Prompts are available in English and Simplified Chinese. Showing English.
Make an old prompt easier to execute
Remove ambiguity and define defaults and a stopping point.References & what we used
- Evaluation best practices
Uses explicit criteria and representative cases; this is a static walkthrough, not an executed comparison or performance evaluation.
Choose an earlier task prompt from this conversation, excluding this rewrite request. Prefer one that caused repeated clarification or rework; otherwise use the latest clear task. If none exists, report that without asking for an ideal answer. Preserve its objective. Specify where to find inputs, how to choose a default scope, what to deliver and when to stop. Do not turn inferable context into blank fields. Keep consequential unknown facts or product decisions unresolved instead of inventing them. Deliver the main issues, the full rewritten prompt and three static cases: ordinary input, missing materials and oversized scope. Walk through the expected handling without claiming independent model testing or measured gains. The rewrite must be usable without a preliminary clarification round. Execution contract: Start directly. Do not ask me to approve a plan, choose files or fill in a template. Use only this project, this conversation and materials already provided; do not search unrelated projects or private history. Complete what the evidence supports and report missing inputs. If the task is inapplicable, explain why and stop without inventing a result. Preserve existing files and uncommitted changes. Include complete proposed content in your reply; standalone artifacts may be saved under a new, unused filename. Do not install dependencies, change the environment, commit, push or deploy. Respect access and permission limits and report blocked work. Check your result once, deliver it and stop without asking whether to continue.