energy-market-intelligence

Overview

Conducts energy market research across multiple authoritative sources to produce cited intelligence briefings on commodities, policy, regulation, technology, and competitive dynamics. Covers natural gas, crude oil, electricity markets (wholesale and retail), renewable energy markets, carbon pricing, liquefied natural gas trade, hydrogen, and energy storage. Produces reports calibrated to audience (executive summary or technical deep-dive) with source citations and confidence assessments.

Workflow

Confidence bands: high (three or more corroborating tier-1 sources), moderate (one or two sources or mixed signals), low (single source, speculative, or conflicting).

Depth levels:

• briefing: 3-5 key findings, 1-2 sources per finding, executive format.
• analysis: 6-10 dimensions, multiple corroborating sources, includes visualizations.
• deep_dive: multi-track parallel research using the deep_research skill.

Audience calibration: executive (3-5 key takeaways), technical (methodology detail), regulatory (compliance implications), investor (risk and return).

Market segments, pricing points, regulatory bodies, analytical frameworks, and the source catalog are defined in references/market-reference.md. Read that file during Research Scoping and Information Gathering rather than relying on internal knowledge.

<Workflow - Research Scoping description="Determine the research scope, verify reference data, identify relevant market segments, and plan information gathering." tools=[get_current_time, web_search, url_fetch, file_read] triggers=["User requests market intelligence or asks about energy market conditions"]

1. [Agent] Read references/market-reference.md to load the segment, pricing, regulatory, and source catalog. Call get_current_time to establish the research date; all findings must be dated. Validate: Reference catalog read and current date captured. If fails: If the reference file is missing, report it and proceed with the source catalog stated by the user. If the date is unavailable, ask the user for date context.

2. [Agent] Verify reference data before any calculation. Identify which time-sensitive values are needed (emission factors, thresholds, benchmarks, costs, regulatory limits) and fetch each current value using web_search or url_fetch from the authoritative source in the reference catalog. Validate: Every time-sensitive value needed downstream has a verified source fetched this session or provided by the user. If fails: Stop and ask the user to confirm or provide the value. Do not proceed with unverified time-sensitive values.

3. [Think] Parse the request ({{topic}}) and determine primary market segment(s), geographic scope, time horizon (near-term 0-6 months, medium-term 6-24 months, long-term 2-10+ years), the key questions to answer, and the audience calibration from {{audience}}.

4. [Decide] Set the plan from {{depth}}:

   • briefing: 3-5 key findings, executive format.
   • analysis: 6-10 dimensions, multiple corroborating sources, visualizations.
   • deep_dive: use the deep_research skill for multi-track parallel research. Validate: Scope is achievable within tool capabilities. If fails: If scope is too broad, ask the user to narrow: "This spans multiple markets. Should I focus on [option A] or [option B]?"

</Workflow - Research Scoping>

<Workflow - Information Gathering description="Collect market data, news, analysis, and regulatory filings from authoritative sources." tools=[web_search, url_fetch, file_read, file_read_pdf, run_python, start_task, create_task_group, get_task_group_result] triggers=["After research scope is defined"]

1. [Agent] Run 3-5 targeted web searches with different angles: current forecasts ("[topic] outlook"), policy ("[topic] regulatory policy"), fundamentals ("[topic] supply demand balance"), regional focus ("[topic] [geography]"), and competitive or stakeholder angle ("[entity] [topic]"). Validate: At least 3 searches return relevant results. If fails: Broaden terms or try alternative terminology (for example "natgas" versus "natural gas").

2. [Agent] For the top 3-5 results per search, fetch full page content using url_fetch. Extract key data points (prices, volumes, percentages), analyst quotes and forecasts, and policy details (effective dates, requirements, exemptions). Assign each source a credibility tier per the reference catalog. Validate: At least 5 high-quality sources accessed, mixing data sources and analysis sources. If fails: If a source is paywalled, note "source behind paywall" and rely on the snippet plus other sources.

3. [Decide] Did the user provide market data files?

   • Yes: analyze them with run_python (moving averages, year-over-year change, rolling volatility, correlations). Read tabular files with file_read and documents with file_read_pdf.
   • No: continue to step 4. Validate: Data analysis completes and trends are identified. If fails: Note "uploaded data analysis failed" and continue with web research.

4. [Decide] For deep_dive depth, use the deep_research skill to run parallel multi-track investigation, one track per dimension, coordinating with start_task, create_task_group, and get_task_group_result. For briefing or analysis, continue to synthesis. Validate: Sufficient source material collected for the requested depth. If fails: Report what is available and note where information was limited.

</Workflow - Information Gathering>

<Workflow - Analysis and Synthesis description="Synthesize gathered information into structured analysis with confidence assessments." tools=[run_python] triggers=["After information gathering completes"]

1. [Think] Organize findings by strategic importance: current state (confirmed facts with data), what is changing (trends with directional confidence), key drivers (supply and demand, policy, technology, geopolitics), scenarios (base, upside, downside), and implications for the user's context. Assign confidence per the bands in .

2. [Decide] Is quantitative data available?

   • Yes: generate trend analysis with run_python (price trends, moving averages, year-over-year comparison, supply-demand balance, scenario modeling for base, high, and low cases).
   • No: present qualitative analysis with data gaps noted. Validate: Quantitative analysis is consistent with qualitative findings. If fails: Present qualitative analysis with data gaps noted.

3. [Think] Identify contrarian signals and risks: where the consensus view has blind spots, what events could invalidate the base case, and what the market is pricing versus the fundamental view. Present these as "risks to outlook" rather than primary findings.

</Workflow - Analysis and Synthesis>

<Workflow - Report Generation description="Produce the final intelligence report calibrated to audience and depth." tools=[run_python, file_write, open_in_session_tab] triggers=["After analysis completes"]

1. [Agent] Structure the report per {{audience}} using the matching structure in . Validate: Report structure matches the requested audience and depth. If fails: Default to the executive structure.

2. [Decide] Does the data support visualizations?

   • Yes: generate charts (price trends with moving averages, supply-demand stacked bars, scenario forecast lines, capacity mix) using the html_design and highcharts skills. Label axes with units.
   • No: include the data in tabular format instead. Validate: Charts render with proper axis labels and units. If fails: Include the data in tabular format.

3. [Ask user] Ask where to save the deliverable, and whether a PDF is wanted. Offer the session workspace directory as the default location. Validate: An output location is confirmed. If fails: Default to the session workspace directory.

4. [Agent] Write the report as markdown to the confirmed location using file_write. Ensure every claim has an inline citation, confidence levels are stated for forecasts, units and dates appear on all data points, fact is separated from opinion, no stale forecast is presented as current without noting its age, and the Rule 12 disclaimer is included. Validate: Report saved with no uncited claims and no undefined acronyms on first use. If fails: Output the report in chat if the file save fails.

5. [Agent] Open the report with open_in_session_tab. If the user requested a PDF, generate it using the canvas_pdf skill. Validate: The file is visible to the user. If fails: Output the key findings directly in chat.

</Workflow - Report Generation>

Executive format:

• Key Takeaways (3-5 bullets, each one sentence)
• Market Context (1-2 paragraphs)
• Outlook and Scenarios (table: base, upside, downside with probabilities)
• Implications and Recommended Actions
• Sources
• Disclaimer (Rule 12)

Technical format:

• Executive Summary
• Market Fundamentals (supply, demand, storage, trade flows)
• Price Analysis (current, historical, forward curve)
• Regulatory and Policy Environment
• Technology and Infrastructure Developments
• Risk Assessment (probability-weighted scenarios)
• Data Appendix
• Full Source List
• Disclaimer (Rule 12)

Regulatory format:

• Summary of Changes
• Affected Parties and Compliance Timeline
• Technical Requirements
• Cost-Benefit Assessment
• Comparison to Prior Rules
• Industry Response and Legal Challenges
• Sources with Docket Numbers
• Disclaimer (Rule 12)

Investor format:

• Investment Thesis (3-5 bullets)
• Market Context and Catalysts
• Risk and Return Scenarios (base, upside, downside with probabilities)
• Key Risks to Outlook
• Sources
• Disclaimer (Rule 12)