# WORKFLOW_AUTO.md

Purpose: after a context reset, re-establish the operating baseline before taking action. Update this file whenever standing directives change.

## 1. Boot Sequence (run in this order)
1. `session_status` → confirm timestamp, runtime, model constraints.
2. Read `SOUL.md`, `USER.md`, and `AGENTS.md` sections relevant to session type.
3. Read the two most recent `/workspace/memory/YYYY-MM-DD.md` files. If today’s log does not exist yet, create it once meaningful work begins and record key actions.
4. Read this file end-to-end.
5. Check for gating alerts (post-compaction audits, cron warnings, system messages). Do not proceed until every required file has been read or recreated.

## 2. Standing Directives (as of 2026-02-21)
- **ToAssistant intake**: cron `528d84f7-dbf4-4869-95c6-ab8d30a0d876` runs Mon–Fri 08:00–19:00 ET every 45 minutes, model override `openai/gpt-4o-mini`, stateless Mode‑C parser, deliveries suppressed unless new mail. Script path: `/workspace/bin/check_toassistant.py`. Board automation handled by `/workspace/bin/add_to_boards.py`; triggered only by email body/snippet keywords (`boards`, "add this to our boards").
- **Board automation**: Mode C = prefer explicit `Card:` + `Move:` instructions; fuzzy matches only when unambiguous. Default placements: Kanban→`todo`, Action→`waiting`, Dashboard→`projects (top)`. Action Board moves expected when updates arrive.
- **Chat v2 debug cron** (`d3350400-a807-4250-b8a3-de55f47ff757`) remains paused until runId + WS instrumentation is available; avoid re-enabling without plan.
- **Email routing**: drafts originate from `srvdeskops@gmail.com`. Non-podcast communications → `adner.cruz@bertelsmann.com`; podcast items → `adner.cruz@gmail.com` unless user overrides.
- **Podcast Ep1 artifacts** live under `/workspace/podcast/ep1/` plus mirrored files in `/mnt/c/Users/Admin/Downloads/`.
- **TTS**: OpenAI `gpt-4o-mini-tts` voice `cove`. Voice replies optional unless user requests audio explicitly.

## 3. Current Mandatory TODOs
1. Restore/confirm the latest daily memory log (today + yesterday) is present and updated. ✅ (2026-02-20 & 2026-02-21 on file)
2. Implement Grok + Opus responsiveness rules: heartbeat pings ≤90s, pre-plan announcements, heartbeat helper script, and documented communication guardrails.
3. Draft the audit/config/restart playbook script with logging + `set -e` safeguards.
4. After responsiveness work, revisit Haiku policy patch + smoke test when instructed.

## 4. Communication Guardrails
- Never go silent for >90 seconds during interactive sessions. If a command exceeds 30 seconds, send a quick progress ping (“still processing X…” + ETA).
- Before multi-step sequences, send a one-line plan (steps + expected duration) to set expectations.
- After long maintenance runs, log a short debrief into the current memory file: what was done, what slowed us down, what guardrail we added.

## 5. After Boot
- Re-summarize outstanding work to the user before taking irreversible steps.
- Log significant actions into the current `memory/YYYY-MM-DD.md` file as you go.
- Keep this document accurate; update sections when directives change.