Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Deep financial research with the FinSight multi-agent system
.claude/skills/brycewang-stanford-finsight-research-guide/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -52% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -24% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -52% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -2% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -42% | 0% |
FinSight is a deep research agent designed specifically for financial analysis. Developed by RUC-NLPIR, it combines multi-source data retrieval, financial reasoning, and report generation to produce publication-ready financial research. It handles market analysis, company fundamentals, sector comparisons, and macroeconomic assessment through specialized agents.
bashgit clone https://github.com/RUC-NLPIR/FinSight.git cd FinSight && pip install -e .
pythonfrom finsight import FinSightAgent agent = FinSightAgent(llm_provider="anthropic") # Generate comprehensive financial analysis report = agent.research( "Analyze the competitive landscape of the global EV battery " "market. Compare CATL, LG Energy, and Panasonic on market " "share, technology, margins, and growth outlook." ) print(report.summary) report.save("ev_battery_analysis.pdf")
| Agent | Role | |-------|------| | Retrieval Agent | Fetches data from SEC filings, financial APIs, news | | Data Agent | Processes financial statements, ratios, time series | | Analysis Agent | Performs fundamental, technical, and comparative analysis | | Reasoning Agent | Synthesizes findings, identifies trends and risks | | Report Agent | Generates structured research reports with citations |
python# FinSight integrates with multiple data sources config = { "sec_edgar": True, # SEC filings (free) "fred": True, # Federal Reserve economic data "yahoo_finance": True, # Market data (free) "news_api": True, # Financial news "world_bank": True, # Macro indicators }
python# Company fundamental analysis report = agent.research( "Provide a fundamental analysis of NVIDIA including " "revenue trends, margin analysis, valuation multiples, " "and competitive moat assessment." ) # Sector analysis report = agent.research( "Compare the top 5 cloud computing companies by revenue " "growth, operating margins, and R&D investment intensity." ) # Macro analysis report = agent.research( "Analyze the impact of rising interest rates on US " "commercial real estate valuations since 2022." )
Generated reports typically include:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-12 | pass→pass | 12,944 | 3,773 | -71% | 1 | 1 | 0% | 1,676 | 1,214 | -28% | 0 | 0 | — |
case-01 | fail→pass | 25,435 | 9,026 | -65% | 1 | 1 | 0% | 4,709 | 2,273 | -52% | 0 | 0 | — |
case-02 | fail→pass | 17,334 | 9,254 | -47% | 1 | 1 | 0% | 3,217 | 2,445 | -24% | 0 | 0 | — |
case-07 | fail→pass | 12,786 | 1,901 | -85% | 1 | 1 | 0% | 2,207 | 1,052 | -52% | 0 | 0 | — |
case-03 | fail→pass | 9,189 | 4,869 | -47% | 1 | 1 | 0% | 1,606 | 1,566 | -2% | 0 | 0 | — |
case-04 | fail→pass | 10,591 | 1,998 | -81% | 1 | 1 | 0% | 1,758 | 1,025 | -42% | 0 | 0 | — |
case-05 | fail→pass | 12,466 | 5,198 | -58% | 1 | 1 | 0% | 2,317 | 1,646 | -29% | 0 | 0 | — |
case-06 | pass→pass | 13,083 | 2,502 | -81% | 1 | 1 | 0% | 2,164 | 1,138 | -47% | 0 | 0 | — |
case-08 | fail→pass | 9,399 | 1,896 | -80% | 1 | 1 | 0% | 1,690 | 1,031 | -39% | 0 | 0 | — |
case-09 | pass→pass | 5,579 | 2,107 | -62% | 1 | 1 | 0% | 950 | 1,055 | +11% | 0 | 0 | — |
case-10 | pass→pass | 10,490 | 1,826 | -83% | 1 | 1 | 0% | 1,431 | 1,019 | -29% | 0 | 0 | — |
case-11 | pass→pass | 9,585 | 2,010 | -79% | 1 | 1 | 0% | 1,529 | 1,061 | -31% | 0 | 0 | — |
case-13 | pass→pass | 9,016 | 2,308 | -74% | 1 | 1 | 0% | 1,374 | 1,066 | -22% | 0 | 0 | — |
case-14 | fail→pass | 10,069 | 2,410 | -76% | 1 | 1 | 0% | 1,433 | 1,142 | -20% | 0 | 0 | — |
case-15 | fail→pass | 19,927 | 13,267 | -33% | 1 | 1 | 0% | 3,437 | 3,127 | -9% | 0 | 0 | — |
case-16 | fail→pass | 13,858 | 7,146 | -48% | 1 | 1 | 0% | 2,881 | 2,038 | -29% | 0 | 0 | — |
case-17 | fail→pass | 21,664 | 13,551 | -37% | 1 | 1 | 0% | 3,527 | 3,146 | -11% | 0 | 0 | — |
case-18 | fail→pass | 17,875 | 4,655 | -74% | 1 | 1 | 0% | 2,745 | 1,525 | -44% | 0 | 0 | — |
case-19 | fail→pass | 44,818 | 1,880 | -96% | 1 | 1 | 0% | 2,675 | 983 | -63% | 0 | 0 | — |
case-20 | pass→pass | 6,952 | 5,162 | -26% | 1 | 1 | 0% | 1,240 | 1,733 | +40% | 0 | 0 | — |
case-21 | pass→pass | 15,790 | 19,374 | +23% | 1 | 1 | 0% | 2,913 | 3,783 | +30% | 0 | 0 | — |
case-22 | pass→fail | 37,424 | 42,487 | +14% | 1 | 1 | 0% | 8,209 | 8,953 | +9% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +55 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.