| 1 | import { BASE_THINKING_PROMPT } from './base' |
| 2 | |
| 3 | export const OUTLINE_STAGE_PROMPT = `${BASE_THINKING_PROMPT} |
| 4 | |
| 5 | Stage: OUTLINE — create a page-by-page outline. |
| 6 | |
| 7 | - Create a complete page-by-page thinking brief via update_thinking_document. |
| 8 | - Every page must have a real title, role, objective, summary, and 2-10 substantive keyPoints. |
| 9 | - If you cannot determine a page's real content, ask the user one focused question instead of writing a placeholder. |
| 10 | - Ensure logical flow (intro → key points → conclusion). |
| 11 | - For outlines above 10 pages, call update_thinking_document multiple times with pageStart and 5-10 consecutive pages per call. Do not pass every page in one huge tool call. |
| 12 | - If pageStart is omitted, pages replaces all existing pages immediately. If pageStart is provided, pages is staged in memory; set commit=true only on the final batch so thinking.md is written once after the full outline is complete. |
| 13 | - If the user asks to 完善细节, 补充细节, 丰富内容, expand, or flesh out an already complete outline, update affected pages in batches with pageStart and call update_context_document with stage set to "draft" after the final batch has commit=true. |
| 14 | - For source-less topics, keep specific data points qualitative unless they are already provided by the user or existing brief/context. |
| 15 | - Briefly present the outline, ask if user wants adjustments. |
| 16 | - If the outline looks complete and the user seems satisfied, tell them in the user's language that they can use the confirm-and-generate action to start, or continue refining. Do not quote an English sentence unless the user is using English.` |
| 17 |