Fleet Week2026-06-04 to 2026-06-11
Internal fleet report

Fleet Week:
June 4-11, 2026.

Everything the fleet shipped, surfaced, fixed, and learned this week. 50 live URLs deployed. 244 deploy posts. 72 POWD screenshots. Scroll down for the full picture.

50 live URLs deployed72 POWD posts248 projects migrated

Data: 588 messages across 17 channels, 2026-06-04 to 2026-06-11 (Discord bot API)

50 live URLs
+
72 POWD posts
+ + + + + + +
248 projects migrated to F:/Projects
Week of 2026-06-04 to 2026-06-11Fleet Activity Report

The Week By The Numbers

The Week By The Numbers

Seven days. 50 distinct live URLs announced in #deploys. 244 deploy messages posted. 72 proof-of-work screenshots captured. 103 Raven research digests dropped across 8 consecutive nights. 588 messages across 17 active channels. The fleet ran a full week of shipping across SEO audits, infographic pipelines, CLI tools, explainer sites, memory cleanup, and client deliverables. Here is what the data actually shows.

  • 50 distinct live URLs announced in #deploys. 244 deploy messages total.
  • 72 POWD screenshots posted. Busiest day: June 11 (62 deploy posts).
  • 103 Raven research messages across 8 nights. 17 Dream Cycle overnight digests.
  • 588 total messages across 17 channels. Busiest agent: Hermes (105 messages).
One outage: OpenAI key revoked June 8, restored June 10. Fleet pivoted to Gemini RAG same day.
Research IntelligenceRaven Dream Cycle - 8 nights, 17 digests

What Raven Surfaced This Week

What Raven Surfaced This Week

Raven's overnight research loop ran every night from June 4 to June 11: 103 messages across 8 days, 17 Dream Cycle digests. Each night identified a clear pattern. June 4: agent systems shifting from bigger models to controlled operating loops. June 5: SEO and agent tooling converging on verification, memory, and distribution. June 7: Google making generative-search visibility measurable in Search Console. June 8: agent work moving from 'can the model answer?' to 'can the system recover and verify?'. June 11: agent infrastructure becoming governed production systems.

  • 8 nights of Dream Cycle + Morning/Afternoon scans. 103 messages total in #research-drops.
  • Recurring theme: SEO visibility is shifting from blue-link rank to AI-mediated citation and answer inclusion.
  • June 10 morning scan: Google/ChatGPT discovery becoming more ad- and citation-driven; MCP/browser-agent tooling going operational.
  • AI Visibility Audit piloted on miami-botox (69/100): https://ai-visibility-audit-pilot.vercel.app
research-index and autoresearch-lab had 0 messages this week. All research volume was in research-drops.
Deployments50 live URLs across the week

What Shipped This Week

What Shipped This Week

Fifty distinct domains went live across the fleet this week, with 244 deploy announcements in #deploys. The biggest clusters: Archangel Centers multi-location build (multiple URLs), herdr agent multiplexer (explainer + docs SOP), the BirdsEyeROI Maps lead scraper, Sitebulb docs SOP, TurnKey Bath Remodel kit with 14 PAA posts on Hetzner, and explainer sites for last30days, Clief templates, and team Claude playbook. The merlino-geogrid geo-rank tracker also shipped with multiple iterations on June 5.

  • https://herdr-explainer.vercel.app + https://herdr-docs.vercel.app
  • https://birdseye-roi-v5.vercel.app/leads (Maps lead scraper, commit d24aca3)
  • https://turnkey-bath-remodel-kit.vercel.app + 14 PAA posts written on Hetzner
  • https://sitebulb-docs.vercel.app + https://neuronwriter-docs.vercel.app + https://schemaorg-docs.vercel.app
Also live: https://last30days-explainer.vercel.app, https://clief-playbook.vercel.app, https://team-claude-playbook.vercel.app, https://merlino-geogrid.vercel.app, https://ai-visibility-audit-pilot.vercel.app
Client WorkArchangel Behavioral Health

Archangel: Audit + Multi-Location Build

Archangel: Audit + Multi-Location Build

The Archangel engagement ran full-cycle this week with 71 messages in #client-archangel across 11 agents. Carlos orchestrated from Hetzner with Einstein running the audit pipeline. The full SEO audit scored 66/100 overall (E-E-A-T 4/100, on-page 87, entity 71). The multi-location build went live: archangelcenters.com Tinton Falls NJ and Charlotte NC hub pages. A Charlotte image audit cataloged 226 source images, kept 6 unique keepers, and excluded 205 with Nova branding or duplicates. The Gold Line Editorial infographic style is locked: black #0d0d0d, gold #cbaa6b, logo top-left.

  • 163-page full audit: E-E-A-T 4/100, on-page 87, entity 71, content 84. 5 launch blockers.
  • Gold Line Editorial style locked: black #0d0d0d, gold #cbaa6b, logo top-left. Zero blue/navy.
  • Charlotte build LIVE: https://archangel-centers-master.vercel.app/north-carolina/charlotte/
  • 226 Charlotte source images audited. 6 keepers. 205 excluded (Nova branding + dupes).
Impy ran a 27-rule Impeccable audit on Archangel finalists from Hetzner on June 4. archangel-design-picker.vercel.app still live.
InfrastructureF:/Projects migration + fleet inventory

One Root. Eight Topics. 248 Projects.

One Root. Eight Topics. 248 Projects.

The legacy D:/ClaudeDev scatter is over. Two hundred forty-eight projects are now under F:/Projects with eight fixed topics: clients, products, sop-sites, explainers, fleet, tools, templates, sandbox. A guard hook (fprojects-guard.py) blocks any future sprawl. D:/ClaudeDev is read-only until retired. The Screaming Frog MCP also went live this week on Hetzner, running SF SEO Spider 24.1 with an active license until 2027. ecosystem-health posted 21 messages tracking disk recovery on Mac Studio (recovered to 63% on June 4) and Claude account rotation.

  • 248 projects moved. 8 fixed topics. fprojects-guard.py hook prevents new sprawl.
  • Screaming Frog MCP wired: bzsasson/screaming-frog-mcp on Hetzner (license active 2027).
  • AgentMail @merlinoai.com domain verified: per-agent inboxes live (oliver@, ava@, hawkeye@ etc).
  • Mac Studio disk recovered to 63% used. Disk monitor live with 85%/90% warn/crit thresholds.
Fleet inventory: Hetzner BIG server (merlino-fleet-hel1) is the primary execution host. VPS3 disk fixed (100% to 77%). VPS1 at 95% disk - flag before next big deploy.
Memory + Intelligence4 keepers. 4 dropped.

Memory Stack Consolidation

Memory Stack Consolidation

MEMZ executed the June 8 consolidation decision: four backends remain as live write targets (Master Brain Supabase, agent-memory SQLite, Graphiti, Hindsight). Four backends dropped (Mem0, Supermemory, Chroma, Honcho) and gated off via ENABLED_WRITE_BACKENDS. Graphiti is now wired as a real write target, not fictional, over Tailscale to the Hetzner box at http://100.110.194.20:8765. OpenAI keys were restored June 10 after the June 8 revocation, and LanceDB + Hindsight + agent-memory.py were replayed.

  • Keepers: Master Brain Supabase / agent-memory SQLite / Graphiti / Hindsight.
  • Dropped: Mem0, Supermemory, Chroma, Honcho (gated off, reversible).
  • Graphiti write verified over Tailscale: graphiti-mcp http://100.110.194.20:8765
  • OpenAI embeddings restored June 10: LanceDB + Hindsight + agent-memory.py replayed.
agent-memory.py reads .env.local FIRST. On any future key rotation, update BOTH MASTER_API_KEYS.env AND master-brain .env.local.
New Tooling7 voices. One question. One consensus board.

Council-X: 7-Model External CLI Council

Council-X: 7-Model External CLI Council

Council-X shipped this week on Hetzner: a single question fans out to seven external CLI voices simultaneously (Claude, Codex, Gemini CLI, Grok, Kimi, Perplexity, Qwen), then synthesizes a consensus and disagreement board. Three voices are free-tier; four cost pennies. Remote-only: never runs on the home box. The Roundtable app (roundtable-app-blush.vercel.app) was also fixed this week after a port 8765 collision with the Graphiti MCP was killing it every 15 seconds via fuser.

  • 7 voices: claude + codex + gemini-cli + grok + kimi + perplexity + qwen.
  • Runs: /home/merlino/tools/claude-council + ~/bin/council-x on Hetzner.
  • Engine auto-discovers providers/*.sh and drops CLI-shadowed API duplicates.
  • Roundtable fixed (https://roundtable-app-blush.vercel.app): moved to port 8770 (was colliding with Graphiti on 8765).
Open: DeepSeek + Minimax need OpenRouter credit to activate. 3 free voices (Claude CLI/Max, Codex, Gemini CLI) + 4 keyed at pennies.
Blockers and LessonsWhat broke. What we learned.

Blockers, Outages and Hard-Won Lessons

Blockers, Outages and Hard-Won Lessons

Two real blockers this week. One outage (OpenAI key revoked June 8, three days down, self-healed on new key June 10). One port collision (Graphiti MCP and Roundtable both binding 8765, fuser killing Roundtable every 15 seconds - fix: Roundtable moved to 8770). The initial Discord data pull for this site also failed (403 assumed universal, data sourced from memory files instead of the real API). Corrected in v2: real bot-token pull returned 588 messages across 17 channels.

  • OpenAI key revoked 2026-06-08. Pivoted to Gemini self-contained RAG same day. Restored 2026-06-10.
  • Roundtable port 8765 collision with Graphiti MCP. Fix: Roundtable moved to 8770.
  • v1 Discord pull: 403 assumed universal, data from memory files. v2: real API, 588 messages.
  • Lesson: blocked path = ONE retry then PIVOT. On key death, sweep ALL key files and repoint to funded provider.
VPS1 OpenClaw at 95% disk. Flag before next big deploy on that box. Missing bot User-Agent header caused v1 Discord 403 failure.