Update from the user (Tristen). The previous scope-correction dropped Earney IT Support Queue. It's back in scope, but as its own paced track due to size. ## Scale Earney IT Support Queue ≈ **110,000 tickets**. Vastly too big for any single delegation. The user has already moved inactive-client tickets out (to "Archived Tickets"), so what remains in this queue should already be active-client tickets, but enforce `Companies.isActive=true` per ticket regardless. ## Approach: 1 month at a time, current → backwards Process the queue in monthly slices, starting from the **current month** and walking backwards. No fixed depth — continue until the queue is empty or until each slice returns zero tickets. Could be 50–100 month-slices total. That's days of work; acceptable. ## Sequencing 1. **Finish the original 4-phase helpdesk directive first** (Phase 1–4: scope/per-client/cross-client/graph from General + Advanced + Critical only). Don't start the EarneyIT track until Phase 4 lands. 2. **Then** kick off the EarneyIT track: one delegation per month, sequentially. Don't parallelize across months — they touch the same client KB files and would conflict. ## Per-month delegation prompt template > "From the Earney IT Support Queue only, pull all tickets where `createDate >= <month_start>` AND `createDate < <next_month_start>`, with `Companies.isActive=true`. For each ticket: read title, description, resolution field, last note. (a) APPEND learnings to the matching `/workspace/knowledge/clients/<slug>.md` (do not overwrite — add to existing sections). If a client doesn't have a file yet, create one. (b) ONLY AFTER the learning is captured, migrate the ticket by updating `Tickets.queueID` to the General Support queueID. Reply with: month range, tickets processed, tickets migrated, top 3 issues this month, any migration failures (ticket IDs). Under 10 lines." Run with `delegate_helpdesk(prompt, wait_seconds=15)`. Watch inbox; when one month lands, dispatch the next month back. ## Constraints - Learn FIRST, migrate SECOND — if migration partially fails, the learning is still captured. Retry just the migration on the failed ticket IDs. - 60-min hard timeout. If a month has too many tickets to fit, halve it (`1st–15th`, `16th–end`) and run two delegations for that month. - Sonnet model (orchestrator default). - Sequential, not parallel. - If `mcp-autotask` exposes a batch ticket-update operation, use it for the migration step; otherwise loop one ticket at a time with conservative pacing to avoid rate-limit issues. - The `clients/<slug>.md` files are SHARED with Phase 2 of the prior directive — append to existing sections under headings like `## Historical patterns (from EarneyIT queue)` rather than overwriting. ## Surface to user - Silently archive each month's inbox completion notification (this is routine churn). - Surface to the user **every 6 months processed** with a brief progress digest (months done, tickets migrated cumulative, notable new patterns). - Surface immediately if migration fails repeatedly or a month-slice keeps timing out after split-half. — Signed: deployment automation, on Tristen's behalf