Publish Analysis

publish_analysis

Overview

This skill publishes a completed data analysis as a structured document in DataHub's hierarchical knowledge base. Before saving, the agent discovers the org's existing document structure and asks the user how widely the analysis should be shared.

Instructions

Step 1 — Discover the org's document strategy

Call search_documents with query "Analysis" to find existing analysis documents. Look at the results and identify any naming patterns or folder structure the org already uses (e.g. "Analyses / Reports / Q1-2024").

If no existing analysis documents are found, note that you will create a default hierarchy:

• Shared → Analyses → Private / {Your Name} (private)
• Shared → Analyses → Teams (team-shared)
• Shared → Analyses → Reports (org-wide)

Step 2 — Ask the user about visibility

Before saving, ask the user:

"Should this analysis be saved privately (only visible to you), shared with your team, or published globally for the whole org?"

Map their answer to the visibility parameter:

Step 3 — Prepare the document body

Structure the analysis body in markdown using this template:

## Summary
<2–3 sentence overview of what was analysed and the top finding>

## Key Findings
- <finding 1>
- <finding 2>
- <finding 3>

## Methodology
<describe the approach: what tables were queried, what logic was applied,
any filters or date ranges used>

## SQL
```sql
<the key query or queries used>

Data Sources

### Step 4 — Collect related dataset URNs

From prior `search` / `get_entities` results, collect URNs of the datasets
that were queried or referenced. Pass these as `related_dataset_urns` so
DataHub links the document back to the relevant assets.

### Step 5 — Call publish_analysis

Call the tool with:
- `title`: clear descriptive title, e.g. "Q1 2024 Revenue by Region"
- `body`: the markdown document prepared in Step 3
- `visibility`: value from Step 2
- `related_dataset_urns`: list from Step 4 (empty list if none)
- `topics`: optional tags, e.g. `["revenue", "q1-2024", "finance"]`

### Step 6 — Report back

After the tool returns, tell the user:
- Whether it succeeded
- The document URN (so they can find it in DataHub)
- Where it was saved (which folder in the hierarchy)