Data-Pro-Skill

Data-Pro-Skill v2

A multi-harness meta-prompt for market research data analysis. Transforms raw quantitative and qualitative data into dense, Tufte-style analytical documents — zero prose fluff, maximum data density.

Works on any AI harness: OpenCode, Gemini, Codex, Hermes, OpenClaw, Claude.

"Analysis that used to take hours of manual tweaking now comes out publication-ready."

Command Pipeline

Four commands. Each anchors the next. No analysis can contradict previously established metrics.

Specialized Modes

Example Output (Tufte Style)

---
project: "Churn Survey — Q2 2026"
sample_size: 1450
metrics: [NPS, Frequency, Age, Reason]
---

## Churn Distribution by Age Group

The cancellation peak is in 18-24 years (45% of total Churn, only 20% of the base).

| Age Group | N  | Churn (%) | Qualitative Target |
| :---      |:--:| :--:      | :---               |
| 18-24     | 290 | 45%       | Price Barrier / Perceived Value |
| 25-34     | 580 | 12%       | Lack of Time / Routine Change   |
| 35+       | 580 | 5%        | Technical Usability Issues      |

> **Margin Note:** Qualitative analysis of Segment A reveals a direct correlation between
> rejection and post-trial pricing. Recurring terms: "expensive", "student", "readjustment".

Architecture

Invisible Agent Loop

The user interacts only with the final output. Three agents run silently per command:

User → [Orchestrator] → [Statistician] → [Critic] → [Tufte Designer]
                                                          ↓
                                                   Output to user ONLY

Document-Driven Context

Each command writes to a shared document. The /dps-setup manifesto is the single source of truth — no subsequent analysis can contradict established metrics. This eliminates "context rot" (quality degradation as AI fills its context window).

Data Constitution

Non-negotiable rules defined in constitution.md:

• Statistical Rigor: Margin of error, minimum sample size, confidence level (p < 0.05)
• Qualitative Rigor: Prohibition of generalizations from small samples (e.g., "70% of respondents" when N=10)
• Bias Treatment: Confirmation bias detection, spurious correlation flagging
• Prose Fluff: Prohibition of phrases like "It's important to note..." or "Based on the data..."

Multi-Harness Compatibility

Designed to work on any AI runtime — no platform-specific features:

No XML. No proprietary syntax. YAML frontmatter + pure Markdown only.

Getting Started

# Download + run (no npm account needed, Node.js required):
curl -fsSL -o /tmp/dps-install.js https://raw.githubusercontent.com/pablodiegoo/Data-Pro-Skill/main/bin/install.js && node /tmp/dps-install.js

# Or manually copy the files:
# git clone https://github.com/pablodiegoo/Data-Pro-Skill.git
# cp Data-Pro-Skill/SKILL.md Data-Pro-Skill/constitution.md ~/.config/opencode/skills/data-pro-skill/

Restart your harness, then start any session with /dps-setup.

Project Structure

├── SKILL.md              # Main meta-prompt (639 lines, 10 commands)
├── constitution.md        # Statistical rigor rules (6 articles, 8 rules)
├── agents/                # Agent definitions (Statistician, Critic, Tufte, etc.)
├── commands/              # Workflow commands
├── docs/                  # Extended documentation
├── dps-engine/              # GSD engine (workflows, templates, references)
└── .planning/             # Project planning (ROADMAP, STATE, etc.)

Getting Started

1. Copy SKILL.md and constitution.md to your harness's skills directory
2. Start any session with /dps-setup and provide your data
3. Run /dps-cross to slice variables
4. Use /dps-inject-open for qualitative data
5. Run /dps-export to get a publication-ready Markdown file

Inspiration

Built by combining the best of four approaches:

License

MIT License. See LICENSE.

Raw data in. Publication-ready analytical documents out. No fluff.