subagent-delegation

Delegating heavy command workflows to a subagent prevents long outputs from consuming your context window. A two-test gate routes output-intensive tasks out of the parent session when both reasoning consumption and higher-judgment loops are required.

You should delegate Bash commands to a subagent when you are about to run the second or third command toward the same unknown target. This applies to check-suite runs, broad codebase discovery, and multi-step diagnostics where parent context cost is significant.

Route codebase exploration tasks to the right subagent by dispatching discovery objectives to Explore or general-purpose subagents. Use objective-based dispatch prompts with an absolute working directory and a unique dispatch-id to reliably map verdicts.

Dispatch-id-based spool retrieval works by assigning a unique identifier to each subagent invocation. This per-invocation dispatch-id reliably maps blocked-path verdicts and spool output back to the correct delegated command workflow.

Handle hook blocks and settings denials by explicitly distinguishing SETTINGS_DENIAL and HOOK_BLOCK verdicts. This prevents unsafe fallback behavior by ensuring blocked-path handling interrupts the delegated workflow instead of executing inline.

You can use subagent delegation for multi-step test execution by routing check suites to a dedicated check-runner subagent. This prevents output-heavy test runs from damaging the parent context while preserving the higher-judgment loop.