research-insight

name	research-insight
description	Research trend analysis and insight generation. Triggers on "趋势", "trend", "这个方向火不火", "research insight", "方向分析".

Research Insight

Flow

1. Context carry-over:

   • If user explicitly references existing papers ("根据刚才的", "用这些做趋势分析"): use those papers directly as landscape
   • If papers in context on the same topic but reference is ambiguous: ASK "刚才找到了 N 篇相关论文，基于这些做趋势分析？还是重新搜索？"
   • If no context → call paper_quick_scan(topic, limit=20) directly

2. Call paper_trend_data(topic, years_back=3) for publication counts and trends

3. Present as structured tables:

   趋势总览: | 子方向 | 2023 | 2024 | 2025 | 趋势 | 热门论文: | # | 标题 | 年份 | 引用 | 一句话 | 主要会议分布: | 会议 | 论文数 | 代表工作 | 结论与建议: 方向整体判断（上升/成熟/衰退）、最有潜力的子方向、入场时机建议

4. Auto-save: call paper_save_report(report_type="insight", content=<insight markdown>, filename="{topic}-{YYYY-MM-DD}.md") to persist the insight report.

5. [CONTEXT-AWARE FORK] — Based on the trend analysis result, suggest next steps:

   • If found a rising sub-direction: "趋势洞察已保存至 {path}。\n💡 下一步建议：\n1. 深入 [{rising_subdirection}] — 这个子方向在快速上升\n2. 做一个 [{rising_subdirection}] 的文献综述\n（说编号或告诉我你想做什么）"

   • If found a dominant/mature direction: "趋势洞察已保存至 {path}。\n💡 下一步建议：\n1. 分析 [{top_paper_title}] — 这个方向的代表工作\n2. 找研究空白 — 看看还有什么没被做过\n（说编号或告诉我你想做什么）"

   • If the direction is declining: "趋势洞察已保存至 {path}。这个方向发文量在下降。\n💡 下一步建议：\n1. 看看相邻方向 [{related_topic}] 的趋势\n2. 分析为什么在下降 — 是被新方法替代了吗？\n（说编号或告诉我你想做什么）"

If user wants to go deeper

Route to literature-survey skill with the selected sub-direction as topic.

Quick mode (default)

The above flow IS the quick mode. Two tool calls + AI formatting. No extra questions.

Full mode

Only when user explicitly asks for "详细分析" / "deep analysis":

• Expand paper_quick_scan(topic, limit=40)
• Expand paper_trend_data(topic, years_back=5)
• Generate a detailed report with method evolution timeline, key contributor analysis

Rules

• Only 1 checkpoint. The data gathering is fully automatic via paper_quick_scan + paper_trend_data.
• Don't ask about time range or sub-topics upfront — use defaults.
• Present trend data as a compact table with arrows (trend: up/down/stable).
• Always auto-save insight report via paper_save_report. Additional actions are opt-in via FORK.
• FORK suggestions must reference actual sub-directions, paper titles, and trend signals from the analysis result.