reference approval queue · SkyRun memory

Top section of brief.josephbowens.com (2026-04-22+)

The PWA now shows an Approvals collapsible as the FIRST thing you see — above the Morning Brief. Any pending items from any of four outreach channels surface here with deep-links to the native tool where approval happens.

URL: https://brief.josephbowens.com
Auto-opens if there are pending items; stays collapsed when queue is clear.

Four independent channels

Emoji	Channel	Source skill	State file	Deep-link target
✉️	Direct drafts	`gmail-deep-scan` (Step 5b)	`pending_drafts.jsonl`	Gmail draft URL
📧	SmartLead	`daily-beenverified-enrichment` (future wiring)	`pending_smartlead_batches.jsonl`	SmartLead campaign URL
📬	Postcards	`grand-county-property-scout` (future wiring)	`pending_postcard_batches.jsonl`	Postcard workbook / Jasmine draft
🤝	Realtor intros	`nightly-consolidation` (future wiring)	`pending_realtor_intros.jsonl`	Gmail search URL for Rachel's intro thread

HubSpot Sequences are NOT a channel — decommissioned 2026-04-22. See project_hubspot_sequences_running.md.

File paths

State files: ~/Library/Application Support/SkyRun/pwa/data/pending_*.jsonl
Schema doc: ~/Library/Application Support/SkyRun/pwa/data/_schema.md
Renderer: /Users/josephbowens/Library/Application Support/SkyRun/build_pwa.py → read_approvals() + render_approvals_section()

Item schema (canonical)

json
{
  "id": "unique-string",
  "channel": "direct_drafts" | "smartlead" | "postcards" | "realtor_intros",
  "title": "short summary < 80 chars",
  "context": "1-3 line elaboration",
  "deep_link": "URL to native tool",
  "created_at": "ISO8601",
  "urgency": "RED" | "YELLOW" | "INFO",
  "source_skill": "skill-name",
  "meta": { / channel-specific fields / }
}

Rendering rules

Sort: urgency DESC (RED → YELLOW → INFO), then oldest created_at first
RED items open by default via section auto-expand
Each item: title + context + urgency chip + optional "Xd pending" age chip + "Open in native tool →" button
Empty queue = renders "No pending approvals. Queue is clear."

Current state (as of 2026-04-22)

1 direct draft (Hadank Thursday walkthrough — awaiting Andy reply)
1 SmartLead batch (3 campaigns in DRAFTED state — Rachel sign-off pending)
0 postcards (no pending batch)
3 realtor intros (Leah Bishop, Megan Waymeyer, Katie Fox — all 43d pending from Rachel)

How source skills populate the queues (writer contract)

Each source skill is responsible for:
1. Appending new items when it detects a pending state (e.g., gmail-deep-scan sees a new draft in Gmail → writes pending_drafts entry)
2. Removing items when the pending state is resolved (draft was sent, batch was activated, etc.)
3. Updating urgency based on age + content (e.g., draft to active-deal person > 24h = RED)

Cleanup rules:

Files are append-mostly but writers may rewrite to remove resolved items
Items older than 14 days are auto-archived to _archive.jsonl (stale queue cleanup)

Wiring status per source skill (all 4 wired 2026-04-22)

✅ gmail-deep-scan — wired via Step 5b. Maintains pending_drafts.jsonl on every run (9am/1pm/5pm/10pm).
✅ daily-beenverified-enrichment — wired post-Step 7b. Maintains pending_smartlead_batches.jsonl with a per-campaign state item (3 items: SWITCH/HOT+WARM/NURTURE+LOW). Removes items when campaign status becomes RUNNING.
✅ grand-county-property-scout — wired as Phase 7. Adds postcard-batch items to pending_postcard_batches.jsonl when new postcard targets are created. Auto-clears stale batches on next run.
✅ nightly-consolidation — wired via Section D2-B. Maintains pending_realtor_intros.jsonl based on Gmail insights + transcripts. Tracks the 4 realtor contacts (Leah, Megan, Katie, Whitney) through states: awaiting_rachel_intro → intro_sent → realtor_replied → live_engagement.

Architectural decision: smart inbox vs swipe-to-approve

v1 is smart inbox (deep links to native tools for native approval). No backend. v2 candidate (future): swipe-to-approve via Cloudflare Pages Functions + KV, where tapping "Approve" executes the action server-side. Held back until v1 is in daily use and the deep-link friction is the actual bottleneck.

How to apply

When asked "what's in my queue" / "what's pending" → check the 4 JSONL files or point to the PWA URL.
When adding a new outreach channel, extend APPROVAL_CHANNELS in build_pwa.py and create matching pending_*.jsonl.
When debugging "why did X not appear in my approvals?" → check if the source skill's wiring is done (see status above).