lesliexqing Market Report
Generate professional-grade financial analysis reports. The core value is not formatting — it is the analytical rigor applied before writing.
Step 1: Gather Material
If the user provides raw material (news, articles, notes), use that directly.
If the user only provides a topic (e.g., "本周A股周报"), search for recent relevant news and market data before proceeding.
Step 2: Signal Filtering — DO NOT SKIP
Before writing anything, evaluate every piece of source material. For each item, answer these four questions:
- Novelty — Is this new information, or already priced in / widely known?
- Expectation gap — Does reality differ from market consensus? (This is where alpha lives)
- Source reliability — Official data (central bank, earnings) > mainstream financial media > social media / rumors
- Reaction potential — Is capital already moving on this, or is it noise?
Action: Discard items that fail on 3+ questions. Prioritize items with clear expectation gaps — these drive the strongest report sections.
Show the filtering result to the user as a brief table before writing:
| Signal | Novelty | Exp. Gap | Source | Keep/Drop |
|---|
This forces transparency and lets the user override your judgment.
Step 3: Thematic Clustering
Group the surviving signals into 3-5 investment themes.
Rules:
- Cluster by causal relationship, not by sector label
- Theme titles MUST be directional and narrative-driven:
- GOOD: "Rate cut cycle fueling liquidity rotation into growth"
- BAD: "Macro economy"
- Each theme needs 2+ signals to justify its existence
- Order themes by market impact significance (highest first)
- If a signal doesn't fit any theme, it's either a standalone highlight or should be dropped
Step 4: Write Each Theme Section
Every theme section follows this three-layer structure:
Layer 1: Macro Context
Why does this theme matter right now? What is the broader environment?
Layer 2: Transmission Mechanism
How does the trigger event propagate through the market? Be specific about the causal chain — do not hand-wave.
Layer 3: Investment Impact
Which specific sectors, tickers, or asset classes are affected? State the direction (bullish/bearish) and the reasoning.
Critical rule: The same event affects different assets differently. A rate hike is bearish for growth stocks but bullish for banks. Never paint an entire theme with one color.
Writing quality standards:
- Every claim needs evidence ("PBOC data shows..." not "the market believes...")
- Distinguish fact from opinion explicitly ("we expect..." vs "data confirms...")
- Every paragraph must pass the "so what" test — if a reader asks "what does this mean for my portfolio?", the paragraph should answer it
- Include counter-arguments. A report that only presents the bull case is not professional — it's marketing
- Be specific: "semiconductor equipment names like ASML, AMAT" not "some tech stocks"
Step 5: Assemble the Report
Choose the report type based on user intent, then read references/templates.md for the matching skeleton structure.
| Type | When | Length | Key Feature |
|---|---|---|---|
| Morning Brief | Daily quick update | 500-800 words | Time-sensitive, 3-5 bullet points, no deep analysis |
| Weekly Report | Weekly market review | 1500-3000 words | Thematic structure, backward + forward looking |
| Deep Dive | Single topic analysis | 2000-5000 words | Exhaustive analysis of one theme, full evidence chain |
| Strategy Note | Forward-looking allocation | 1500-3000 words | Asset allocation recommendations, risk scenarios |
Assembly checklist:
- Executive summary that stands alone (reader grasps key insights without reading full report)
- Risk factors section with at least 3 distinct risks and counter-arguments
- Clear heading hierarchy (H1 title, H2 themes, H3 sub-sections)
- Follow user's language (Chinese input produces Chinese report)
Step 6: Output
Choose the output format based on report type:
| Report Type | Output Format | Source |
|---|---|---|
| Weekly Report | HTML (multi-page, styled) | assets/template.html |
| Deep Dive | HTML (adapted from template) | assets/template.html |
| Morning Brief | Markdown (lightweight, single page) | references/templates.md |
| Strategy Note | Markdown (unique structure) | references/templates.md |
For HTML output (Weekly / Deep Dive):
Read assets/template.html, then replace all {{PLACEHOLDER}} variables with report content from Steps 2-5. The template provides:
- Cover page with headline, summary box, and key data
- Table of contents (colored background)
- Signal filtering table (Step 2 output)
- Theme sections with data cards, three-layer analysis, and counter-argument boxes
- Risk grid + data calendar + scenario analysis
- References page with disclaimer
Generate one <div class="page"> per logical section. Typical report has 7-9 pages:
- Cover → 2. TOC → 3. Signal filtering → 4-6. Themes → 7. Risk & Outlook → 8. References
Theme colors — The template uses CSS variables. Pick one that matches the report tone:
| Theme | --c-primary |
Best For |
|---|---|---|
| Ember (default) | #ea4821 |
General market reports, high-energy topics |
| Ocean | #1a56db |
Corporate strategy, fixed income, institutional |
| Forest | #0d7c3e |
ESG, wealth management, positive outlook |
| Amethyst | #7c3aed |
Fintech, AI/tech, innovation themes |
| Slate | #334155 |
Neutral/bearish tone, risk-heavy reports |
To switch: uncomment the desired :root block in CSS, comment out the others.
For Markdown output (Morning Brief / Strategy Note):
Read references/templates.md for the matching skeleton structure. Output Markdown directly in conversation. These report types have distinct structures that don't fit the multi-page HTML layout.
Format conversion
If the user needs docx/pdf from any output, suggest the skywork-doc skill.
Quality Self-Check Before Delivery
- Signal filtering was performed and shown to user (Step 2 not skipped)
- 3-5 themes, each with directional title
- Every theme has all three layers (macro, transmission, impact)
- Mixed impact directions across themes (not all bullish or all bearish)
- Specific tickers/sectors named, not vague references
- Counter-arguments and risks included
- Evidence cited for key claims
- Report type matches user intent