deepwork

What problem does it solve? Applying Deepwork's autonomous multi-agent workflow inside Codex requires manually mapping roles, models, reasoning efforts, and dispatch tools, which is error-prone and inconsistent across sessions. ## Core Features & Use Cases - Request Classification and Role Routing: Classifies incoming requests into categories like quick, coding, deep, research, or frontend and selects the matching Deepwork role or generated dw-* Codex agent profile. - Callable Dispatch Contract: Routes delegation through exact profile selection, direct model composition, generic/flat dispatch, or local execution based on what the current callable tool schema actually exposes. - Tiered Model and Review Selection: Assigns flagship, mid, and mini model lanes with reasoning-effort floors, plus ordered Reviewer and Oracle cross-check profiles for implementation acceptance. - Use Case: When asked to plan and implement a cross-module feature, the skill classifies the work as deep, dispatches dw-planner and dw-builder profiles at xhigh effort, and runs dw-reviewer plus dw-oracle cross-checks on the resulting diff before reporting completion. ## Quick Start Ask the agent to plan and implement a feature using deepwork-style multi-agent execution with plan review and code verification.