⌂ Home ☷ Board

Specialised Agents Recommendation — Based on Ticket History

Source: 669 MC tickets (396 done, 209 cancelled, 51 inbox) across 27 projects Date: 2026-04-17 Context: Follow-up to MC-655 ClaudeClaw analysis — which named agents map to Elmar's real work?

Method

Analysed ticket volume by project, theme keywords (code/dev, finance, safair, legal, email, etc.), and domain keyword frequency across all ticket titles + descriptions. ClaudeClaw's generic Main/Comms/Content/Ops/Research split doesn't match our data — our work clusters differently.

Data snapshot

Lane signal Volume
Mission Control (project) 190 tickets
Life (project) 123 tickets
LegalMind (project) 83 tickets
SafairBru + FlySafair 81 tickets
PKA 53 tickets
code/dev theme 365 tickets
data/reports theme 226 tickets
infra/ops theme 186 tickets
email/comms theme 162 tickets
finance/money theme 110 tickets
research/analysis theme 92 tickets

Recommended Specialist Agents (ranked by ROI)

Tier 1 — Must-have (clearest signal, highest volume)

1. luci-dev — Developer / Platform Ops ⭐⭐⭐⭐⭐ - Lane: MC dashboard, skills, systemd services, workers, cron jobs, bug fixes, deploys, infrastructure - Why: 190 MC tickets + 365 code/dev keyword + 186 infra/ops = the single biggest workstream - Tools: dev-loop skill, git/gh CLI, systemctl, playwright, review/qa, health - Personality: Terse, precise, commit-discipline, verification-before-completion - Ticket keywords that route here: fix, deploy, skill, dashboard, service, cron, worker, bug, build, PKA, MC, Luci Infrastructure

2. luci-finance — Money / Entities / Portfolio ⭐⭐⭐⭐⭐ - Lane: Portfolio.xlsx, Werda Ops/Inv, ENCFT, bank statements, FICA/KYC, fund contributions, group-financials, X3, IATA fuel, entity bookkeeping - Why: 37 FICA + 110 finance theme + dedicated skill stack (encft-finance, x3-finance, iata-fuel, portfolio-collect, form-filler). Life project (123) is largely personal finance. - Tools: encft-finance, x3-finance, portfolio-collect, form-filler, xlsx, pdf, iata-fuel-update - Personality: Conservative, precise with numbers, never rounds, proactively captures account details - Ticket keywords: portfolio, FICA, KYC, Werda, ENCFT, trust, bank, invoice, fund, statement

3. luci-safair — FlySafair CEO Work ⭐⭐⭐⭐⭐ - Lane: Board pack, monthly forecast, booking curve, ACSA stats, QlikView, Radixx API, FlightAware, Safair dashboards, fuel, ASL Board reports - Why: 81 tickets across SafairBru + FlySafair projects, 35 flysafair keyword, 9 ACSA, dedicated skill stack (10+ safair-specific skills) - Tools: safair-dashboard, safair-data, booking-curve, board-pack, radixx-api, flightaware-data, acsa-stats-update, acsa-monthly, flysafair-monthly-forecast, qlikview-export, flysafair-brand - Personality: RTM for trends, precise numbers, brand-compliant outputs - Ticket keywords: safair, flight, booking curve, yield, load factor, ACSA, Radixx, board pack, fuel

Tier 2 — High value (clear dedicated project)

4. luci-legal — LegalMind ⭐⭐⭐⭐ - Lane: LegalMind-specific work (contracts, compliance, legal docs) - Why: 83 dedicated tickets, 87 "legalmind" keyword, 98 "legal" keyword — a clearly separate workstream - Tools: pdf, docx, defuddle, form-filler, wiki-ingest, cso (security), review - Personality: Careful, cites sources, conservative on interpretations - Ticket keywords: legalmind, legal, contract, compliance, clause

5. luci-comms — Inbox / Email / Replies ⭐⭐⭐⭐ - Lane: Gmail + M365 triage, drafting replies as Elmar, WhatsApp, Telegram answers, meeting notes, calendar scheduling - Why: 162 email/comms tickets, 133 email keyword, 34 whatsapp, 16 telegram, 16 meeting. You already have a detailed email style guide in your rules. - Tools: inbox-assist, m365, email-index, meeting-notes, pst-search, google-drive, m365 calendar - Personality: Writes like Elmar (casual-professional, short, Afrikaans code-switching, "Elmar" sign-off — your existing communication-style rule already encodes this) - Ticket keywords: email, reply, draft, inbox, meeting, calendar, whatsapp, telegram

Tier 3 — Nice-to-have (meaningful but lower volume)

6. luci-research — Research / Brain / Wiki ⭐⭐⭐ - Lane: Deep research reports, wiki compilation, SecondBrain ingestion, NotebookLM, YouTube analysis, knowledge graph - Why: 92 research theme + 14 wiki + 11 brain + 23 "research" keyword. Dedicated skill stack (deep-research, notebooklm, wiki-ingest, wiki-compiler, wiki-query, search-brain, youtube-analyzer) - Tools: deep-research, notebooklm, wiki-compiler, wiki-ingest, search-brain, youtube-analyzer, firecrawl, defuddle, knowledge-graph - Personality: Cites sources, bullet points preferred, runs long tasks in background - Ticket keywords: research, wiki, brain, notebooklm, deep research, compile, analyze video

7. luci-home — Personal / Home / Hobbies ⭐⭐ - Lane: Home Assistant, Spotify Radio, Padel rackets/bookings, Radio Show, family calendar, personal reminders - Why: Spotify Radio 17 + Home Assistant 4 + Padel 4 + Radio 23 keyword + 76 home/personal theme. Lower volume but distinct domain. - Tools: home-assistant, spotify, radio-show, brain (personal capture) - Personality: Casual, brief, conversational - Ticket keywords: home assistant, spotify, padel, radio show, music, light

Skip for now

Proposed routing (classifier rules)

  1. Project field winsSafairBru/FlySafair → safair; LegalMind → legal; Life → finance or comms (sub-classify)
  2. Keyword prefix@dev, @finance, @safair, @legal, @comms, @research, @home
  3. Content classifier — Haiku sees title+description, outputs one of 7 lanes. Fallback to luci-main (triage) if confidence < 0.6.
  4. Broadcast keywords — "everyone", "team", "status update" → all agents weigh in
  5. Pin override — if ~/workspace/.pin-agent exists, route all to that agent (for focused sessions)

Implementation sketch

Each agent = separate systemd unit with: - Its own Claude Code tmux session + memory directory - Curated skill set (not all 200+ skills loaded — per-agent .claude/settings.json) - Shared vault.db write access (Hive Mind) - Separate Telegram chat thread / topic per agent - One shared "main" agent for triage + fallback

Recommendation

Start with 3 agents: luci-main (triage/fallback) + luci-dev + luci-safair. These cover ~50% of ticket volume and have the strongest skill clustering. Add luci-finance, luci-legal, luci-comms, luci-research in subsequent ships once the split-agent pattern is proven.

This avoids the ClaudeClaw mistake of launching 5 generic agents without domain fit — ours are data-driven from 669 real tickets.