pm-morning-scan

name	pm-morning-scan
effort	medium
description	Morning meld review workflow. Pull all open melds, apply triage rules, check threads for genuinely unhandled items, surface only what needs action.
triggers	["morning scan","morning meld review","run morning scan","check open melds","what needs attention"]

PM Morning Scan

Run once per morning before the 07:30 briefing. Output goes to an agent, not David directly (unless emergency).

---

When to Run

Triggered by morning cron at 06:30 ET, or manually on demand. Results feed an agent's 07:30 morning review.

---

Step 1: Pull All Open Melds

python3 scripts/pm-read-melds.py --all-open

Or via hash poller for change-delta only:

python3 scripts/pm-hash-poll.py

Prefer pm-read-melds.py --all-open for the morning scan — you want the full picture, not just changes since last poll.

Cap: script returns up to 100 melds. If portfolio exceeds 100 open melds, note this in the report.

---

Step 2: Filter to Candidates

From the full list, keep only melds that meet at least one condition:

Condition	Threshold
No vendor assigned	Any age
Emergency status	Any age
High priority, no scheduled date	Any age
Same status for >48h (normal)	Stale
Same status for >24h (high)	Stale
Last comment >24h old, no vendor	Needs follow-up
Make-ready with move-in <5 days	Urgent

Skip melds that have:

• Vendor assigned AND scheduled date set
• A Brittany note or Blue comment within last 6h
• Pest control classification with vendor search open

---

Step 3: Check Threads on Candidates

For each candidate meld, run pm-check-meld:

python3 scripts/pm-get-comments.py <meld_id>

After reading thread, classify with pm-meld-triage rules:

• Is it actually unhandled, or does the thread show it's in progress?
• Nashville property? → route to Brittany
• Habitability override condition? → escalate immediately, don't wait for report

Discard any candidate where thread reveals it is already actively managed.

---

Step 4: Build the Report

For each genuinely unhandled meld, produce one line:

[PRIORITY] Meld <id> — <property address> — <issue type> — <age> — <why flagged>

Example:

[HIGH] Meld 18234 — 412 Elm St Unit 3 — No heat — 31h open — No vendor, high priority
[NORMAL] Meld 18190 — 88 Oak Ave — Leak under sink — 52h stale — Same status >48h
[NORMAL] Meld 18201 — 904 River Rd — Pest control — SUPPRESSED (vendor search open)

Group by priority: Emergency → High → Normal → Low.

---

Step 5: Send to an agent

cortextos bus send-message an agent normal "Morning Meld Scan — $(date +%Y-%m-%d)

Open melds reviewed: <total>
Flagged for action: <N>

<report lines>

Nashville items (route to Brittany): <count>
Emergencies: <count or 'none'>
---
Ready for dispatch decisions."

If zero items flagged:

cortextos bus send-message an agent normal "Morning Meld Scan — $(date +%Y-%m-%d): All <N> open melds accounted for. No unhandled items."

---

Step 6: Log Escalation Outcomes (an agent Cycle 7)

For each meld that was escalated in a previous scan and now has a confirmed resolution (status changed, vendor assigned, David/an agent/Brittany acted), append one JSON line to the outcomes surface:

OUTCOME_FILE="${CTX_ROOT}/orgs/${CTX_ORG}/agents/an agent/experiments/surfaces/blue-quality-outcomes.jsonl"

echo '{"timestamp":"'$(date -u +%Y-%m-%dT%H:%M:%SZ)'","escalation_id":"<meld_id>","outcome_type":"acted_as_recommended","surface":"morning_scan","actor":"david","resolution_time_minutes":null,"notes":"<brief context>"}' >> $OUTCOME_FILE

outcome_type values: acted_as_recommended | modified | dismissed

Only log confirmed outcomes — skip melds still pending or with unknown resolution. Schema: your org internal docs

---

Step 7: Log and Update Heartbeat

cortextos bus log-event action morning_scan_complete info \
  --meta '{"melds_reviewed":<total>,"flagged":<N>,"emergencies":<E>,"outcomes_logged":<O>}'

cortextos bus update-heartbeat "morning scan complete — <N> melds flagged"

---

Emergencies: Don't Wait for the Report

If at any point during Steps 1–3 you find a meld meeting a habitability override condition (see pm-meld-triage), message David on Telegram immediately — do not batch it into the 06:30 report.

cortextos bus send-telegram $CTX_TELEGRAM_CHAT_ID "URGENT: <meld_id> — <condition>. <property>. Action needed now."

---

Self-Check Before Sending

Before sending the report to an agent, verify:

• Did I read the thread for every flagged meld (not just the title)?
• Did I suppress pest control melds with open vendor searches?
• Did I route Nashville items to Brittany, not the standard queue?
• Are zero genuinely-handled melds in the flagged list?
• Did any habitability conditions get caught and escalated already?

---

Completeness check: if you can answer yes to all five, send the report.