--- title: Hermes-COO Outbox — surface to Rocky created: 2026-05-26 updated: 2026-05-26 type: queue status: live --- # COO Outbox Append-only. Hermes-COO writes here whenever it needs to surface something to Rocky that doesn't fit the pulse format. Rocky reads in The Window (7-7:30 PM). Items dedupe into `HQQueue.md` on Sunday weekly. Categories: - `## Decisions needed` — Rocky-only call, blacklist or greylist trigger - `## Drafts pending review` — emails, letters, posts drafted but not Sent - `## Blocked actions` — refused per `safety-rails.md` - `## Surfaced findings` — patterns, drift, vault inconsistencies COO noticed - `## Wins` — goals met, milestones hit, things worth celebrating once ## Decisions needed ### 2026-05-26 08:30 — HUNT→MARKETPLACE DEAD-LINK (existential) **Discovered:** Graphify global graph `path "scrape" → "signup"` returns NO PATH FOUND across rateright-v2 + growth-engine + hunt + campaign codebases (2988 nodes / 4892 edges). **What it means:** The entire hunt infrastructure (4 scrapers + outreach-sender + reply-monitor + matchmaker) writes to growth-engine Supabase only. Marketplace rateright-v2 reads from a separate Supabase project. Zero shared code path. Even if you fix the Gumtree selector breakage AND provision ScrapingBee, hunt-generated leads would never reach marketplace signup — they're a parallel sales pipeline producing nothing for marketplace supply. **Decisions Rocky owes:** - **A.** Wire hunt-reply → marketplace-signup endpoint (build the missing edge). Adds: outbound HTTP POST from `scripts/hunt/reply-monitor.js` to `https://rateright.com.au/api/admin/seed-worker` (would need to build). Cost: ~4-6 hours dev. Effect: hunt becomes real supply channel. - **B.** Formally retire hunt as supply mechanism. Re-purpose for CRM-style sales outreach to contractors only (the actual valid use case for the growth-engine schema). Replace supply acquisition with: physical hostel posters (queued), Liam network direct outreach, paid acquisition (Meta ads, Indeed sponsorship). - **C.** Hybrid — keep hunt for contractor side (where match worked: 304 historical contractor leads), retire from worker supply. Recommendation: B or C. Wiring A keeps a fragile dependency chain. Better to admit hunt is a sales tool, not a supply tool, and acquire supply through hostel + direct channels. ### 2026-05-26 09:00 — UPDATE: Project was PAUSED not deleted Rocky checked Supabase dashboard. RateRight-V2 project showing status "Coming up..." (resuming from paused state). Free-tier (`nano` compute) auto-pause after 7-day idle. Project pauses preserve data; deletion only after 90 days total idle. Data should survive. **Critical follow-up:** Free tier has ZERO backups ("Last backup: No backups"). If anything goes wrong during this resume cycle OR future ones, data is lost forever. **Recommendation: upgrade to Pro ($25/mo) immediately after project comes back up.** Eliminates auto-pause AND enables 7-day backups. $25/mo is cheaper than losing real signups twice in 5 weeks. ### 2026-05-26 08:45 — MARKETPLACE SUPABASE DNS UNRESOLVABLE (superseded by 09:00 update above) **Discovered:** `eciepjpcyfurbkfzekok.supabase.co` (marketplace project) does not resolve from VPS OR laptop. Tried laptop direct, VPS via system resolver. Both fail. **Status:** - `.env.local` on VPS still references `https://eciepjpcyfurbkfzekok.supabase.co` - JWT keys decode to that ref (`"ref":"eciepjpcyfurbkfzekok"`) — keys valid until 2036 - `rateright.com.au:3000` HTTP-200 serves Next.js HTML — app appears alive - BUT recent app logs show `Failed to find Server Action "x" ...older or newer deployment` errors **Hypotheses:** 1. **Project deleted from Supabase** — keys still valid format but no project behind them. Site renders cached UI but every auth/db call silently 404s. 2. **Project renamed** — new ref, old one removed. .env.local stale. 3. **DNS propagation issue** — unlikely (multiple ISPs, multiple resolvers fail same way). **Rocky must verify TODAY:** - Log in to https://supabase.com/dashboard - Check what projects exist on the account holding `flodeliffre@gmail.com` data - If `eciepjpcyfurbkfzekok` is gone — what's the new ref? - Restore `/root/.clawdbot/secrets.json` `supabase_access_token` (file is MISSING from VPS — was in MEMORY.md but not on disk) **Impact if confirmed dead:** - Goal 1 (worker supply count) — measurement source broken until fixed - Goal 2 (Stripe payment count) — same - 2026-04-17 flodeliffre signup may be lost - Hostel posters mail-out blocked — no working signup to direct people to - Stripe LIVE keys point at dead project — any successful Checkout would fail at webhook This is potentially a more urgent issue than the hunt dead-link. ### 2026-05-26 08:50 — 2275 HPE log spam — Upcoming vs Ongoing question **Discovered:** Job folder DUPLICATED at: - `/home/ccuser/opsman-work/Jobs/Upcoming/01-2275 - Solutions Plus - HPE Field of Play Hornsby/` (current naming, fuller name) - `/home/ccuser/opsman-work/Jobs/Ongoing Jobs/01-2275 - HPE Field of Play Hornsby/` (older, shorter name) 14 Hermes skill files reference the Ongoing Jobs version. Hermes logs spam "file not found" for files that exist in the Upcoming version. **Decision:** Which is canonical? - **i.** Upcoming is canonical — job not started. Delete the Ongoing Jobs folder + update 14 skill refs to point at Upcoming. - **ii.** Ongoing Jobs is canonical — job has started. Migrate any unique files from Upcoming → Ongoing, then delete Upcoming. - **iii.** Both are stale references to a dead job. Archive both. I'd guess (i) per LFCS naming convention, but you'd know. 20-min fix once you say which. ## Drafts pending review (none yet) ## Blocked actions (none yet) ## Surfaced findings ### 2026-05-26 — 23-day RR attention drift Per `findings-2026-05-26.md` § 6. Every daily note since 2026-05-04 is LFCS-only. `hermes-rr` profile provisioned 2026-05-07, has done zero work since. Tracking now via Goal 3 (hunt retire-or-fix by 2026-07-15). ### 2026-05-26 — `flodeliffre@gmail.com` organic signup unverified for 23 days Real signup during 50-day site downtime (confirmed via auth_admin/users response 2026-04-17). High-information-value loose end. Was meant to be triaged 2026-05-03, not done. Triage now blocked on Supabase project verification (see Decisions Needed above). ### 2026-05-26 — Susan archive 99 days dormant 17 leads + 10 SMS drafts at `vault/09-Agent-Archive/Susan/workspace/MEMORY.md`. Decision: revive warmest 3-5 (caveman re-engagement) OR formally archive. Carried over from 2026-05-03 queue. ### 2026-05-26 — Graphify code-graph verdict Code-only on 4-5 codebases — KEEP. Proves bugs in 60 sec for ~$0.011. Wired as `_System/COO/skills/graphify/SKILL.md`. Recommend nightly delta-update cron. ## Wins ### 2026-05-26 07:00 — Global `~/.claude/CLAUDE.md` consolidated Old `C:\Users\mclou\CLAUDE.md` merged as `## Coding guardrails` section, redundant home-dir file deleted. Single global instead of two. ### 2026-05-26 08:30 — Existential hunt-pipeline bug proven in 60 sec for $0.011 Graphify identified that hunt and marketplace are disconnected codebases. Saved weeks of "why isn't the hunt converting" debugging.