What problem does it solve? Running a full backlog maintenance cycle requires coordinating multiple promotion pipelines (learning, intake, inspect) plus docs inspection and backlog review, each with its own state, failure modes, and resume contracts. Doing this manually risks skipped stages, duplicate execution, and lost progress after interruptions. ## Core Features & Use Cases - Three-stage orchestration: Runs inspect-docs serially first, then executes the three promotion pipelines (learning, intake, inspect) concurrently with serialization of conflicting Git operations, shared artifact writes, and user interactions, then runs backlog-review after a fan-in gate. - Fan-in gating and stop propagation: Starts backlog-review only when all three promotion lines finish normally or with no targets; blocked, failed, or incomplete lines halt downstream stages without cascading stops to independent lines. - State-based resume: Reconstructs progress from durable state under .agentdev/ (inbox files, promoted artifacts, RU files, backlog_auto_started_at) so reruns skip completed stages without relying on conversation context. - Use Case: After accumulating learning entries, intake items, and inspect findings, invoke the backlog-auto command once to promote everything, generate requirement units (RUs), and receive a per-stage report with resume commands if anything stops. ## Quick Start Run the /agentdev/backlog-auto command to execute the full backlog maintenance cycle and review the per-stage results report.
