Already executed: 18 cleared, 6 conversations, 5 users, read-back clean.

Why

The vendor-bucket cleanup only covered todo. A recommendation carries the same misrouted content as a task , same wrong object, same customer named in a vendor's notification , and surfaces in the agenda's recommendation section.

So after Xander's six PandaDoc tasks were cancelled, six PandaDoc recommendations were still sitting there, all overdue:

Mark fully-executed deals Closed Won in CRM , WZMH A…     08-21
Approve Utile order form in PandaDoc , unblocks Xand…     08-24
Follow up with Matt Bedee re: Middough Executive Sum…     08-25
Mark Geis Companies as Closed Won in CRM , order for…     08-31

Those would have reappeared on Monday's agenda looking exactly like the problem he reported.

<email>'s "Small Business Loans" is newly caught: 286 inbound, 285 from DocuSign.

Safety

The predicate is unchanged and is what makes this safe: >= 5 inbound, >= 70% from a single vendor sender, and zero inbound from a real counterparty. Nothing holding a genuine customer email is touched , the same test that correctly spared a conversation carrying four PandaDoc notices and one message from the actual customer.

Test added pinning that both statuses are covered, so a later edit can't silently narrow it back.

🤖 Generated with Claude Code

https://claude.ai/code/session_01NVHnt2pEEjhVwSyhkcaxoC

Greptile Summary

This PR extends the vendor-bucket cleanup to remove both open tasks and recommendations while retaining the existing conversation qualification predicate.

• Counts both todo and recommended rows during candidate selection.
• Marks both statuses as agent_deleted during an applied cleanup.
• Verifies that neither status remains after mutation.
• Adds regression coverage and records the completed cleanup outcome.

Confidence Score: 4/5

The implementation appears safe to merge, with a non-blocking weakness in the new regression test’s ability to detect partial status-filter regressions.

The cleanup consistently handles both target statuses, but the added test checks duplicated source literals rather than the behavior or consistency of all four filters.

Files Needing Attention: apps/server/tests/scripts/cancel-vendor-bucket-tasks-selection.test.ts

Important Files Changed

Flowchart

### Issue 1
apps/server/tests/scripts/cancel-vendor-bucket-tasks-selection.test.ts:59-60
**Status test misses regressions**

These assertions only verify that each status-list literal appears so