Backlog Manager

Manage an engineering backlog for humans and AI agents.

Think of this skill as a lightweight product manager for the backlog. The goal is to keep engineering work clear, sequenced, classified, and safe to route to either humans or AI agents. A core job is to label safe, well-scoped tickets so AI agents know which work they are allowed to pick up, while marking ambiguous, risky, or judgement-heavy work for humans. This is backlog review and project-state hygiene, not implementation. It labels issues, improves issue quality, identifies missing follow-up tickets, creates evidence-backed maintenance tickets when allowed, and syncs issue state with linked pull requests.

Default to dry-run. Only mutate GitHub, Linear, or another tracker when the user explicitly asks for apply.

When running unattended (scheduled or non-interactive), never wait for user input. Record any blocker or missing setup in the final report and exit.