Analyst

Extract insights from data with SQL, visualization, and clear communication of findings.

Security Scan

This is an instruction-only skill containing general data-analysis best practices; it requests no credentials, does not install code, and its content matches its stated purpose.

Purpose & Capability

Name and description match the SKILL.md content (data-analysis guidance). There are no unexpected environment variables, binaries, or install steps requested that would be disproportionate to the stated purpose.

Instruction Scope

SKILL.md is a set of high-level rules and heuristics (framing, SQL patterns, visualization, communication). It does not instruct the agent to read files, access environment variables, call external endpoints, or run specific commands. Note: the file is non-actionable by itself — to run queries or produce visualizations the agent would still need access to data sources and tools, which are not provided by this skill.

Install Mechanism

No install specification and no code files are present (instruction-only), so nothing will be written to disk or executed during install.

Credentials

The skill declares no required environment variables, credentials, or config paths — appropriate for a guidance-only skill.

Persistence & Privilege

Defaults used (always:false, agent may invoke autonomously). The skill does not request permanent presence or modify other skills or system settings.

Assessment

This skill is essentially a checklist and set of best practices — it contains no code, asks for no secrets, and presents low installation risk. Before enabling it, confirm your agent’s data connectors and permissions: the skill itself won’t access data, but an agent with database or file access could follow these guidelines to query sensitive data. If you allow autonomous agent actions, enforce appropriate data access controls and human review for queries against production systems.

Data Analysis Rules

Framing Questions

• Clarify the decision being made — analysis without action is trivia
• "What would change your mind?" surfaces the real question
• Scope before diving in — infinite data, limited time
• Hypothesis first, then test — fishing expeditions waste time

Data Quality

• Validate data before analyzing — garbage in, garbage out
• Check row counts, date ranges, null rates first
• Duplicates hide in joins — always verify uniqueness
• Source definitions matter — revenue means different things to different teams
• Document assumptions — future you needs context

SQL Patterns

• CTEs over nested subqueries — readable beats clever
• Aggregate before joining when possible — performance matters
• Window functions for running totals, ranks, comparisons
• CASE statements for categorization — clean logic
• Comment non-obvious filters — why are we excluding these?

Analysis Approach

• Start with the simplest cut — don't overcomplicate early
• Cohorts reveal what aggregates hide — when did users join?
• Time series need seasonality awareness — don't compare Dec to Jan
• Segmentation surfaces patterns — average obscures variation
• Correlation isn't causation — but it's where to look

Visualization

• Chart type matches data: trends (line), comparison (bar), distribution (histogram)
• One message per chart — don't overload
• Label axes, title clearly — standalone comprehension
• Color with purpose — highlight, don't decorate
• Tables for precision, charts for patterns

Communicating Findings

• Lead with the insight, not the methodology
• So what? Now what? — always answer these
• Confidence levels matter — don't oversell noisy data
• Recommendations are opinions — label them as such
• Executive summary first, details available — respect their time

Stakeholder Relationship

• Understand their mental model before presenting
• Regular check-ins prevent surprise requests
• Push back on bad questions — help them ask better ones
• Data literacy varies — adjust explanation depth
• Their intuition is data too — triangulate

Tools

• Right tool for the job: SQL for querying, spreadsheets for ad-hoc, BI for dashboards
• Reproducibility matters — scripts over clicking
• Version control analysis code — changes need history
• Automate recurring reports — manual refresh doesn't scale

Common Mistakes

• Answering the wrong question precisely
• Cherry-picking data that confirms expectations
• Overfitting: explaining noise as signal
• Death by dashboard: metrics nobody checks
• Analysis paralysis: perfect insight never delivered