What problem does it solve? When facing multiple independent failures, investigations, or edits, working them sequentially wastes time, while careless parallelization causes conflicting edits. This Skill defines when and how to safely fan out work to concurrent agents while the main session coordinates integration and verification. ## Core Features & Use Cases - Independence Criteria: Checklists for deciding when tasks qualify for parallel dispatch (disjoint write scopes, no shared state, no sequential dependencies) and when to avoid it. - Prompt Shape Template: A structured agent prompt format covering problem domain, owned files, forbidden files, and expected report format. - Runtime Mapping: Guidance for Claude Code (multiple Agent calls in one message) and Codex (spawn_agent or sequential fallback). - Use Case: Three unrelated test files fail with different root causes; dispatch three agents, each owning one test file, then review summaries and run the combined test suite after integration. ## Quick Start Ask the agent to split these independent failing tests across parallel workers with disjoint file ownership and then verify the combined result. Note: this Skill is deprecated and superseded by workbench:subagent-driven-development.
