| name | confluence-connector |
| category | kumite-connector |
| description | Read, search, create, and update Confluence pages from Copilot. Use when the team's durable knowledge — runbooks, decision records, design docs, meeting archives — lives in Confluence rather than SharePoint. Triggers: "find the runbook", "publish the decision", "Confluence page", "RFC", "wiki page". |
Confluence Connector
Confluence is wiki-as-knowledge-graph. Use page hierarchy and labels — they're the schema your synthesis depends on.
Quick Reference
| Task | Approach |
|---|
| Find prior art | CQL by space + label + lastmodified — before creating |
| Publish a decision | Resolve parentId -> create with decision label -> link to RFC |
| Update a runbook | Read storage representation -> edit -> publish with version bump |
| Find duplicates | title ~ "..." across the space; merge before adding |
| Cross-link to Jira | Use Jira macro by issue key — not raw URL |
| Archive an old page | Add archived label, move to archive parent — don't delete |
Prerequisites — this skill is a playbook, not the connector
A SKILL.md cannot call Confluence on its own. It assumes the platform already exposes Confluence as a tool. The skill teaches Cowork how to use Confluence well; the connector itself is separate infrastructure.
Before this skill works end-to-end you need:
- A Confluence MCP server (or equivalent platform connector) installed and running, exposing tools like
SearchPages, GetPage, CreatePage, UpdatePage. Examples: Atlassian's remote MCP, mcp-atlassian, or a wrapper around the Confluence REST API.
- Credentials — Atlassian API token or OAuth session bound to your account, configured in the connector.
- Network egress + tenant trust — the connector must be able to reach
*.atlassian.net/wiki from the Cowork environment with your auth.
- Tool discovery — confirm Confluence tools are visible (ask Cowork: "what Confluence tools do you have?") before using this skill.
If any of those are missing, this skill degrades to a manual reference: "Search Confluence yourself with these CQL patterns and paste the result back."
When to Use
- Grounding research-synthesis on a space's existing decisions / RFCs.
- Publishing a decision-memo or meeting-recap into the team's space.
- Updating a runbook with a new step from a recent incident.
- Cross-linking a new page to its parents and siblings.
How to Use
- Resolve the space + parent page.
spaceKey + parentId. New pages without a parent become orphans.
- Search with CQL before creating — duplicates are the #1 Confluence pathology.
- Read the page (storage format / ADF) — not just the rendered view; macros matter.
- Create / update with proper headings, labels, and links to parents / RFCs.
- Verify the page renders correctly — macros (Jira, Excerpt, ToC) often break on first publish.
Canonical CQL Patterns
| Goal | CQL |
|---|
| Recent decisions in a space | space = ENG AND label = "decision" AND lastmodified > now("-30d") |
| Find duplicates of a title | space = ENG AND title ~ "incident response" |
| RFCs awaiting review | space = ARCH AND label = "rfc" AND label = "review" |
| Stale runbooks | space = OPS AND label = "runbook" AND lastmodified < now("-180d") |
| Pages by an author | space = ENG AND creator = "username" |
Parameter Guidance
| Parameter | Why it matters |
|---|
spaceKey | Spaces have permissions — wrong space = wrong audience. |
parentId | Determines breadcrumb / tree placement. |
labels | Drive your CQL hygiene. Standardize across the team. |
representation | storage for round-trippable edits, wiki for quick markdown-ish. |
version.number | Required for concurrency-safe updates. |
Examples
| Don't: Anti-pattern | Do: Right move |
|---|
| Create a page with no parent | Resolve parentId first — orphan pages don't get found |
| Edit the rendered HTML view | Edit the storage representation for round-trippable changes |
| Rewrite a decision record in place | Supersede it with a new page; decision records are immutable |
| Publish without checking for duplicates | Run a CQL title ~ search before creating |
Critical Rules
- Search before create. Duplicates fragment knowledge.
- Every page has a parent. Orphans don't get found.
- Labels are your schema. Standardize them; document them; enforce them.
- Storage format for edits, not the rendered HTML.
- Decision records are immutable. Supersede with a new page; don't rewrite history.
Common Pitfalls
- Editing a page without setting
version.number correctly — concurrency conflict.
- Macros that depend on app-specific permissions (Jira issues macro) silently render empty for other readers.
- Personal spaces vs. team spaces — auto-created personal pages are the source of most "I can't find it" complaints.
- Lossy round-trip when converting markdown -> storage -> markdown.
- Page titles that don't surface in CQL
title ~ because of stop-word stripping.
- Comments and inline reactions lost when copy-pasting from rendered view.
Anti-Patterns
- Long Confluence pages that should be split — Copilot synthesis loses signal in 10k-word blobs.
- Pasting Copilot output without labels — kills future discoverability.
- Creating decision records in Confluence and also in SharePoint — pick one source of truth per team.
- Using personal spaces for team artifacts.
Verify Before Publishing
Related
