data-stack

Lightweight Claude Code (+ others) skill suite for data analysts and data engineers, built for use with the Upsolve AI MCP.

9 focused skills. Minimal dependencies. Structured workflows for data exploration, quality, anomaly investigation, and KPI reporting.

Quick Start: Your First 5 Minutes

1. Install data-stack (30 seconds, see below)
2. Open your project in Claude Code with the Upsolve MCP connected
3. Run /dstack-setup to verify the connection
4. Run /dstack-advisor if you're not sure where to start
5. Run /dstack-explore to map your data

Install — Takes 30 Seconds

Install on your machine

Open Claude Code and paste this. Claude does the rest.

Install data-stack: clone https://github.com/Upsolve-Labs/data-stack.git to ~/.claude/skills/data-stack (try HTTPS first, fall back to SSH with git@github.com:Upsolve-Labs/data-stack.git if auth fails), then run cd ~/.claude/skills/data-stack && ./install.sh. The script links skills and prints an INSTALL_STATUS report and NEXT_STEPS. Do NOT install anything yourself — read the status, then walk the user through each missing tool one AskUserQuestion at a time. Follow the NEXT_STEPS in the output.

Upload to Claude (claude.ai Skills)

Each skill is available as an individual zip on the Releases page. Download the skill zip you want and upload it via Customize → Skills → Create skill → Upload a skill.

Available skill zips: dstack-explore.zip, dstack-investigate.zip, dstack-profile.zip, dstack-compare.zip, dstack-funnel.zip, dstack-pipeline-health.zip, dstack-metric-brief.zip, dstack-advisor.zip, dstack-setup.zip, dstack-upgrade.zip

Install via zip (no-git environments)

Download the latest bundle zip from the Releases page, then unzip and run the installer:

# Download latest zip
curl -L https://github.com/Upsolve-Labs/data-stack/releases/latest/download/data-stack-latest.zip -o data-stack-latest.zip

# Unzip into skills directory and install
unzip data-stack-latest.zip -d ~/.claude/skills/
cd ~/.claude/skills/data-stack && ./install.sh

Codex, Gemini CLI, or Cursor

data-stack uses the SKILL.md standard. Clone the repo, then copy the skills into your tool's skill directory:

git clone https://github.com/Upsolve-Labs/data-stack.git ~/.claude/skills/data-stack

# Codex
cp -r ~/.claude/skills/data-stack/skills/* .agents/skills/

# Cursor
cp -r ~/.claude/skills/data-stack/skills/* .cursor/skills/

# Gemini CLI
cp -r ~/.claude/skills/data-stack/skills/* .gemini/skills/

Skills are plain markdown — they work in any agent that reads SKILL.md files.

See It Work

you:    I want to understand why revenue dropped last Tuesday
you:    /dstack-investigate
claude: [confirms the anomaly, narrows by segment, tests 4 hypotheses, surfaces root cause with numbered findings]

you:    /dstack-explore
claude: [maps all available tables, profiles row counts and schemas, outputs a data map with recommended next steps]

you:    /dstack-profile orders
claude: [audits nulls, duplicates, distributions, date ranges — outputs an OK/WARN/FAIL scorecard]

you:    /dstack-funnel
claude: [asks for steps, shows drop-off at each stage, identifies the biggest opportunity, segments by device/channel]

you:    /dstack-metric-brief
claude: [shows current value, WoW/MoM trend, top segments, outliers — formatted for Slack or a doc]

Workflow

/dstack-explore → /dstack-profile → /dstack-investigate (if anomaly found)
/dstack-explore → /dstack-compare or /dstack-funnel   (for targeted analyses)
/dstack-pipeline-health                  (standalone, on-demand or scheduled)
/dstack-metric-brief                     (standalone, for reporting)

Skills

Skill	What It Does
/dstack-explore	Discover available tables, map schemas, profile row counts and freshness. Start here.
/dstack-investigate	4-phase anomaly root cause analysis. Confirm → narrow → hypothesize → validate. Numbered findings.
/dstack-profile	Data quality audit: nulls, cardinality, duplicates, distributions. OK/WARN/FAIL scorecard.
/dstack-compare	Period-over-period or A/B comparison. Breaks down delta by dimension to find what drove the change.
/dstack-funnel	Funnel drop-off analysis. Conversion rates and absolute counts per step, segmented by dimension.
/dstack-pipeline-health	Pipeline SLA check: freshness, row count anomalies, schema drift. Per-table health dashboard.
/dstack-metric-brief	KPI brief: current value, WoW/MoM trend, top segments, outliers. Shareable Slack/doc format.

Independent skills:

Skill	What It Does
/dstack-advisor	Asks what you're trying to do and recommends the right skill.
/dstack-setup	Verify Upsolve MCP connection and data-stack installation.
/dstack-upgrade	Pull latest data-stack with backup.

Dependencies

Tool	Required?	Why	Install
Upsolve MCP	Yes	All data skills use it to run queries, return results, and maintain conversation threads.	Connect via Claude Code MCP settings
gh	Optional	Useful if you contribute to this repo or build on top of it.	macOS: brew install gh

Upgrade

/dstack-upgrade

Or manually: cd ~/.claude/skills/data-stack && git pull

Philosophy

• One thread per analysis. Upsolve conversations build context — reuse thread_id within a session.
• Confirm before concluding. Always verify an anomaly exists before investigating its cause.
• Structured findings. Numbered codes (1A CONFIRMED, 1B CANDIDATE, 2A RULED_OUT) make results auditable.
• No bloat. 10 skills, ~400 lines total. No build system, no templates, no binaries.

License

MIT License. Free and open source.