Company Tech Profile
Provided by Patsnap Eureka.
Produce an evidence-driven technology profile for a single company in a defined
technology topic. The deliverable is a structured report — conclusion first,
evidence second — suitable for technical leaders, analysts, and investment
screeners.
Use When
- A company's technical strength, R&D direction, or core technology layout in a topic
- Technology due diligence or partnership evaluation for a named company
- Investment screening that needs patent/paper/product evidence for one entity
- A company-only prompt where you can infer a safe technology topic from context
Do Not Use
- Multiple companies need comparison or tiering → route to
competitive-landscape
- Pure technology-route comparison with no company anchor → route to
tech-route-comparison
- The task is a project/proposal review → route to
rd-initiation-review
- Legal patent opinion, FTO, infringement, or litigation analysis
- Generic company overview with no technology depth requirement
Tool Routing And Fallback
This skill works across multiple tool environments. Before retrieval, detect
which capabilities are available and select the highest tier that is reachable.
Tier 1 (Recommended): Structured Patent/Paper Retrieval
- Use the host's best structured patent and paper retrieval stack.
- This usually means fielded search with company/entity filtering plus
record-level deep fetch.
- Evidence grade: S/A (primary structured evidence)
- Required capabilities: company/entity filtering, topic search, structured
patent/paper retrieval, and deep-read of shortlisted records.
- Typical query count per run: 4-8 searches.
Tier 2 (Fallback): Web Research + Companion Lanes
- Use the host's best broad web research tool plus scholarly or finance
companion lanes when available.
- Exa, Tavily, Brave, official filings, and domain-specific research databases
are good examples, not hard requirements.
- Evidence grade: A/B (web-sourced but structured)
- Coverage loss vs Tier 1: no structured assignee filtering, no IPC/CPC faceting,
and weaker paper-graph style coverage
Tier 3 (Fallback): Generic Web Tools
- Use generic web search and page/PDF reading tools available in the host.
- Evidence grade: B/C (unstructured web)
- Coverage loss vs Tier 2: no academic-focused ranking, broader noise
Tier 4 (Minimum): Pure LLM + User Materials
- No external tools required
- Evidence grade: C/U (LLM knowledge + user-provided only)
- Coverage loss vs Tier 3: no current information, no verifiable citations
Routing Rules
- Detect available tools at the start of the workflow.
- Select the highest available tier as the primary retrieval channel.
- When a tier is unavailable, explicitly state the downgrade in the output
and in
method_decisions.md.
- Do not pretend lower-tier results have the same coverage as higher tiers.
- When using Tier 3 or 4, increase the proportion of explicit gap statements
and reduce confidence language in conclusions.
Minimum Working Files
Create or update these files in a writable run folder:
request.md
workplan.md
method_decisions.md
query_log.csv
source_index.csv
claim_ledger.csv
report.md
Recommended subfolders are described in references/workflow.md.
Default Workflow
Step 0: Freeze Scope
Confirm or infer the following before any retrieval:
- company_name: the primary entity to profile
- topic: the technology area to anchor the analysis
- purpose: what decision this profile supports (orientation / due diligence /
partnership / investment screening)
- time_window: default last 3 years unless user specifies otherwise
Scope routing decision tree:
User input shape?
├─ Company + technology topic → proceed as single-company profile
├─ Company only, no topic → infer topic from company's main business domain;
│ confirm with user if ambiguous
├─ Multiple companies mentioned → redirect to competitive-landscape
├─ Technology topic only, no company → redirect to tech-route-comparison
└─ "Compare company A with B" → redirect to competitive-landscape
If the user only names a company without a topic, do not stall — infer the most
likely technology topic from the company's known domain and state the inference
explicitly.
Step 1: Anchor The Entity
- Normalize the company name: resolve parent vs subsidiary, Chinese vs English
aliases, brand name vs legal entity.
- Run a confirmation search to verify the company has presence in the topic.
- Tier 1: structured entity-filtered patent/paper search
- Tier 2+: targeted web or scholarly search across patent indices, official
domains, and research pages
- If the company has no patent presence in the topic, note this as a finding
and shift emphasis to papers and public signals.
- Search for the company's official site to capture basic profile, product
lines, and recent announcements.
Step 2: Patent Portfolio Analysis
- Run 2-3 patent searches varying keywords to cover the topic's sub-routes.
- Tier 1: structured patent search with entity filters and keyword variations
- Tier 2: targeted web research across patent indices, official domains, and
technical publications
- Tier 3/4: generic web or user-provided materials
- Analyze from search results: filing volume trend (by year), IPC distribution,
key technology clusters, filing jurisdictions.
- Decide whether direct deep-reading is stable enough, or whether representative
sampling is required.
Sampling decision tree (Tier 1 with structured retrieval):
Patent evidence shape?
├─ Scope is small and route concentration is obvious
│ └─ Skip sampler; deep-read the strongest 3-5 patents directly
├─ Scope is medium/large and route coverage matters
│ └─ Run the optional representative patent sampling script
├─ Scope is large and highly heterogeneous across time/jurisdiction/CPC
│ └─ Run the optional sampling script; treat result as default evidence pack
└─ Task is drifting toward legal completeness
└─ Do not use sampler; redirect to a legal/FTO-oriented workflow
Default thresholds:
<= 12 patents after filtering: usually skip sampler
13-80 patents: sampler recommended if route interpretation is part of the goal
> 80 patents: sampler should be the default
- Deep-read top 3-5 representative patents to extract: core technical
approaches, key claims, innovation points.
- Identify the company's primary technology routes and any emerging pivots.
Step 3: Paper And R&D Signals
- Run 1-2 paper searches with company name + topic.
- Tier 1: structured paper retrieval
- Tier 2: scholarly companion search or targeted web research
- Tier 3/4: generic web or user-provided materials
- Analyze: publication trend, collaborating institutions, research focus areas.
- Cross-reference with patent routes — consistency strengthens confidence;
divergence is a signal worth noting.
Step 4: Product And Public Signals
- Search for: product launches, benchmark results, partnerships, funding
rounds, key hires, conference presentations.
- Read 2-3 high-signal pages for deeper extraction.
- Corroborate patent/paper findings with product reality — a strong patent
portfolio without shipped products is a different signal than one with them.
Step 5: Synthesis
Adjust analysis emphasis by industry type:
Industry type?
├─ Semiconductor / Hardware → emphasize: patent clusters by process/architecture,
│ fab vs fabless signals, IPC H01L/H04 distribution
├─ Biopharma / Medtech → emphasize: pipeline stage, clinical trial signals,
│ target coverage, IPC A61K/C07/A61B distribution
├─ AI / Software → emphasize: papers and benchmarks, open-source presence,
│ model performance, API/platform adoption signals
├─ New Energy / Materials → emphasize: material composition patents, capacity
│ signals, cost trajectory, IPC H01M/C01/H02 distribution
└─ General Manufacturing → balanced patent + product + standard participation
Synthesize all evidence into the output skeleton. Separate every claim into:
- Verified fact: directly supported by patent, paper, or official filing
- Evidence-backed inference: reasonable conclusion from multiple signals
- Open gap: insufficient evidence, state what is missing and suggest next steps
Output Skeleton
1. Lead With Conclusion
1-2 sentences: the company's overall technical position and trajectory in the topic.
2. Technology Layout Highlights
2-4 bullet points: core technology routes, key patents or papers, differentiation
from the broader field. Use a compact markdown table when comparing 3+ dimensions.
3. Trend Assessment
1 paragraph: R&D direction evolution over the time window — acceleration,
deceleration, pivots, emerging focus areas. Cite specific year-over-year changes
when data supports it.
4. Risks And Recommendations
1 set of bullet points:
- Key risks (technology gaps, dependency, competitive pressure)
- Opportunities (white spaces, collaboration potential, emerging routes)
- Recommended next actions for the user's decision context
Every claim must cite its source type and identifier, e.g., [Patent: CN1234567B],
[Paper: DOI or title], [Web: source name].
Completion Gates
All must pass before delivering the final answer:
Guardrails
- Do not equate patent volume with technical strength without considering route
quality, claim scope, and recency.
- Do not treat company marketing claims, recruiting signals, or press releases as
verified facts until corroborated by stronger sources.
- Do not mix parent company and subsidiary data without an explicit scope note.
- This is a single-company skill — do not produce competitive comparison matrices.
If the user needs multi-company comparison, redirect to
competitive-landscape.
- If the company has minimal patent/paper presence in the topic, say so directly
rather than padding the profile with tangential evidence.
- Do not use representative sampling for legal-opinion tasks or when exhaustive
recall is required.
- Do not present the sampled patent set as "all relevant patents"; state clearly
that it is a structured representative subset.
- Do not let the sampled set silently override full-scope facts such as total
patent count, jurisdiction count, or exact filing volume.
- Do not force a strong conclusion when the evidence base is thin — prefer
explicit uncertainty over false confidence.
- Keep charts or visual aids to 1-2 maximum; retain only those that directly
support the analytical conclusion.
- For non-evidence dimensions (TCO, regulation, teardown), write "evidence
pending" rather than fabricating a judgment.
Optional Host-Specific Automation
Some hosts may provide extra helpers for representative sampling, quantitative
aggregation, or patent-card normalization. These helpers are optional and are
not part of the core open-source contract.
If you use host-specific automation:
- record the helper and its role in
method_decisions.md
- keep the final evidence compatible with
query_log.csv, source_index.csv,
and claim_ledger.csv
- do not make the report depend on a helper that other hosts cannot reproduce
Load These Files Only As Needed
- references/method-benchmark.md
- references/domain-playbooks.md
- references/workflow.md
- references/source-routing.md
- references/deliverables.md
- references/evidence-schema.md
- references/quality-gates.md
- templates/request-template.md
- templates/workplan-template.md
- templates/report-outline.md
1---2copyright: "Copyright © Patsnap. All rights reserved."3name: company-tech-profile4description:5 Single-company technology profile and R&D assessment for a defined technology6 topic. Use when the user asks for a company technology profile, company7 technical analysis, R&D direction assessment, technical due diligence8 briefing, or any single-company-by-topic technical evaluation — even if they9 only mention a company plus a technology area without explicitly asking for a10 "profile".11argument-hint: "[company + topic + optional purpose/time window]"12provider: "Patsnap Eureka"13compatibility: "Designed for Claude Code, Codex, and similar agent runtimes that can read/write local files and call whatever search tools are available."14deliverable-default: "Structured Markdown report plus traceable evidence files; docx/pdf export is optional."15fallback-policy: "Prefer structured patent/paper retrieval when available; otherwise downgrade through domain-specific sources, Exa, Tavily, Brave, web, and a known-URL reader without blocking the run."16metadata:17 source: https://skillmd.com/skills/patsnap/company-tech-profile18---1920# Company Tech Profile2122Provided by Patsnap Eureka.2324Produce an evidence-driven technology profile for a single company in a defined25technology topic. The deliverable is a structured report — conclusion first,26evidence second — suitable for technical leaders, analysts, and investment27screeners.2829## Use When3031- A company's technical strength, R&D direction, or core technology layout in a topic32- Technology due diligence or partnership evaluation for a named company33- Investment screening that needs patent/paper/product evidence for one entity34- A company-only prompt where you can infer a safe technology topic from context3536## Do Not Use3738- Multiple companies need comparison or tiering → route to `competitive-landscape`39- Pure technology-route comparison with no company anchor → route to `tech-route-comparison`40- The task is a project/proposal review → route to `rd-initiation-review`41- Legal patent opinion, FTO, infringement, or litigation analysis42- Generic company overview with no technology depth requirement4344## Tool Routing And Fallback4546This skill works across multiple tool environments. Before retrieval, detect47which capabilities are available and select the highest tier that is reachable.4849### Tier 1 (Recommended): Structured Patent/Paper Retrieval5051- Use the host's best structured patent and paper retrieval stack.52- This usually means fielded search with company/entity filtering plus53 record-level deep fetch.54- Evidence grade: S/A (primary structured evidence)55- Required capabilities: company/entity filtering, topic search, structured56 patent/paper retrieval, and deep-read of shortlisted records.57- Typical query count per run: 4-8 searches.5859### Tier 2 (Fallback): Web Research + Companion Lanes6061- Use the host's best broad web research tool plus scholarly or finance62 companion lanes when available.63- Exa, Tavily, Brave, official filings, and domain-specific research databases64 are good examples, not hard requirements.65- Evidence grade: A/B (web-sourced but structured)66- Coverage loss vs Tier 1: no structured assignee filtering, no IPC/CPC faceting,67 and weaker paper-graph style coverage6869### Tier 3 (Fallback): Generic Web Tools7071- Use generic web search and page/PDF reading tools available in the host.72- Evidence grade: B/C (unstructured web)73- Coverage loss vs Tier 2: no academic-focused ranking, broader noise7475### Tier 4 (Minimum): Pure LLM + User Materials7677- No external tools required78- Evidence grade: C/U (LLM knowledge + user-provided only)79- Coverage loss vs Tier 3: no current information, no verifiable citations8081### Routing Rules8283- Detect available tools at the start of the workflow.84- Select the highest available tier as the primary retrieval channel.85- When a tier is unavailable, explicitly state the downgrade in the output86 and in `method_decisions.md`.87- Do not pretend lower-tier results have the same coverage as higher tiers.88- When using Tier 3 or 4, increase the proportion of explicit gap statements89 and reduce confidence language in conclusions.9091## Minimum Working Files9293Create or update these files in a writable run folder:9495- `request.md`96- `workplan.md`97- `method_decisions.md`98- `query_log.csv`99- `source_index.csv`100- `claim_ledger.csv`101- `report.md`102103Recommended subfolders are described in [references/workflow.md](references/workflow.md).104105## Default Workflow106107### Step 0: Freeze Scope108109Confirm or infer the following before any retrieval:110111- **company_name**: the primary entity to profile112- **topic**: the technology area to anchor the analysis113- **purpose**: what decision this profile supports (orientation / due diligence /114 partnership / investment screening)115- **time_window**: default last 3 years unless user specifies otherwise116117Scope routing decision tree:118119```120User input shape?121├─ Company + technology topic → proceed as single-company profile122├─ Company only, no topic → infer topic from company's main business domain;123│ confirm with user if ambiguous124├─ Multiple companies mentioned → redirect to competitive-landscape125├─ Technology topic only, no company → redirect to tech-route-comparison126└─ "Compare company A with B" → redirect to competitive-landscape127```128129If the user only names a company without a topic, do not stall — infer the most130likely technology topic from the company's known domain and state the inference131explicitly.132133### Step 1: Anchor The Entity1341351. Normalize the company name: resolve parent vs subsidiary, Chinese vs English136 aliases, brand name vs legal entity.1372. Run a confirmation search to verify the company has presence in the topic.138 - Tier 1: structured entity-filtered patent/paper search139 - Tier 2+: targeted web or scholarly search across patent indices, official140 domains, and research pages1413. If the company has no patent presence in the topic, note this as a finding142 and shift emphasis to papers and public signals.1434. Search for the company's official site to capture basic profile, product144 lines, and recent announcements.145146### Step 2: Patent Portfolio Analysis1471481. Run 2-3 patent searches varying keywords to cover the topic's sub-routes.149 - Tier 1: structured patent search with entity filters and keyword variations150 - Tier 2: targeted web research across patent indices, official domains, and151 technical publications152 - Tier 3/4: generic web or user-provided materials1532. Analyze from search results: filing volume trend (by year), IPC distribution,154 key technology clusters, filing jurisdictions.1553. Decide whether direct deep-reading is stable enough, or whether representative156 sampling is required.157158Sampling decision tree (Tier 1 with structured retrieval):159160```161Patent evidence shape?162├─ Scope is small and route concentration is obvious163│ └─ Skip sampler; deep-read the strongest 3-5 patents directly164├─ Scope is medium/large and route coverage matters165│ └─ Run the optional representative patent sampling script166├─ Scope is large and highly heterogeneous across time/jurisdiction/CPC167│ └─ Run the optional sampling script; treat result as default evidence pack168└─ Task is drifting toward legal completeness169 └─ Do not use sampler; redirect to a legal/FTO-oriented workflow170```171172Default thresholds:173174- `<= 12` patents after filtering: usually skip sampler175- `13-80` patents: sampler recommended if route interpretation is part of the goal176- `> 80` patents: sampler should be the default1771784. Deep-read top 3-5 representative patents to extract: core technical179 approaches, key claims, innovation points.1805. Identify the company's primary technology routes and any emerging pivots.181182### Step 3: Paper And R&D Signals1831841. Run 1-2 paper searches with company name + topic.185 - Tier 1: structured paper retrieval186 - Tier 2: scholarly companion search or targeted web research187 - Tier 3/4: generic web or user-provided materials1882. Analyze: publication trend, collaborating institutions, research focus areas.1893. Cross-reference with patent routes — consistency strengthens confidence;190 divergence is a signal worth noting.191192### Step 4: Product And Public Signals1931941. Search for: product launches, benchmark results, partnerships, funding195 rounds, key hires, conference presentations.1962. Read 2-3 high-signal pages for deeper extraction.1973. Corroborate patent/paper findings with product reality — a strong patent198 portfolio without shipped products is a different signal than one with them.199200### Step 5: Synthesis201202Adjust analysis emphasis by industry type:203204```205Industry type?206├─ Semiconductor / Hardware → emphasize: patent clusters by process/architecture,207│ fab vs fabless signals, IPC H01L/H04 distribution208├─ Biopharma / Medtech → emphasize: pipeline stage, clinical trial signals,209│ target coverage, IPC A61K/C07/A61B distribution210├─ AI / Software → emphasize: papers and benchmarks, open-source presence,211│ model performance, API/platform adoption signals212├─ New Energy / Materials → emphasize: material composition patents, capacity213│ signals, cost trajectory, IPC H01M/C01/H02 distribution214└─ General Manufacturing → balanced patent + product + standard participation215```216217Synthesize all evidence into the output skeleton. Separate every claim into:218219- **Verified fact**: directly supported by patent, paper, or official filing220- **Evidence-backed inference**: reasonable conclusion from multiple signals221- **Open gap**: insufficient evidence, state what is missing and suggest next steps222223## Output Skeleton224225### 1. Lead With Conclusion2262271-2 sentences: the company's overall technical position and trajectory in the topic.228229### 2. Technology Layout Highlights2302312-4 bullet points: core technology routes, key patents or papers, differentiation232from the broader field. Use a compact markdown table when comparing 3+ dimensions.233234### 3. Trend Assessment2352361 paragraph: R&D direction evolution over the time window — acceleration,237deceleration, pivots, emerging focus areas. Cite specific year-over-year changes238when data supports it.239240### 4. Risks And Recommendations2412421 set of bullet points:243- Key risks (technology gaps, dependency, competitive pressure)244- Opportunities (white spaces, collaboration potential, emerging routes)245- Recommended next actions for the user's decision context246247Every claim must cite its source type and identifier, e.g., `[Patent: CN1234567B]`,248`[Paper: DOI or title]`, `[Web: source name]`.249250## Completion Gates251252All must pass before delivering the final answer:253254- [ ] Company entity confirmed and aliases resolved (not confused with subsidiary or homonym)255- [ ] At least 1 patent search and 1 paper search executed with results analyzed256- [ ] If patent scope is broad, representative sampling was executed or explicitly skipped with reason257- [ ] Core technology routes identified with patent or paper evidence258- [ ] Every major claim has a traceable source citation259- [ ] Verified fact / evidence-backed inference / open gap clearly separated260- [ ] Output follows the 4-part skeleton (conclusion → layout → trend → risks)261- [ ] No drift into generic company biography without technical substance262- [ ] Tool tier and coverage limitations explicitly stated in the output263- [ ] If evidence is insufficient for a confident profile, explicitly stated with follow-up suggestions264- [ ] Company marketing claims treated as signals, not verified facts, unless corroborated265266## Guardrails267268- Do not equate patent volume with technical strength without considering route269 quality, claim scope, and recency.270- Do not treat company marketing claims, recruiting signals, or press releases as271 verified facts until corroborated by stronger sources.272- Do not mix parent company and subsidiary data without an explicit scope note.273- This is a single-company skill — do not produce competitive comparison matrices.274 If the user needs multi-company comparison, redirect to `competitive-landscape`.275- If the company has minimal patent/paper presence in the topic, say so directly276 rather than padding the profile with tangential evidence.277- Do not use representative sampling for legal-opinion tasks or when exhaustive278 recall is required.279- Do not present the sampled patent set as "all relevant patents"; state clearly280 that it is a structured representative subset.281- Do not let the sampled set silently override full-scope facts such as total282 patent count, jurisdiction count, or exact filing volume.283- Do not force a strong conclusion when the evidence base is thin — prefer284 explicit uncertainty over false confidence.285- Keep charts or visual aids to 1-2 maximum; retain only those that directly286 support the analytical conclusion.287- For non-evidence dimensions (TCO, regulation, teardown), write "evidence288 pending" rather than fabricating a judgment.289290## Optional Host-Specific Automation291292Some hosts may provide extra helpers for representative sampling, quantitative293aggregation, or patent-card normalization. These helpers are optional and are294not part of the core open-source contract.295296If you use host-specific automation:297298- record the helper and its role in `method_decisions.md`299- keep the final evidence compatible with `query_log.csv`, `source_index.csv`,300 and `claim_ledger.csv`301- do not make the report depend on a helper that other hosts cannot reproduce302303## Load These Files Only As Needed304305- [references/method-benchmark.md](references/method-benchmark.md)306- [references/domain-playbooks.md](references/domain-playbooks.md)307- [references/workflow.md](references/workflow.md)308- [references/source-routing.md](references/source-routing.md)309- [references/deliverables.md](references/deliverables.md)310- [references/evidence-schema.md](references/evidence-schema.md)311- [references/quality-gates.md](references/quality-gates.md)312- [templates/request-template.md](templates/request-template.md)313- [templates/workplan-template.md](templates/workplan-template.md)314- [templates/report-outline.md](templates/report-outline.md)