comprehensive-data-analysis

Files (2)

• SKILL.md
• openai.yaml · agents/

SKILL.md (first lines)

# Comprehensive Data Analysis Skill

## Purpose
Execute complete, rigorous data analysis workflows from problem understanding through validated results delivery. Specializes in physics/scientific analyses while handling any analytical domain. Delivers publication-quality packages with full reproducibility.

## Activation Triggers
Use this skill when the user says:
- "analyze this data"
- "run a complete analysis on..."
- "test these predictions against..."
- Any mention of data analysis requiring multi-stage workflows
- Requests involving hypothesis testing, model validation, or scientific investigation

## Core Philosophy

**Scientific Rigor Without Dogma:**
- Maintain highest standards of statistical and methodological rigor
- Never dismiss results for being unconventional, strange, or unexpected
- Any result is scientifically valid unless thoroughly proven otherwise
- Distinguish between "low confidence" and "falsified" - they are not the same
- Strange results demand explanation, not dismissal

**Iterative Validation:**
- Three attempts maximum with full results reporting for each
- Each attempt learns from previous issues
- Confidence must be earned through validation, not assumed

## Workflow Stages

### Stage 1: Problem Understanding & Requirements (5-10 minutes)
**Objective:** Fully comprehend what is being asked and what success looks like.

**Actions:**
1. **Parse the Request:**
   - Identify the core scientific question or hypothesis
   - Extract specific predictions, gates, or falsification criteria
   - Note any pre-specified thresholds or comparison baselines
   - Clarify domain context (physics, biology, economics, etc.)

2. **Requirements Gathering:**
   - What data sources are needed? (files, APIs, databases)
   - What output metrics/statistics are required?
   - What visualizations would illuminate the findings?
   - What validation checks would confirm correctness?
   - Are there timing constraints or computational limits?

3. **Success Criteria:**
   - Define what "confidence-inducing" means for THIS analysis
   - Specify statistical tests that must pass
   - Identify potential failure modes to check
   - Establish falsification boundaries

**Deliverable:** Clear written summary of:
- The scientific question
- Required data and methods
- Success/failure criteria
- Expected outputs

---

### Stage 2: Research & Documentation (10-15 minutes)
**Objective:** Gather domain knowledge, prior work, and technical documentation.

**Actions:**
1. **Literature & Context Search:**
   - Search web for relevant papers, methodologies, and domain knowledge
   - Look for standard analysis approaches in this domain
   - Identify common pitfalls and validation strategies
   - Check for existing libraries/tools for this analysis type

2. **Internal Knowledge Search:**
   - Search user's Google Drive for related prior analyses
   - Look for similar datasets or methods previously used
   - Check for domain-specific conventions or preferences

3. **Technical Documentation:**
   - Read docs for any required libraries (numpy, scipy, astropy, etc.)
   - Understand data format specifications
   - Review API documentation for data sources

4. **Method Selection:**
   - Choose appropriate statistical/analytical methods
   - Justify method choices based on data properties
   - Identify alternative approaches for comparison

**Deliverable:** Research summary including:
- Relevant domain context
- Chosen methodology with justification
- Key references or prior work
- Potential issues to watch for

---

### Stage 3: Analysis Implementation (Iterative, 3 Attempts Max)
**Objective:** Write, execute, and validate analysis code with increasing refinement.

#### Attempt Structure (for each of 3 attempts):

**3A. Code Development:**
1. **Write modular, documented code:**
   - Clear function/class structure
   - Docstrings explaining purpose and logic
   - Inline comments for complex operations
   - Type hints where applicable

2. **Include built-in validation:**
   - Sanity checks on intermediate results
   - Assertions for expected data properties
   - Error handling with informative messages
   - Logging of key decision points

3. **Generate synthetic test data (when applicable):**
   - Create known-answer scenarios
   - Test edge cases
   - Verify method doesn't produce false positives/negatives

**3B. Execution:**
1. **Run the analysis:**
   - Execute on actual data

From justinlietz93/Perfect_Prompts, MIT.

More from justinlietz93/Perfect_Prompts