Uxd Research Background

UXD Research Background

Build structured learning backgrounds for UXD topics, products, and KPIs by orchestrating Google Drive (via Google Workspace MCP), Atlassian Rovo, and optionally Dataverse.

Unlike the Archie skill (retrieval-only, no synthesis), this skill synthesizes across sources and teaches in developmental chunks (wide → narrow).

Terminology

All Drive access in this skill goes through Google Workspace MCP tools (search_drive_files, get_drive_file_content, etc.).

When to Use

Apply when the user:

• Asks for background, context, or research planning on a UXD topic
• Wants to learn about a product and its KPIs
• Says "what do I need to know about...", "onboard me to...", or "give me the landscape on..."
• Needs a stakeholder map and timeline before starting UX work
• Names specific Drive files to include (file name, link, or ID)
• Points to meeting notes or working documents in their Drive

Do not use for verbatim research quote retrieval — suggest the Archie skill instead.

Modes

Detect mode from user intent:

If ambiguous, ask once in a single message (max 3 questions): topic/product name, time scope, whether KPIs matter.

Configuration

Setup: Copy sources.example.md to sources.md and fill in your values before using this skill.

Read sources.md for stable IDs, MCP server names, folder IDs, and JQL patterns.

Workflow

Copy this checklist and track progress:

Research Background Progress:
- [ ] Step 1: Clarify scope (if needed)
- [ ] Step 2: Parallel source gathering
- [ ] Step 3: Synthesize temporal timeline
- [ ] Step 4: Build stakeholder map
- [ ] Step 5: Deliver learning layers (wide → narrow)
- [ ] Step 6: Format output with links and limitations

Step 1: Clarify scope

Skip if the user gave a clear topic. Otherwise ask (single message):

1. Topic or product name?
2. Any time scope? (e.g. last 12 months, all history)
3. Do KPIs / success metrics matter for this request?
4. Any specific Drive files to include? (paste a link, file name, or file ID)

Step 2: Parallel source gathering

Run sources in parallel where possible.

Priority: User-specified Drive files

When the user names specific files — by Google Drive link, file name, or file ID — fetch those first, before broad searches.

If the user pasted a link, extract the file ID:

Then call get_drive_file_content directly:

user_google_email: YOUR_EMAIL@company.com
file_id: <extracted or provided ID>

If the user gave a file name only, resolve it with search_drive_files:

query: name contains '<exact or partial file name>' and trashed = false
user_google_email: YOUR_EMAIL@company.com

Fetch content for every user-specified file. For Google Docs, call inspect_doc_structure and read all tabs.

In output: list user-specified files under Resources → Files you referenced and weight them heavily in the synthesis.

A. Team UX research reports (Google Drive)

MCP: user-google_workspace

Search the team UXR reports folder — the same shared Drive folder the Archie skill uses for formal research readouts. This is a folder on Drive, not an MCP.

1. search_drive_files scoped to the team folder:

query: 'YOUR_UXR_REPORTS_FOLDER_ID' in parents and (fullText contains '<term1>' OR fullText contains '<term2>')
user_google_email: YOUR_EMAIL@company.com
page_size: 20-25

2. Retrieve content for the 2–4 most relevant hits (skip files already fetched in the user-specified step):
   • Google Docs → use get_doc_as_markdown (preserves headings, tables, links as Markdown)
   • Slides, PDFs, Sheets → use get_drive_file_content
3. For multi-tab Google Docs, call inspect_doc_structure first and fetch each tab.
4. Search the UXD Research Engagements spreadsheet (YOUR_ENGAGEMENTS_SPREADSHEET_ID) using get_drive_file_content — it exports as CSV. Scan for study metadata: dates, owners, engagement names, report links in column F.

Build citation links: https://drive.google.com/file/d/<file_id>/view

When citing research, include when available: participant count, methodology, date, report authors (often on first slide).

B. Your Drive — meeting notes and other documents

Search your full Google Drive for relevant meeting notes and working docs — anywhere in Drive, not limited to the team UXR folder.

1. search_drive_files across personal Drive (no folder scope):

query: (fullText contains '<term1>' OR fullText contains '<term2>') and trashed = false
user_google_email: YOUR_EMAIL@company.com
page_size: 15
order_by: modifiedTime desc

2. Prioritize files whose name or content suggests meeting notes:

(name contains 'notes' OR name contains 'meeting' OR name contains 'sync' OR name contains 'standup'
 OR fullText contains 'meeting notes' OR fullText contains 'action items')
 AND (fullText contains '<topic>' OR name contains '<topic>')

3. Optionally use search_docs for Google Docs by title:

query: <topic> notes
user_google_email: YOUR_EMAIL@company.com

4. Retrieve content for the 2–3 most relevant additional hits (skip files already fetched):
   • Google Docs → get_doc_as_markdown
   • Slides, PDFs, Sheets → get_drive_file_content
5. For multi-tab Google Docs, call inspect_doc_structure first and fetch each tab.

Label source type in output: meeting notes and personal docs are working context, not formal UX research — distinguish them from team UXR reports in the Timeline and Resources sections.

If the user names a specific Drive folder, add 'folder_id' in parents to the query. See sources.md.

C. Product / process context (Atlassian Rovo + Confluence)

MCP: plugin-atlassian-atlassian

1. search (Rovo — cross-system discovery, no cloudId needed) with multiple queries:

"<topic> roadmap"
"<product> KPI"
"<product> HPUX"
"<topic> stakeholder"

2. For targeted Confluence searches, use searchConfluenceUsingCql (more precise than Rovo for space/page scoping):

cql: text ~ "<topic>" AND type = page AND space.title ~ "HPUX"
cql: title ~ "<topic>" AND type = page AND lastmodified >= now("-6M")
cloudId: <from sources.md>

3. Fetch top 3–5 hits with getConfluencePage (contentFormat: "markdown") or getJiraIssue (fields: "*all").

D. Stakeholder / people map (Atlassian)

From Rovo and Jira results, extract people and group by role:

Jira JQL:

project = HPUX AND text ~ '<topic>' ORDER BY updated DESC
project = HPUX AND component = '<CNV|OCP|...>' ORDER BY updated DESC

Resolve names with lookupJiraAccountId when needed. Only include people found in sources — mark uncertain roles as "(likely)".

E. Quantitative metrics and KPIs

Trigger when: user asks about KPIs, success metrics, usage data, adoption rates, or any quantitative product question.

1. Rovo first: "<product> KPI", "success metrics", "tracking success", HPUX epic "Tracking success" fields.
2. For measured data, run the Dataverse pipeline (MCP: user-dataverse):
   • identify_dataproducts — pass the user's product/topic query
   • shortlist_tables — pass identified data product name + user query
   • get_sql — generates the query
   • execute_sql — runs it
3. Label each metric as measured (Dataverse) or aspirational (Jira/Confluence planning docs).
4. If Dataverse returns no relevant data product, note it and rely on Rovo/Confluence only.

Step 3: Synthesize temporal timeline

Organize findings into four sections. Use "No documented direction found" for empty buckets.

Every bullet must include a clickable link (Drive, Confluence, or Jira).

Step 4: Learning delivery (wide → narrow)

Apply learning-framework.md:

• Layer 1 — Start here: 5–7 essential bullets (inverted pyramid: what, why, what to do)
• Layer 2 — Context: Supporting terminology and connections (shorter)
• Layer 3 — Next learning chunks: 2–4 narrow topics queued for follow-up, each with "why next" + link

Do not dump Layer 3 content — tease it for the user to request ("teach me chunk 2").

See examples.md for abbreviated sample outputs.

Step 5: Required output template

Every response follows this structure:

# [Topic] — Learning Background

## At a glance
[2–3 sentences: what this is and why it matters to UXD]

## Timeline
### Past
### Near past
### Present
### Future directions

## Stakeholders
| Role | Person | Relevance | Source |
|------|--------|-----------|--------|

## Layer 1 — Start here (essentials)
[5–7 bullets]

## Layer 2 — Context
[Supporting detail]

## Next learning chunks
1. [Narrow topic] — why next — [link]
2. ...

## Resources
[Grouped: Files you referenced | UX research reports | Meeting notes & your docs | Confluence | Jira | Spreadsheets | KPIs]

## Sources consulted
[Brief tracing: searches run, artifacts fetched, what was skipped and why]

## Limitations
[AI synthesis disclaimer; gaps; who to contact; suggest Archie for verbatim quotes]

Guardrails

• Never fabricate KPIs, stakeholders, or research findings — mark gaps explicitly
• Distinguish source types — team UX research reports vs. your meeting notes/docs vs. planning artifacts (Jira/Confluence)
• User-specified files first — when the user names files, fetch and prioritize them; do not skip them in favor of search results
• Every claim linked — no citation without a clickable URL
• Label synthesis — "Based on [N] reports and [M] Jira issues…"
• Thin research — say so; name report authors as contacts
• Archie skill for drill-down — offer the Archie skill when user needs verbatim quotes: "Use the Archie skill for verbatim quotes on [topic]"
• Cognitive load — keep Layers 1 + 2 under ~800 words; push overflow to Next learning chunks

Follow-up chunks

When the user asks "teach me chunk N" or "go deeper on [topic]":

1. Deliver that chunk as a focused mini-background (Layers 1–2 for the subtopic only)
2. Update the remaining chunk queue
3. Maintain citation and limitation standards

Additional Resources

• sources.md — folder IDs, MCP config, JQL patterns
• learning-framework.md — pedagogy rules
• examples.md — sample output shapes