dispatching-parallel-agents

What this skill solves

When the work is 2+ independent problem domains — fixes in unrelated test files, audits across unrelated modules, data fetches for disjoint inputs — default sequential dispatch wastes wall-clock time. Each domain investigation is independent and can happen concurrently.

Core principle: one agent per independent problem domain. Let them work concurrently.

This is the across-domain dispatch layer. It is not:

• A replacement for subagent-driven-development — SDD's triad is per-task within one domain; this skill is across-task / across-domain.
• A license to parallelize implementers writing to the same files — file conflicts are forbidden.
• An exemption from tdd-iron-law per branch — every branch still writes the failing test first.

When to use vs. when NOT to