mk:confluence-collaborate

Forks to the confluence-collaborate agent. The agent prefers footer comments over inline; asks before posting an inline comment.

Triggers

• "add comment to page 12345: '<text>'"
• "list comments on page 12345"
• "upload screenshot.png to page 12345"
• "label page 12345 as 'urgent'"
• "remove label 'draft' from page 12345"
• "watch page 12345" / "stop watching page 12345"
• "list watchers of page 12345"

Examples

• Footer comment: "add comment to 12345: 'awaiting backend fix'"
• Attachment upload: "attach /tmp/repro.mp4 to page 12345"
• Label add: "label 12345 as 'rfc'"
• Watcher list: "who watches page 12345?"

See also

• Agent: ../../agents/confluence-collaborate.md
• Shared: ../confluence/references/{install-and-auth,cli-idioms,safety-framework}.md
• Domain refs:
  • references/comment-formats.md — inline vs footer, markdown vs storage format
  • references/attachment-conventions.md — file size limits, allowed types, download paths
• Peer leaves: mk:confluence-page (page CRUD; create-with-attachment), mk:confluence-bulk (bulk-label across 10+ pages), mk:confluence-spec-analyst (downloads attachments for image analysis)

Gotchas

• validate_file_path in confluence-as rejects .. traversal but delegates to opaque assistant-skills-lib. Don't trust delegation alone — the agent independently validates upload paths at the agent boundary too. [from research]
• Confluence Cloud comments do NOT have a strict internal/public flag like Jira Service Management. Inline vs footer is the closest semantic boundary — agent prefers footer. [from jira-collaborate parallel pattern]
• Inline comment requires an anchor selection. If the selection text doesn't match unique content on the page, the comment fails with a 400-class error.
• Bulk-label across 10+ pages goes through mk:confluence-bulk, NOT this skill. Single-page label add/remove only here.
• Grow this list as new edge cases surface.