What problem does it solve? Large, high-risk coding efforts—cross-cutting refactors, unsafe-to-partially-ship migrations, multi-phase architectural changes—often fail because a single agent tries to implement everything without structure, review, or persistent state. Deepwork turns the AI into a scheduler that plans phases, delegates to specialists, and enforces review gates before work proceeds. ## Core Features & Use Cases - Persistent Session State: Maintains a task-specific markdown progress file under .slim/deepwork/ capturing goals, research, plans, validation results, and blockers across the session. - Phased Delegation with Oracle Gates: Splits work into coherent delivery phases, delegates to specialist agents (oracle, librarian, explorer, designer, fixer), and requires an Oracle review after each phase with a bounded re-review budget. - Scheduler Discipline: Tracks task/session IDs, waits for background job completion before consuming results, and blocks phase advancement until terminal results are reconciled. - Use Case: Migrating a monolith's authentication layer across multiple services—plan the phases, delegate each to specialists, gate each with an Oracle review, and commit only at validated delivery boundaries. ## Quick Start Ask the AI to use the deepwork workflow to plan and execute a large multi-phase refactor with specialist delegation and review gates after each phase.
