Task lists as instruction mirrors — Claude's task interpretation reveals prompt quality¶
Insight: Claude Code's task lists serve a dual purpose: tracking progress and revealing how the model interprets instructions. When tasks mirror the user's intentions, instructions are properly understood. When they diverge (out of order, missing steps, extra steps, wrong granularity), it flags communication gaps. The TodoWrite system was replaced by Tasks in v2.1.16 (Jan 2026) adding dependency tracking, file-system persistence, and cross-session collaboration — while Opus 4.5's improved autonomy means simple work no longer needs explicit tracking.
Detail: Common divergence patterns: tasks listed out of order vs. instructions, missing steps (e.g., testing omitted), extra steps (e.g., backup added without instruction), wrong granularity (single instruction expanded to per-file tasks). Using task lists as a benchmarking tool for instruction quality is a meta-technique that improves prompt engineering over time. The new Tasks system provides dependency metadata for complex multi-session projects where Todos were insufficient.