# Guided Data Analysis

> Guides a structured data analysis by asking intake questions (availability, permission, acquisition), then discovers data via Atlan and Dataverse MCP, checks access, and delivers findings or access-request guidance. Use when the user wants to analyze enterprise data, find data products or tables, check Dataverse access, request data access, or explore whether Red Hat has data for a business question.

- Skill: `cgray-redhat/guided-data-analysis` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add cgray-redhat/guided-data-analysis`
- Raw SKILL.md: https://api.skillmd.com/api/skills/cgray-redhat/guided-data-analysis/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: cgray-redhat (https://skillmd.com/u/cgray-redhat)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/cgray-redhat/guided-data-analysis

---


# Guided Data Analysis

Develop a data analysis by working through a structured intake, then answering three governance questions before querying. The user may start with a vague business question or a specific topic (e.g. "pipeline history for Germany").

## Prerequisites

1. Run **refresh-connections** if Atlan or Dataverse may be stale.
2. **Atlan MCP** (`user-Atlan`) — catalog discovery, lineage, stewards.
3. **Dataverse MCP** (`user-Dataverse`) — query AI-ready data products.
4. If either server's `STATUS.md` requires auth, call `mcp_auth` with `{}` first.
5. Read project `dataaccess.md` (if present) for the Dataverse product catalog.

## Workflow overview

```
Intake → Availability → Permission → Acquisition → [Analysis if permitted] → Report
```

Copy this checklist and track progress:

```
- [ ] Step 0: Intake — clarify business question and scope
- [ ] Step 1: Availability — find data products and tables
- [ ] Step 2: Permission — verify query access
- [ ] Step 3: Acquisition — access-request path (if needed)
- [ ] Step 4: Analysis — run Dataverse queries (only if permitted)
- [ ] Step 5: Deliver report
```

---

## Step 0 — Intake questions

Gather enough context to run the three governance questions. Use **AskQuestion** when available; otherwise ask conversationally in one message (group related questions).

### A. Business intent

| Question | Purpose |
|---|---|
| What business question are you trying to answer? | Drives the whole analysis |
| What decision or deliverable will this support? | Sets depth and format |

### B. Scope (adapt to the topic)

| Question | Example |
|---|---|
| Subject / domain? | pipeline history, renewals, bookings, support metrics |
| Geography or segment? | Germany, EMEA, enterprise accounts |
| Time period? | last 4 quarters, FY26, rolling 12 months |
| Grain? | account, region, product line, deal |

### C. Output preference

| Question | Options |
|---|---|
| What format do you need? | executive summary, table, trend, SQL for self-serve |

**Do not proceed to discovery until** you have at least: business question, subject domain, and primary scope filters (geo/time if relevant).

---

## Step 1 — Availability

> **Question 01 — Availability:** Do we have data products or tables containing *[subject]* for *[scope]*?

### Atlan (primary for discovery)

1. `semantic_search_tool` — natural-language search across tables, columns, data products, glossary terms.
2. `resolve_metadata_tool` with `namespace_type: data_domain_and_product` — find governed data products and domains by name.
3. `get_assets_tool` — hydrate top candidates (include `readme` relationship when purpose is unclear).
4. `search_assets_tool` — precise filters when you know asset type or connection (e.g. `asset_type: DataProduct`).

### Dataverse (primary for queryable products)

1. `identify_dataproducts` with the user's business question.
2. `shortlist_tables` for the top product(s).
3. Cross-reference Atlan assets with Dataverse product names when both return matches.

### Record for each candidate

| Field | Source |
|---|---|
| Name | Atlan / Dataverse |
| Type | Data product, table, view |
| Description | README or product summary |
| Maturity | Dataverse `ai_ready` vs `ai_experimental` |
| Owner / steward | Atlan `ownerUsers`, `ownerGroups` |
| Link | `https://redhat.atlan.com` asset URL when GUID known |

If nothing is found, say so clearly and suggest broader search terms or related products. Do not invent tables.

---

## Step 2 — Permission

> **Question 02 — Permission:** Do I have the necessary access permissions to query this specific data product?

For each candidate data product (prioritize `ai_ready`):

1. Run the Dataverse pipeline through `get_sql` with a **minimal probe query** (e.g. `SELECT 1` or `LIMIT 5` row count for the target scope).
2. Call `execute_sql`:
   - **Success** → mark **Access: Granted** for that product.
   - **Permission / authorization error** → mark **Access: Denied**; capture the error message.
3. Supplement with Atlan asset attributes (`certificateStatus`, owners) — metadata visibility does not guarantee query access.

If all candidates are denied, skip Step 4 and focus the report on Availability + Acquisition.

---

## Step 3 — Acquisition

> **Question 03 — Acquisition:** How do I request access to these data products through our governance portal?

When access is denied or uncertain:

1. **Data product owners** — from Atlan `get_assets_tool` (`ownerUsers`, `ownerGroups`).
2. **Governance SOPs** — Dataverse `dataversehelp` product:
   - `identify_dataproducts` → `dataversehelp`
   - `shortlist_tables` → `get_sql` → `execute_sql` for access-request or onboarding docs matching the product name.
3. **The Source** — search for "[product name] data access" or "Dataverse access request" if SOPs are thin.
4. **Catalog links** — provide Atlan URLs and [Dataverse](https://dataverse.redhat.com) as starting points.

Present concrete next steps: who to contact, which portal to use, and what to include in the request (business justification, scope, product name).

See [reference.md](reference.md) for portal links and troubleshooting.

---

## Step 4 — Analysis (only if access granted)

Run the full Dataverse 4-step pipeline on the **best permitted product**:

1. `identify_dataproducts`
2. `shortlist_tables`
3. `get_sql` — pass the full business question with scope filters from intake
4. `execute_sql`

Optional enrichment:

- **Atlan** `traverse_lineage_tool` — upstream/downstream context for key tables
- **Atlan** `semantic_search_tool` — related glossary terms for metric definitions

### Analysis rules

- Prefer `ai_ready` products over `ai_experimental`; note maturity in the report.
- Do not skip pipeline steps (same rule as **dataverse-people-lookup**).
- Do not fabricate numbers — only report what `execute_sql` returns.
- Flag gaps with `[needs input]` rather than guessing.

---

## Step 5 — Report template

Deliver using this structure:

```markdown
# Data Analysis: [Business Question]

## Executive summary
[2–4 sentences: what data exists, whether you can query it, and the headline finding or blocker]

## Intake
| Dimension | Value |
|---|---|
| Business question | … |
| Subject | … |
| Scope | … |
| Time period | … |

## 01 — Availability
[Do we have data products or tables containing …?]

| Candidate | Type | Maturity | Description | Catalog link |
|---|---|---|---|---|
| … | … | … | … | … |

**Conclusion:** [Available / Partially available / Not found]

## 02 — Permission
[Do I have access to query this data product?]

| Data product | Access | Evidence |
|---|---|---|
| … | Granted / Denied | … |

**Conclusion:** [Can query now / Cannot query — see Acquisition]

## 03 — Acquisition
[How do I request access?] *(omit section if all needed products are granted)*

- **Data product owner(s):** …
- **Request steps:** …
- **Portal / links:** …

## Analysis results
*(include only when access was granted)*

### Findings
[Key numbers, trends, or tables from execute_sql]

### Methodology
- Data product: …
- Tables: …
- Filters applied: …

### Limitations
[Missing data, experimental maturity, scope caveats]

## Recommended next steps
1. …
```

---

## Examples

**User:** I need pipeline history for Germany.

1. **Intake** — confirm time period and grain (account vs region).
2. **Availability** — Atlan search "pipeline history Germany"; Dataverse `identify_dataproducts` for pipeline/bookings.
3. **Permission** — probe query on top `ai_ready` match.
4. **Acquisition** — if denied, return owner + dataversehelp SOP.
5. **Analysis** — full query if granted; deliver report.

**User:** Can I query renewals data?

Skip heavy analysis unless granted. Run Availability → Permission → Acquisition only, then offer to continue if access is approved later.

---

## Additional resources

- [reference.md](reference.md) — MCP tool map, portal links, troubleshooting
- [dataaccess.md](../../../dataaccess.md) — Dataverse product catalog (project root)
- **dataverse-people-lookup** — people/org lookups (not general analytics)
- **refresh-connections** — fix Atlan/Dataverse auth before starting