| name | cowork-with-wiki |
| description | Use when the user works with or mentions Confluence wiki, product docs, documentation pages, or knowledge base content |
| argument-hint | [search term or page title] |
| allowed-tools | Read, Glob, Grep, Bash, Agent, AskUserQuestion |
Wiki Workflow
Confluence wiki workflow for Claude Code via the Atlassian Rovo MCP server.
Prerequisites
Requires globally installed Atlassian MCP server (named atlassian). If not set up, run /cowork-with:cowork-with-onboarding.
Product Documentation Root
- Space: ONEKEY
- Root page ID: 489095189
For product doc queries, navigate from root via getConfluencePageDescendants.
Write Safety
Before ANY write operation to an existing page (`updateConfluencePage`, `createConfluenceFooterComment`, `createConfluenceInlineComment`), the `explorer` agent MUST first read the target page using `getConfluencePage` and verify the page creator. If the page was NOT created by the current user, REFUSE the write and explain why. Only pages created by "me" (the current authenticated user) may be modified. This applies to ALL write operations on existing pages without exception.
New page creation skips the creator check (current user is the creator), but always confirm with the user before creating.
Tool Strategy
ALL MCP read tool calls MUST go through the `explorer` agent. The main model MUST NOT call any MCP read tool directly — always spawn the `explorer` agent to do it. This is mandatory because the explorer agent enforces correct parameter types and names.
Spawn via: Agent tool → name: "explorer" (the plugin ships agents/explorer.md — haiku model, read-only: MCP reads + codebase + git).
| Task | Who |
|---|
| Read pages, search, list spaces | explorer agent |
| Browse page tree (descendants) | explorer agent |
| Read comments (footer, inline) | explorer agent |
| Verify page creator before writes | explorer agent |
| Draft page content | Main model |
| Preview to user and get confirmation | Main model |
| Write to Confluence (create, update, comment) | Main model |
MCP Tools
Read (explorer agent):
getConfluencePage — get a page by ID, body returned as Markdown
getConfluencePageDescendants — list child pages under a parent
getConfluencePageFooterComments — list footer comments on a page
getConfluencePageInlineComments — list inline comments on a page
getConfluenceCommentChildren — list replies to a comment
getConfluenceSpaces — list available spaces
getPagesInConfluenceSpace — list pages in a space
searchConfluenceUsingCql — search using CQL (Confluence Query Language)
search — cross-service Atlassian search (searches Confluence + Jira together)
Write (main model, after creator check + user confirmation):
createConfluencePage — create a new page (Markdown body)
updateConfluencePage — update existing page (title, body, location)
createConfluenceFooterComment — add a footer comment
createConfluenceInlineComment — add an inline comment
Update Page Flow
1. `explorer` agent → Read page to get current content AND verify creator (getConfluencePage)
2. HARD-GATE → If creator != current user, REFUSE and stop
3. Main model → Draft updated content
4. Main model → Preview changes to user for confirmation
5. Main model → Apply update after user approves (updateConfluencePage)
Create Page Flow
1. `explorer` agent → Read parent page to verify location (getConfluencePage)
2. Main model → Draft page title and content
3. Main model → Preview to user for confirmation
4. Main model → Create page after user approves (createConfluencePage)
Comments on existing pages follow the same gate as page updates: verify creator first, then confirm with user.
Before creating or updating page content, you MUST use `AskUserQuestion` to present a structured confirmation. Show title, parent page, and body content in the `markdown` preview field. Options: "Submit" (Recommended), "Edit", "Cancel". Do NOT call any Confluence write tool without explicit user approval via this confirmation. This is NOT optional — text-based confirmation does NOT satisfy this gate. You MUST call the `AskUserQuestion` tool.
CQL Quick Reference
| Goal | CQL |
|---|
| Pages in a space | space = "ONEKEY" AND type = page |
| Search by title | space = "ONEKEY" AND title ~ "search term" |
| Pages containing text | space = "ONEKEY" AND text ~ "search term" |
| Recently modified | space = "ONEKEY" AND lastModified > now("-7d") |
| Pages I created | space = "ONEKEY" AND creator = currentUser() |
| Pages I modified | space = "ONEKEY" AND contributor = currentUser() |
Cross-Skill Integration
- If product documentation is missing or outdated, suggest creating a Jira issue via
/cowork-with:cowork-with-jira with type Task and label documentation.
- When reading a wiki page, use the
explorer agent to search Jira for related issues via searchJiraIssuesUsingJql.
- If a wiki page references Figma designs, use
/cowork-with:cowork-with-figma to extract design specs for implementation.
Limitations
The following operations are not available via MCP and require the Confluence web UI:
- Page permissions management
- Page labels / tags
- Attachments (uploading files to pages)
- Page templates
- Page history / version comparison
- Space administration
- Macros and dynamic content
- Page restrictions