setting-up-data-catalog — Claude Code skill: data catalog setup
setting-up-data-catalog fills a project's data catalog: canonical metric definitions, trust marks on warehouse tables, and reviewed join proposals.
no license 182 installs ★ 101 updated 2026-10-24
npx skills add PostHog/ai-plugin --skill setting-up-data-catalog -a claude-codeWhat setting-up-data-catalog does
The skill keeps a per-project inventory of three things: what a metric means, which of several similar tables can be trusted, and how tables join. Your data stays where it is. Reads go through SQL against system.information_schema; writes go through the data-catalog MCP tools.
Setup runs top-down. It proposes certifications for the tables your team leans on and deprecation for stale or duplicate copies, samples candidate keys with posthog:execute-sql before any join is proposed, and builds metrics out of the project's most-used insights, linking each to its source_insight_short_id so later drift can be detected.
Nothing an agent writes counts until a person approves it. Promotion tools come in pairs: the -prepare call shows a confirmation message, you type confirm, then the -execute call takes the returned hash. Without the data-catalog MCP tools there is no write path at all — plain SQL only reads.
What it helps with
- certify the warehouse tables the team actually relies on
- measure a join key's match rate before proposing the relationship
- turn a heavily used insight into a canonical metric
- clear the review queue of pending certifications and joins
- catch a metric that drifted from its source query
Example prompts for the agent
- Seed the data catalog: propose certifications for the main tables and the joins between them
- Show the pending relationship proposals with their match-rate evidence and wait for my confirmation
- Create a metric from the project's most-used insight and record its grain
How to install setting-up-data-catalog
1Skills CLI (Vercel)
npx skills add PostHog/ai-plugin --skill setting-up-data-catalog -a claude-code2Manually: put the files in place
~/.claude/skills/setting-up-data-catalog/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/setting-up-data-catalog/
1Skills CLI (Vercel)
npx skills add PostHog/ai-plugin --skill setting-up-data-catalog -a codex2Manually: put the files in place
.agents/skills/setting-up-data-catalog/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add PostHog/ai-plugin --skill setting-up-data-catalog -a cursor2Manually: put the files in place
.cursor/skills/setting-up-data-catalog/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install PostHog/ai-plugin skills/setting-up-data-catalog2Manually: put the files in place
.github/skills/setting-up-data-catalog/SKILL.md1Skills CLI (Vercel)
npx skills add PostHog/ai-plugin --skill setting-up-data-catalog -a gemini-cli2Platform CLI
gemini skills install https://github.com/PostHog/ai-plugin --path skills/setting-up-data-catalogManually: put the files in place
.devin/skills/setting-up-data-catalog/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add PostHog/ai-plugin --skill setting-up-data-catalog -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about setting-up-data-catalog
How do I install setting-up-data-catalog in Claude Code?
Run npx skills add PostHog/ai-plugin --skill setting-up-data-catalog -a claude-code. The skill file lands at ~/.claude/skills/setting-up-data-catalog/SKILL.md.
What does setting-up-data-catalog do?
It builds and curates the semantic layer: canonical metrics, certifications on warehouse tables and views, and table relationships backed by measured evidence.
Why don't the agent's changes apply right away?
Everything an agent writes is stored unapproved. Promotion is a two-step call: -prepare returns a confirmation message, and after you type confirm, -execute applies it with the returned hash.