task

What problem does it solve? Running a backlog task with an AI agent usually means untracked token spend, no isolation between concurrent work, and no record of which model produced what. This Skill runs one task end-to-end as a tracked unit of work: it derives a task ID, creates an isolated git worktree, delegates planning and review to high-reasoning models, routes implementation to persona-mapped agents, and records token usage per dispatch. ## Core Features & Use Cases - Token-tracked task lifecycle: task_tracker.py registers start/finish, records per-subagent token usage, and reports model mix so delegation decisions become measurable. - Effort-tiered loops: low-effort and high-effort pipelines, where high-effort adds MoE (mixture-of-experts) review panels, a mandatory integration package, and a QA test-quality gate. - Worktree isolation and recovery: every task runs in its own worktree with a recorded resume command, so a dead session can be reattached and continued. - Use Case: A developer says "/task PROJ-123" — the skill creates a worktree and draft PR, delegates the plan to an architect persona, dispatches TDD implementation lanes, runs the quality gate, merges, and closes tracking with a token-cost report. ## Quick Start Ask the agent to start working on a backlog task by invoking the task skill with a task ID, for example "/task aib-add-slugify-helper", and choose low or high effort when prompted.