SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillstechnology-mapping

technology-mapping

Agents & models

View repository

Updated 10-15 19:55

About this skill

Technology Mapping Skill (v2.1.1)

Overview

Fully automatic generation of a technology panorama map (technology origin mapping) for a specified frontier technology field. Through a systematic workflow of 6 phases, it automatically completes field analysis, target discovery, lineage tracing, pruning and clustering, visualization output, and quality verification.

Core Features：

  • 🧠 Field Adaptivity: The LLM dynamically analyzes field characteristics at runtime and generates customized search strategies (replacing fixed templates)
  • 🔄 Bidirectional Mining: Forward tracking of academic origins + reverse tracing of star project founding teams
  • 🔗 Wikidata Lineage Query: Integrates 380,000 doctoral advisor relationship records, greatly reducing lineage tracing breakpoints
  • ✂️ Aggressive Pruning + Connection Strength Grading: Three-level connection strength system, retaining only Tier 1/2 connections
  • 📊 Measurable Lineage Quality: Automatically calculates lineage completion rate, verification rate, and orphan rate

When to Use

Use when the user needs to conduct a panoramic study of atechnology-centric frontier track:

  • "Help me do a technology mapping of neutral atom quantum computing"
  • "Generate a technology panorama of LLM Agent frameworks"
  • "Track map of the nuclear fusion field"

Not applicable: Highly mature non-technology-driven industries.

Input and Output

Input

The user provides a technology field keyword:"中性原子量子计算" / "LLM Agent 框架" / "核聚变磁约束"

Output

  1. [Topic]_TechMap.svg — SVG vector map generated by Graphviz dot
  2. [Topic]_TechMap_Report.md — Accompanying text report (node summaries + confidence annotations + lineage statistics + VC analysis conclusions)
  3. generate_techmap.py — Map generation script (reproducible, iteratively modifiable)

Data Schema

For the node type table, edge type table, and confidence rules, see data_schema.md For the three-level connection strength layering system, see connection_strength.md For visualization node color scheme/edge styles/timeline/Legend specifications, see visual_spec.md For the VC analysis framework (founder evaluation + technology route + China-foreign comparison + traceability), see vc_framework.md

Search Tool Selection Specifications

In any step that requires executing a web search, select tools according to the following process:

1. 尝试调用 tavily-search 技能
2. IF tavily-search 调用成功 → 使用结果
3. IF tavily-search 未安装 OR 调用报错（ImportError / HTTP 错误 / 超时）:
   → 记录降级原因
   → 调用系统原生 search_web 工具（相同 query）
   → 使用 search_web 结果
4. IF search_web 也失败 → 标记该搜索步骤为「未完成」，继续下一步

Execute Workflow

[!CAUTION] Strictly execute in the order Phase 0 → 1 → 2 → 3 → 4 → 5. After each Phase is completed, internally check the output quality before proceeding to the next Phase.

Phase 0: Field Adaptation & Strategy Generation 🧠

Purpose: Have the LLM analyze the characteristics of the technology field at runtime and dynamically generate adapted search strategies and lineage tracing paths.

Steps：

  1. Field Characteristic Analysis: Based on the LLM's own knowledge, analyze the following dimensions and output domain_profile：

    {
  "domain": "量子计算-中性原子",
  "knowledge_carrier": {
    "论文": 0.8, "专利": 0.3, "开源代码": 0.1, "产品": 0.05
  },
  "talent_flow_pattern": "大学PhD → 博后 → 教授创业 or 大厂",
  "known_anchor_persons": ["Mikhail Lukin", "Antoine Browaeys", "..."],
  "known_anchor_companies": ["QuEra", "Pasqal", "..."],
  "china_pattern": "海归教授 + 中科大/清华系",
  "upstream_path_type": "PhD advisor chain (2-3 layers)",
  "special_relations": ["同实验室师兄弟", "共同导师", "技术迁移(冷原子→中性原子)"],
  "search_templates": {
    "company_overseas": "[keyword] startup unicorn funding series",
    "company_china": "[关键词] 中国 创业公司 融资 A轮 B轮",
    "academic": "[person] PhD advisor [field] university",
    "academic_china": "[中文名] 博士 导师 留学 [领域]",
    "lab_alumni": "[advisor] lab alumni students notable startup"
  }
}
    • ⚠️ AI/open source fields:talent_flow_pattern should be "大厂核心团队 → 独立创业" or "顶级实验室 → 开源项目 → 创业"
    • ⚠️ Hard tech fields (quantum/nuclear fusion/biology): should be "大学PhD → 博后/教授 → 创业"

  2. Anchor Verification: Use a search engine to quickly verify 2-3 anchors and correct deviations.

  3. User Confirmation ⚠️: Display the domain_profile summary and ask whether any corrections or additions are needed.

Phase 0 Deliverables：domain_profile(Domain characteristics + search strategy + anchor list), confirmed by the user.

Phase 1: Global Target Discovery

Purpose: Find the "North Star" nodes of the track—overseas leading companies, domestic startups, core open-source projects, and academic milestones.

Steps：

  1. Search leading companies (overseas): Use search_templates.company_overseas, filter for unicorns valued at >$1B or benchmark projects of big tech companies.
  2. Search domestic startups: Use search_templates.company_china, filter for total funding >100 million yuan or with public funding in the past two years.
    • ⚠️ The above templates target the Chinese market by default. If coverage of other entrepreneurial ecosystems such as Japan/South Korea/Israel is needed, adapt to the local language.
  3. Search core open-source projects(when knowledge_carrier.开源代码 > 0.3 ): industry-recognized foundational frameworks/models.
  4. Search academic milestones and major awards(when knowledge_carrier.论文 > 0.3 ): use the openalex-database skill. ⚠️ Nobel Prize/Wolf Prize/National Highest Award winners must be included.
  5. Search domestic academic origins: Chinese Academy of Sciences system, Tsinghua/Peking University/Fudan/USTC and other important research institutions.
  6. Build a technology evolution timeline: extract 3-7 key technology milestone years.
  7. Interactive confirmation ⚠️: Show the discovered list to the user and wait for additions.

Phase 1 Deliverables: Target list (12-25 high-value nodes) + technology evolution timeline, confirmed by the user.

Phase 2: Upstream Tracing

Purpose: Starting from all Targets, trace upward to discover source figures and lineages of mentorship. Tracing depth: 2-3 layers.

⚠️ Connection strength rules: All relationships must be classified by connection_strength.md . Only Tier 1/2 enter the graph. Core determination: citation ≠ lineage。

Steps：

  1. Wikidata P184 quick query (priority): For each person, first query scripts/wikidata_client.py for query_doctoral_advisor(). If there is a result → skip search; if there is no result → enter the progressive search funnel. At the same time, use query_students() to laterally discover more entrepreneurs.

    • For batch queries use batch_query_advisors()(SPARQL VALUES batch query, avoid N+1)

  2. Progressive search funnel(when Wikidata has no result):

    • Layer 1(<30s/person): Search engines directly query PhD advisor / educational background
    • Layer 2(<60s/person): OpenAlex co-author cross-verification + university official websites
    • Layer 3(<2min/person): Chinese-specific templates + news/interviews + Wikidata indirect inference
    • Still not found → mark as orphan node

  3. Continue tracing the advisor's advisor (2-3 layers): Retention conditions—Nobel Prize level / trained 2+ core figures / bridge nodes.

  4. Tracing termination rules: Stop when any one is met — domain founder / Nobel or Turing Award / cross-domain boundary / information unavailable / age exceeds threshold (60 years for hard tech, 20 years for AI/software).

    ⚠️ For each leaf-node founder, if the technology is not independently original, it must be traced back to the independent proposer or domain founder.

  5. Lab horizontal scan: For core advisors, search all important students/postdocs to discover fellow-lab founders.

  6. AI/open-source domain adaptation(when talent_flow_pattern contains "big tech"): Build a traceability chain of "big tech Lab → startup".

  7. VC analysis framework: Each company node must be tagged with a technology source label according to vc_framework.md .

  8. Traceability completeness check ⛔: Check nodes one by one, count traceability completion rate / verification rate / orphan rate. Goal: completion rate >70%, orphan rate <20%.

  9. Entity disambiguation: Cross-validate from multiple sources. Use query_person_info(name, expected_field=...) to perform Wikidata disambiguation.

Phase 2 deliverables: Origin node list + relationship edges + orphan node list + traceability metrics. Expected 30-50 nodes.

Phase 3: Aggressive pruning & forward fill (Pruning & Forward Fill)

Purpose: Clean data quality, aggregate into technology schools.

Steps：

  1. Connection strength filtering: Tier 1 → keep; Tier 2 → keep but mark as weak connection; Tier 3 → delete.
  2. Node pruning: Keep — pioneers of subdirections / trained 2+ founder students / founded open-source projects / top-tier awards / bridge nodes. Delete — pure optimization papers / purely academic with no commercialization / intermediate nodes covered by upstream nodes / isolated nodes with only Tier 3 relationships.
  3. Forward fill: For grandmaster-level nodes, forward-search for omitted commercialization transitions. ⚠️ Especially check domestic scholars/students.
  4. Clustering and grouping: Aggregate into 3-8 schools by core technology framework/route name. ⛔ Grouping by country/region is absolutely not allowed。
  5. Generate technology route difference annotations: Generate a "route feature card" for each Group (format see data_schema.md）。

Phase 3 deliverables: Clean graph data (nodes + edges + groups + route_notes + timeline), expected 40-60 nodes.

Phase 4: Visualization output (Graphviz SVG)

Purpose: Convert graph data into a high-quality, structured SVG vector graph.

Technical approach: Use Python graphviz library + dot layout engine.

[!CAUTION] The Graphviz dot engine must be used to generate SVG. Obsidian Canvas (.canvas), NetworkX + PyVis HTML, or any alternative is prohibited. The generation script generate_techmap.py must use import graphviz and call graphviz.Digraph, outputting .svg file.

⚠️ Prerequisites: The system needs Graphviz installed (winget install Graphviz.Graphviz). Python libraries:pip install graphviz。

Steps：

  1. Create generate_techmap.py, use graphviz.Digraph(format='svg', engine='dot')
  2. Graph global settings: white-background scientific style,rankdir: TB，dpi: 150，fontname: Microsoft YaHei
  3. node colors, edge styles, timeline, Legend — strictly follow visual_spec.md
  4. layout constraints — Strictly follow visual_spec.md layout constraint specification(edge direction, cluster grouping, weight/constraint, rank anchoring)
  5. Node/edge type definitions — Strictly follow data_schema.md
  6. VC analysis annotations — Strictly follow vc_framework.md
  7. Layout validation — After generating the SVG, follow debugging checklist to check layout correctness

Phase 4 deliverables: SVG vector graph + generation script.

Phase 5: Quality validation & report generation 📊

Purpose: Generate a quality report so users can understand the credibility and completeness of the graph.

Steps：

  1. Generate accompanying report ([Topic]_TechMap_Report.md)：

    • Track overview + domain adaptation analysis + technology route list + all node inventory tables + key findings
    • VC analysis conclusions(independent section, must include): in-depth founder rankings, technology route maturity comparison, clarity rating of overseas traceability for Chinese companies, summary of investment risk signals

  2. Traceability verification table ⚠️ (must include):| 人物 | 关系声明 | Source 1 (URL) | Source 2 (URL) | 置信度 | 状态 |

  3. Traceability quality statistics: traceability completion rate / multi-source verification rate / orphan rate / average confidence

Phase 5 deliverables: text report (including traceability verification table + quality statistics).

Dependency skills

Skill Purpose Required
tavily-search Preferred search: if the environment is missing, degrade to search_web Optional (has fallback)
openalex-database Query academic papers, authors, institutions, citation networks Optional (recommended for academic domains)
graphviz (system + Python library) Only diagramming solution for Phase 4: dot engine generates SVG vector graphs ⚠️ Required

⛔ Deprecated: As of v2.0.1, Obsidian Canvas (.canvas) and NetworkX + PyVis HTML output are no longer supported.

Auxiliary scripts

Script Purpose
scripts/wikidata_client.py Query Wikidata doctoral advisor (P184) relationships, reverse-lookup students, obtain academic information for people, batch queries

Reference documents

Document Content
data_schema.md Node/edge type definitions, confidence rules, route feature card format
connection_strength.md Three-level connection strength criteria + 7 boundary cases
visual_spec.md Graphviz node color scheme, edge styles, timeline, Legend specification
vc_framework.md VC analysis framework (founder evaluation + technology route + traceability)

Important principles

  1. Fully automatic + interactive confirmation: Users can start by providing only keywords; automatically pause for confirmation at the end of Phase 0 and Phase 1
  2. Domain adaptive: Dynamically generate strategies through Phase 0, without relying on fixed templates
  3. Moderately dense: target node count 40-60
  4. Domestic coverage: Must actively search for domestic startups (financing > 100 million RMB or financing in the past two years)
  5. Visible technical differences: Each technology route Group must have a route feature card annotation
  6. Nobel Prize retention exception: Top-tier major award scholars are retained as long as they are related to the technical route
  7. Wikidata priority: When tracing, check Wikidata first, and use the result directly if there is one; for batch queries use batch_query_advisors()
  8. Confidence transparency: Mark confidence when it cannot be confirmed, and never fabricate mentorship relationships
  9. Color is semantics: Strictly follow the visual_spec.md color system
  10. Chinese annotation: Keep the original English names for people plus Chinese annotations, and use 🇨🇳 to mark Chinese companies/scholars
  11. Timeline required: There must be a timeline anchored on the left side
  12. VC analysis required: Each company node must be labeled with 🏷️ technical source, see vc_framework.md
  13. Graphviz dot standard: SVG output, white-background scientific research style color scheme