., retrieving facts, analyzing logs, validating plans).
- Context Encapsulation: Spawning sub-agents prevents the primary agent's context from becoming cluttered. Synthesize results from sub-agents into the final response.
SUBAGENT DELEGATION POLICY
The use of the task tool and spawning sub-agents is permitted for complex, multi-step tasks, but delegation is limited to the project agents explicitly configured by this runtime.
- Allowed Delegation Targets: Use only the named project agents exposed by this runtime configuration.
- Built-in Subagents Disabled: Built-in
explore and general subagents are not approved for this runtime and MUST NOT be used, even if a host tool still lists them.
- Maintain Context: Ensure that the primary agent remains the coordinator and synthesizes the results from sub-agents into the final response.
- Strategic Delegation: Delegate only when the task involves broad codebase analysis or independent sub-tasks that can be executed in parallel.
