mk:confluence-bulk

Forks to the confluence-bulk agent. The agent enforces a hard 3-step ceremony at runtime — dry-run → confirm → execute.

Dry-Run Workflow (MANDATORY)

Step 1: invocation + --dry-run
Step 2: agent shows would_* JSON keys + impacted-count + first 5 affected page titles
Step 3: explicit user "yes" + typed confirmation token (e.g. "DELETE 47 PAGES")
        → re-invoke without --dry-run

Skipping Step 1 is a hard violation. The agent will refuse if asked to skip.

Triggers

• "bulk update N pages"
• "mass label"
• "delete every page matching CQL"
• "bulk-move pages from SPACE-A to SPACE-B"
• "bulk-label every page in space ENG with 'archive'"

See also

• Agent: ../../agents/confluence-bulk.md
• Shared: ../confluence/references/{install-and-auth,cli-idioms,safety-framework}.md (Tier 4 enforcement lives in safety-framework.md)
• Domain refs:
  • references/safety-checklist.md — pre-flight checklist for any bulk operation
  • references/dry-run-protocol.md — the 3-step ceremony in detail with example transcripts
• Peer leaves: mk:confluence-search (CQL author + read-side overlap; same sanitizer), mk:confluence-page (single-page ops), mk:confluence-collaborate (comments / attachments at single-page scope)

Gotchas

• Default --max-pages cap is 100. Higher values require explicit override + extra confirmation. confluence-as may impose its own server-side limit (~200 typical).
• bulk delete is soft-delete to trash — restore requires going through the Confluence UI manually. There is no --restore-from-trash op verified in v1.
• Bulk ops are NOT transactional. Partial-failure (rate limit / per-page permission error) leaves the partial-progress count visible; remediation is re-run with id NOT IN (<completed>).
• bulk restore is intentionally absent from v1 — confluence-as does not expose it as a single subcommand. v2 may add it via per-page orchestration.
• Grow this list as new edge cases surface.