implement-next-tasklisted

# Implement Next Task Select and implement one eligible task in full on its feature branch, then stop and hand off to the manual review loop. Use one implementation workflow at a time in the current checkout. The master plan controls order and status; the selected task file is the contract for what to build. This skill covers selection, setup, and implementation. It does not review, update the README, record final completion, or open a PR. When it ends, tell the user to run: 1. `/skill:task-review` — writes a findings file under `reviews/`; 2. `/skill:review-fix-worker` — fixes the findings one by one; 3. repeat 1–2 until the review is clean or only accepted/skipped findings remain; 4. `/skill:finish-task` — README, final proof, completion record, and PR. Without explicit user approval, modify only repository-scoped files, isolated local/test databases, and the task log directory defined below; never run destructive shell commands, access production systems or data, read `.env` files, or delete or move anything else outside the repository root. ## Efficient evidence handling Keep the main context for decisions, behavioral TDD evidence, targeted source inspection, and final integration proof. Delegate broad, read-only interpretation to isolated subagents when available and require concise structured results. Do not delegate primary RED-GREEN decisions, concurrent edits, or stateful checks that share a database, port, simulator, fixture, generated output, or external serv