@axiom-lattice/examples-deep_research 1.0.34 → 1.0.35

package/src/agents/data_agent/skills/business-reporting/SKILL.md

@@ -0,0 +1,115 @@
+---
+name: business-reporting
+description: defines how to transform "raw data" and "fragmented insights" into Executive-ready business decision reports.
+The core objective of the report is NOT to "display data," but to "drive decisions."
+metadata:
+  category: global
+---
+High-Impact Business Reporting Guide
+
+0. Skill Definition and Objective
+
+This skill defines how to transform "raw data" and "fragmented insights" into Executive-ready business decision reports.
+The core objective of the report is NOT to "display data," but to "drive decisions."
+
+1. Narrative Structure: SCQA and the Minto Pyramid Principle
+
+Do not write the report in the chronological order of your data analysis. You must write it following the logical flow of the reader's (decision-maker's) decision-making process.
+
+1.1 The Opening SCQA Framework (For the Executive Summary)
+
+The first three sentences of your opening must clearly address the following four points:
+
+Situation: Describe the stable business background (e.g., "For the first three quarters of this year, our revenue maintained a steady 20% growth.").
+
+Complication: Point out the anomaly or disruption that broke the stability (e.g., "However, in November, revenue in the East Region suddenly plunged by 35%.").
+
+Question: Introduce the core issue (e.g., "What caused this plunge, and will it spread to other regions?").
+
+Answer: Directly provide your analytical conclusion and core recommendation (e.g., "The drop was caused by a core competitor's price war. We recommend immediately matching their subsidy strategy in the East Region.").
+
+1.2 Top-Down Approach (Conclusions First)
+
+Strictly Prohibited: Placing the conclusion at the very end of the report.
+
+Mandatory Action: The first sentence of every section and every paragraph MUST be the core argument of that section. All subsequent data and charts exist solely to prove that first sentence.
+
+2. Elevating Insights: The "Soul 3 Questions" Formula
+
+To upgrade mediocre "data descriptions" into valuable "business insights," you must strictly execute the following formula:
+High-Value Insight = What (Phenomenon & Data) + So What (Business Impact) + Now What (Actionable Recommendation)
+
+❌ Incorrect Example (Only "What")
+
+"In November, the Average Order Value (AOV) dropped from $150 to $120, a 20% decrease."
+
+Why it's wrong: This makes you a "data broadcaster." Executives can see this on their dashboards; they don't need a report for it.
+
+🟡 Mediocre Example (What + So What)
+
+"In November, the AOV dropped from $150 to $120. This is because, during the major promotional campaign, we acquired a massive number of new users from lower-tier markets, which dragged down the overall average."
+
+Why it's mediocre: You found the cause, but you didn't tell the business team what to do about it.
+
+✅ Perfect Example (The Complete Formula)
+
+"In November, AOV dropped by 20% (to $120), primarily driven by a massive influx of new users from lower-tier markets during the promotion. (What + So What)
+Despite the lower AOV, the absolute gross profit increased by 15%. For this newly acquired, price-sensitive cohort, we recommend shifting next month's marketing strategy from 'threshold discounts' to 'high-frequency, low-value repurchase coupons' to boost their retention and LTV. (Now What)"
+
+3. Visual & Copywriting Constraints: High Signal-to-Noise Ratio
+
+3.1 Chart Titles MUST be "Action Titles"
+
+Chart titles should never be purely descriptive; they must be conclusive. The reader should know what you want to convey without even looking at the numbers in the chart.
+
+❌ Prohibited: "2024 Regional Sales Trend Chart"
+
+✅ Mandatory: "East Region Sales Suffered a Cliff-like Drop in Nov, Dragging Down National Performance"
+
+3.2 Eliminate Vague Adverbs, Force Quantification
+
+The use of subjective adjectives or adverbs is strictly prohibited. You must replace them with specific numbers or probabilities.
+
+❌ Prohibited: "Sales increased significantly," "Refund rates seem a bit abnormal," "The effect is probably not great."
+
+✅ Mandatory: "Sales increased by 34.2% MoM," "Refund rates breached the 5% warning line," "Confidence level is < 80%; cannot currently prove the strategy is effective."
+
+3.3 Visual "Zebra Crossings" (Formatting)
+
+Currency/Percentages: In Markdown tables, these must be right-aligned (---:) to facilitate easy comparison of digits.
+
+Status Indicators: Use Emojis (✅/❌/⚠️) to build visual anchors for quick scanning.
+
+Bold Critical Metrics: In a long paragraph of text, ONLY bold the most critical metrics (e.g., "dropped by 35.4%").
+
+4. Defensive Writing: Building a Baseline of Data Trust
+
+If an executive spots a single common-sense data error in your report, the credibility of your entire conclusion will be overturned.
+
+4.1 Force Data Definitions (In the Appendix or Intro)
+
+Never assume that everyone has the same definition for a metric.
+
+Example: "Note: 'Active User' in this report is defined as a deduplicated Device ID that generated at least one click event on that day, excluding internal test accounts."
+
+4.2 Strictly Distinguish Between "Correlation" and "Causation"
+
+Without rigorous A/B testing or attribution models, the use of absolute causal words (e.g., "caused," "led to") is strictly prohibited.
+
+❌ Prohibited: "The homepage redesign caused the conversion rate to increase by 5%."
+
+✅ Mandatory: "Following the launch of the redesigned homepage, the conversion rate increased by 5%, showing a high positive correlation. However, this overlapped with the weekend effect; we recommend extending the observation period."
+
+5. Final Deliverable Checklist
+
+After drafting the report, you must perform the following 5 checks:
+
+[ ] Elevator Pitch Test: Can the reader figure out what happened to the business and what action needs to be taken within the first 30 seconds of reading the first page?
+
+[ ] Chart-less Reading Test: If you cover all charts and tables and only read the text, is the logical chain still coherent and self-consistent?
+
+[ ] Closed-Loop Verification: For every "question/anomaly" raised in the report, is there a corresponding "hypothesis" or "clear attribution" provided later in the text?
+
+[ ] Number Alignment: Do the numbers cited in the main text match the numbers in the charts/tables 100%?
+
+[ ] Actionable Landing: Do the final recommendations include specific actions, expected goals, and execution owners? (Avoid "correct nonsense" like "we need to strengthen management").