What problem does it solve? Executing a queue of independent coding tasks one at a time is slow, and running them in parallel without structure leaves a pile of unmerged branches and manual cherry-pick cleanup. This Skill drains task files from /tmp/backlog/ by running multiple tasks concurrently in isolated git worktrees, each with an implementer and reviewer subagent, then serially integrates approved branches onto one linear integration branch. ## Core Features & Use Cases - Parallel task execution: Maintains a configurable number of concurrent feature slots (default 3), launching the next pending task the moment any slot frees. - Implementer-reviewer pipeline: Each task gets a fresh implementer subagent, a read-only reviewer subagent, and up to 3 review-fix rounds before being approved or failed. - Serial integration lane: A dedicated integrator subagent cherry-picks approved branches onto a single linear backlog/integration branch, with an escalation path to a repair subagent when mechanical cherry-picks fail. - Use Case: After generating 15 independent task files in /tmp/backlog/, invoke this Skill to grind through all of them in parallel and receive one clean integration branch ready to merge to master, plus a summary of integrated, failed, and blocked tasks. ## Quick Start Run backlog-execute to drain all task files in /tmp/backlog/ with parallel implementer and reviewer subagents and integrate the approved branches onto backlog/integration.
