This skill is used to deliver "already clarified requirements and design constraints" through Gemini CLI as actionable UI/UX solutions and code changes.
1) Gemini CLI must be used (hard requirement)
- When executing UI/UX design and implementation tasks, it must be completed through Gemini CLI; do not fall back to ordinary chat generation.
- If Gemini CLI is unavailable (command does not exist, permissions are restricted, execution fails, or the task cannot be completed), you must immediately stop further implementation and first clearly inform the user of the current blocking reason.
2) Model selection and fallback strategy (must be explicit)
Preferred model chain (from high to low):
proauto
During execution, specify the model explicitly whenever possible, for example:
gemini -m pro -p "<YOUR_PROMPT>"
If the current Gemini CLI version does not support -m or model selection is restricted by platform policy at runtime:
- Continue using the default model to complete the delivery (do not interrupt the process)
- In the final reply, explain: that a model could not be specified this time, the actual usage method (default/platform-managed), and the reason
3) Use local paths to read long context (reduce paste noise)
- When requirement documents, specification documents, or context are long, prioritize providing local file paths in the instructions and let Gemini CLI read the files.
- Do not paste large blocks of document text directly into the conversation.
4) Delivery method: modify code directly, reply with only a summary
- Do not return complete code blocks as chat replies.
- Write the output directly to the target files, then reply only with the necessary summary and results.
5) Final reply format (fixed)
- Keep only the following content:
- List of files written/updated
- Key change points for each file (brief)
- Questions still requiring user confirmation (if any)