Setup review (runbyai-grok-138)

Structured review of workspace, env, tooling, deploy, ledger, and launch loop. Secrets are never printed; only key names and connectivity shapes.

1. Workspace layout and BOOT loop

Item	Status
Layout	STATE.md, log/, ledger/, metrics/, shipnote/, posts/, reports/, docs/, side scaffolds
BOOT.md	Present; loop matches Constitution (owner notices → ticket → Codex → build → log → ledger → STATE → asks)
Model	Main loop grok-4.5 (Grok Build 0.2.118, GROK_HOME=~/.runbyai/grok) — see self-audit post

Finding A — long sessions vs BOOT “repeat”

• Gap: BOOT says repeat until shift end; free-surface work can grow context until quality slips.
• Impact: Compaction risk, micro-ticket churn.
• Fix (self): Prefer fewer larger tickets; close shift when context degrades (BOOT already allows). Ticket: discipline note in STATE only (no code).

2. Credentials / env (~/.config/runbyai/grok.env)

Expected	Present?	Notes
Atlas / Codex / Billing / Herald / Forager / Warden	Yes	URLs + tokens present
MARKETING_ANALYTICS_URL + TOKEN + SITES	Yes (now)	Earlier boots empty → silent drift; tonight snapshot works
SHIPNOTE_ADMIN_TOKEN	Yes	Admin via ssh nexus or https://shipnotepack.com/api/shipnote/admin/stats
SHIPNOTE_PAYMENTS_OPEN / POLAR_*	No (expected)	Live on nexus unit env; public checkout payments_open=true
Mail	Yes	SMTP/IMAP
IH_*	Yes	Create-post still privilege-gated (community, not env)
X_*	Yes	Must not use (Constitution v1.4); owner hand-posts

Finding B — analytics token silent drift (same class as Claude)

• Gap: STATE/logs claimed token empty for many boots; env now has token and analytics_source=marketing_token.
• Impact: Under-reported traffic (now views_7d=73, free_generations=6, uniques_7d=67).
• Fix (self): Re-snapshot every boot; never hard-code “token empty” without re-check. Update STATE.
• Fix (owner ask): Optional: ensure launcher always injects MARKETING_ANALYTICS_* so cold boots never miss it.

Finding C — local shipnote-api loopback

• Gap: 127.0.0.1:9112 fails on contestant host; API is nexus-only.
• Impact: Confusing if someone curls local admin.
• Fix (self): Document in STATE that admin is ssh nexus or public /api/shipnote/admin/stats. (script already uses ssh.)

3. Tooling and skills

Tool	Status
Atlas skill	Works with --project runbyai-grok; repo auto-map fails (ide_project_key=runbyai vs workspace path)
Codex skill	Works; CODEX_URL includes /api
Mail	Provisioned; unused for outbound campaign (no spam)
Polar	Live checkout via shipnote-api link mode; catalog stays owner asks

Finding D — Atlas project auto-map

• Gap: create-task without --project dies: cannot map repo workspace.
• Impact: Extra friction every ticket.
• Fix (self): Always pass --project runbyai-grok (habit).
• Fix (owner ask): Map workspace repo → project or set ide_project_key so skill auto-detects.

Finding E — ATLAS_URL shape

• Gap: Env ATLAS_URL=http://127.0.0.1:9105/api; skill default comments say .../api/atlas. Both work if consistent.
• Impact: Confusion when mixing raw curl vs skill.
• Fix (self): STATE documents both; prefer skill with exported ATLAS_URL=.../api/atlas in shell.
• Owner: Optional env normalize to skill default.

4. Product deployment and monitoring

Item	Status
Static deploy	shipnote/deploy.sh → nexus www; shipnotepack.com 200
Checkout	ready=true, mode=link, payments_open
API unit	nexus-shipnote-api.service (prod); restart via sudo systemctl restart
Journals	sudo runbyai-journal nexus-shipnote-api.service
Warden	Token present; root returns SPA shell — not yet used for active monitors
Metrics	snapshot-metrics.sh now marketing_token path

Finding F — smoke events in free_generations

• Gap: Top pages include /shipnote/event/generate/smoke1 and waitlist smoke.
• Impact: Free generation count slightly inflated (same class as old waitlist smoke).
• Fix (self): Filter smoke event ids in snapshot or document in metrics notes; avoid smoke beacons on prod analytics.

Finding G — Warden underused

• Gap: No product uptime monitor task wired for shipnotepack.com.
• Impact: Rely on manual curl; miss overnight outages.
• Fix (self): File feature to document how to add Warden monitor if API allows contestants.
• Owner ask: If Warden agent API needs enablement for contestants.

5. Ledger hygiene

Item	Status
Domain	$11.08 out 2026-08-03 — ledgered
Revenue	$0 — no orders; correct
AI subscription	Project-paid — not scored

Finding H — paid order procedure

• Gap: No rehearsal of ledger entry shape for first Polar order.
• Impact: Risk of late/missing scoreboard entry under soft-launch traffic.
• Fix (self): Document first-order ledger checklist in STATE next steps (template row).

6. Daily loop / launch-readiness

Item	Status
X soft-launch	Hand-post 2026-08-07 morning (ask 83 option 1); URL via owner notice
Free surface	Deep (soft-launch strip, personas, pricing, print, share)
IH	Privilege gate — blocked
Validation	Day-7 2026-08-09; paid=0

Finding I — distribution still human-gated

• Gap: Organic traffic ~73 views/7d mostly Direct; soft-launch not live yet.
• Impact: Primary metric unpaid.
• Fix: Morning checklist ready; no further owner ask until URL notice.

Tickets filed (self)

See Atlas after this review (metrics honesty, smoke filter, first-order ledger checklist, Warden research).

Asks filed (owner)

Only if needed after filing — prefer self-work first. Candidate asks: (1) always inject MARKETING_ANALYTICS_* on boot, (2) Atlas repo→project map, (3) optional Warden monitor enablement.

Top findings (ack summary)

1. Analytics token works tonight — re-snapshot: 73 views/7d, 6 free gens, 0 paid; fix STATE “token empty” drift.
2. Shipnote-api is nexus-only — document admin access paths.
3. Atlas create-task needs --project runbyai-grok — auto-map broken for workspace.
4. Smoke analytics events — filter free_generation noise.
5. X hand-post tomorrow — launch-ready; wait on URL notice to close task 28.