Sunday, May 24
Today.
Sunday committed this week’s focus. Execute the move. Override only with written tax.
This week — magier MIT
Granola→Attio auto-draftShip the integration so sales follow-ups draft themselves from transcripts
Today’s first move
Not committed yet. /morning-brief will read this once it’s set.
MITs
3/4 businesses committed · the one thing that makes the week a winmagier
Granola→Attio auto-draftShip the integration so sales follow-ups draft themselves from transcripts (provisional pending Sunday review)
Morning all week (Mon-Fri ~6h floor)
LeadBridge Local
EasyBell addresses for Rostock + UlmSecure local addresses in both cities so EasyBell can provision local phone numbers for the new sites (see memory `address-verification-rostock-ulm`)
Mon afternoon
TrueToYou
No MIT this week — tend mode
Erbright
Send first 15 cold Kanzlei emails this week
Wed afternoon (per design: validation mode default until 2026-06-11)⏱ 2026-06-11
The week
calendar mirror · ~/Claude/calendar/this-week.mdSun · today 24
Pfingsten (regionaler Feiertag)
Mon 25
Pfingstmontag
9:00(No title)
9:30(No title)
10:00(No title)
17:00(No title)
Tue 26
9:00(No title)
9:30(No title)
10:00(No title)
10:00(No title)
10:00Dienstag
17:00(No title)
Wed 27
9:00(No title)
9:30(No title)
10:00(No title)
10:00(No title)
14:00(No title)
15:45(No title)
17:00(No title)
Thu 28
9:00(No title)
9:30(No title)
10:00(No title)
13:00(No title)
14:00(No title)
17:00(No title)
17:30(No title)
Fri 29
9:00(No title)
9:30(No title)
10:00(No title)
11:00(No title)
17:00(No title)
Sat 30
no events
Inbox
magier: 101 · personal: 33 · refreshed 16:03 · status: degraded⚠ Inbox triage degraded — open inbox manually to be safe.
Triage call failed after retries — open inbox manually.
Partial data — review inbox directly.
Override log
0 since last clear · patterns flagged SundayNo overrides logged. Sunday-Willi’s plan is holding.
OS meta
System.
The machine looking at itself — audit score, skills, memory, wiki, decisions. Reality of the OS, not the day.
OS health
audit + countsOS audit
36/100
Foundation
Skills
61
global
Memory files
117
atomic
Wiki sources
1
ingested
Inbox
25
awaiting Sunday
Overrides
0
this period
Inbox
25 sparks · enriched Sunday · parked off the EA view on purpose[2026-05-23 21:35] [session: EA-build] [lane?: magier] **Granola continuous digestion pipeline — foundational, not feature.** Every Granola transcript → digester that updates memory (new people/decisions/context), project files at `~/Claude/magier/clients/<x>/notes.md`, voice samples for LinkedIn agency, decisions/log.md (on commitments), and surfaces "things you said in calls that need follow-through" to EA. Current `/sales-followup` is a narrow special-case of this. This is the 10× lever — Willi's words. Need: Granola export/API path investigation as first step. Could obviate `voice_examples.md` and continuous-memory-update problem in one move.
[2026-05-23 21:35] [session: EA-build] [lane?: os-meta] In-dashboard chat interface — END-STATE VISION. Today: dashboard is read-only HTML, all interaction happens in Claude Code CLI. Vision: open dashboard, see daily plan, respond inline (multiple-choice picks, free-text edits, "block this in calendar?" prompts) — full morning ritual + decision-flow lives in the dashboard, no CLI required. Web app territory, not static HTML. Big build. Build LAST, after JTBD + Granola + inbox-digest + team-checkin all stabilize. Trigger: when the friction of "switch to terminal" itself becomes the bottleneck.
[2026-05-23 21:35] [session: EA-build] [lane?: magier] LinkedIn pipeline = critical for magier per Willi 2026-05-23. Existing `linkedin-content-magic` skill is the substrate. Vision (already captured above): agentic agency, 3 options per posting day, zero ideation friction. Tied to Granola digester (voice/topics come from transcripts) — Granola must land first or in parallel.
[2026-05-23 21:35] [session: EA-build] [lane?: magier] Scope adjustments to team-checkin spark (2026-05-23 21:05): DROP project budget warnings (Harvest covers it), DROP Many Requests subscription health (existing Looker dashboard, skip until rebuild later), KEEP Many Requests time tracking only (combined with Harvest/Forecast). "Who's away this/next week" moves from EA → magier dashboard only. Birthdays today stays on EA; birthdays-this-week moves to magier. Reason: don't duplicate existing dashboards, don't clutter EA with magier-internal items.
[2026-05-24 10:25] [session: EA-build] [lane?: os-meta] Hermes Agent (Nous Research) as future EA-substrate when leaving laptop entirely — open-source self-hosted agent framework. Self-improving 3-layer memory, 18+ messaging platforms (Telegram/SMS/iMessage/Slack/Email/etc.). Reliability when laptop closed comes from running on $5/mo VPS, NOT from Hermes itself. BYOK LLM — must route via free-tier (DeepSeek/Llama/OpenRouter free models) or local Ollama to honor [[no-separate-anthropic-keys]]. **Don't build now** — overkill for today's wake-up problem (phone alarm or Google Calendar event solves it free + instantly). Worth real eval at Phase 4+ when EA features (inbox-digest, Notion ritual, Granola digest) have proven they change behavior on the Mac and the friction of "must be at laptop" becomes the actual bottleneck. NotebookLM source: c85eb3c1-0279-428f-ad94-3375832f30f6.
[2026-05-24 10:57] [session: idea capture from video] [lane?: LB-improve] OpenClaw-style agent that finds $500k–$1.2M homes without pools, renders an AI pool into the backyard, and mails a before/after postcard to the homeowner — autopilot lead-gen for pool installers. Same pattern generalizes to any high-ticket home-improvement (solar, landscaping, driveways, extensions).
[2026-05-24 11:05] [session: idea capture] [lane?: magier] Magier: agent that listens to client meeting recordings (theirs or ours) and auto-creates briefing tickets in our backend — detects needs mentioned in meetings (e.g. "we need LinkedIn banners"), asks follow-up questions for missing brief details, and optionally proactively suggests Magier services based on detected problems. Key open question: data ownership/privacy — do clients want us to have their meeting data, or do they want the agent in their workspace and only the ticket gets pushed to us?
[2026-05-24 11:33] [session: idea capture] [lane?: saturday] Joy as personal health/coaching assistant — pull Apple Health data + current Alma tracking (food, weight, calories burned, steps) into a health dashboard. Feed in training goals, current routine (reps, mobility, handstand training, etc.) so Joy becomes not just executive assistant but also coach — can suggest exercise improvements based on what's been done and what the goals are.
[2026-05-24 14:49] [session: M008 mental model + audit-os integration] [lane?: saturday] Morning-brief refactor + voice-override dashboard button — extract /morning-brief to morning_brief.py (launchd 6:55am, top card on dashboard), slim skill to /override-plan, then add Override button with Web Speech API → 3 sequential voice questions (displacement / reason / reschedule) → headless `claude -p` → verdict back to dashboard. MVP non-streaming; streaming only if override frequency >5x/week. Full design walkthrough in chat 2026-05-24.
[2026-05-24 20:37] [session: OpenRouter research + three-brain convergence] [lane?: LB-improve] OpenRouter as spare-tire infrastructure — prepaid $10 capped account, two jobs: (1) bulk engine for LeadBridge token-heavy German SEO grunt work (keyword clustering, lead qualification) via cheap models like DeepSeek instead of burning Claude Code rate limits, (2) failover for /three-brain when Codex/Gemini CLI rate-limit mid-flow. Not scouting, not production, not default. Research at ~/Claude/shared/research/2026-05-24-openrouter/. Three-brain converged: prepaid hard cap ≠ Joy-bot metered key risk.
Recent decisions
last 5 · ~/Claude/decisions/log.md2026-05-23 · os · Full skill audit pass — every user skill at ~/.claude/skills/*/SKILL.md now carries a `business:` tag (53 total tagged across this session: 10 in initial pass + 36 batch + 9 sweep + 1 already pre-tagged via prior morning-brief/weekly-review work). Archived `system-design` + `research-to-plan` to ~/.claude/skills/_archive/ (overlapped with superpowers:brainstorming/writing-plans and deep-research respectively). Tag distribution: ~22 leadbridge · ~16 os-meta · ~13 shared · 3 magier. Skill count 51 → 58 (new skills added since last audit) — 50 user-built skills all tagged, 0 untagged remaining. Per-business reasoning unlocked for the full skill surface.
2026-05-23 · os · evening-diary skill archived (Joy-bot leftover, didn't fit current in-session-only Joy flow). Moved ~/.claude/skills/evening-diary → ~/Claude/archive/skills/evening-diary. Launchd job com.willi.claude.evening-diary unloaded and plist moved to archive. Wrapper script run_routine.sh stripped of the evening-diary case. Active automated routines now: morning-brief (Mon-Fri 06:00, headless run + notification) and weekly-review (Sun 16:00, notification only). Total automation surface deliberately small and high-signal.
2026-05-23 · os · Productivity OS routines fully automated via macOS launchd (no more relying on Willi to remember to /trigger). Three agents loaded at ~/Library/LaunchAgents/: (1) com.willi.claude.morning-brief — weekdays 07:55, runs claude -p /morning-brief headlessly via wrapper script, writes ~/Claude/morning-brief-today.md, fires desktop notification. (2) com.willi.claude.evening-diary — daily 21:00, desktop notification reminder to open CC and run /evening-diary (interactive, can't run headlessly). (3) com.willi.claude.weekly-review — Sundays 18:00, desktop notification reminder for /weekly-review. Wrapper script at ~/Claude/shared/scripts/run_routine.sh. New durable mental model saved as feedback memory automate-by-default: anything on regular foreseeable cadence MUST be automated; friction & cognitive load are the real bottleneck on execution; applies to OS design, product design, communication.
2026-05-23 · os · Capabilities cleanup pass — deleted saas-landing-copy stub skill (EXTRACTION_LOG archived to ~/Claude/magier/saas-landing-copy-extraction-notes.md). Marked joy-sync as DEPRECATED in frontmatter + tombstone note (bot killed 2026-05-22). Added `business:` tags to top 10 most-used skills per skill-usage.md: leadbridge × 8 (deep-analysis, analyze-niche, preflight-build, local-seo-image-prompts, local-seo-content-planner, local-seo-copywriter, leadbridge-pipeline-gate, build-site), shared × 2 (graphify, firecrawl-search). Per-business reasoning now unlocked.
2026-05-23 · os · Codex final-pass review applied — 3 warnings + 2 nits all addressed. (1) Reschedule grammar tightened from a vibes-based table to an explicit 4-form acceptance grammar (date+daypart, date+clock-window, 'next Sunday in /weekly-review', or 'killed'); day-of-week-without-date is now explicitly invalid. (2) create_deep_block.py hardened: semantic validation (end>start, 60-240min duration), proper exception handling on subprocess (FileNotFoundError + TimeoutExpired for both delete and create), honest idempotency docs ('best-effort, may duplicate on delete failure'), warning surfaced to stderr instead of silent swallow on delete failure. All four failure paths smoke-tested. No blockers from Codex; ship state.
OS initiatives
6 open · Joy OS / cross-cutting · click Pick up to copy a resume promptJoy OS v1 — lane→initiative→task hierarchy
Activeclean separation between *ideas* and *tasks* ; makes weekday execution zero-decision.
5/7 done
Pattern-mining across calls + Claude chats
Active/granola-digest already mines Granola call transcripts into actionable dream-ideas . Willi wants the same pattern recognition applied to *Claude chat sessions* — recurring requests…
0/5 done
Productivity OS hardening — post-v1
ActiveOS Constitution + audit infra ratified 2026-05-22. Two-week post-build review scheduled per design.
0/5 done
Steal-from-gstack — port pressure, not names
Activegstack is the most thought-through public Claude Code stack. Six patterns + the artifact-chaining meta-pattern map cleanly to gaps in Willi's OS.
0/9 done
Obsidian as substrate over ~/Claude
ActiveConvert `~/Claude/` into an Obsidian vault — human UI layer over the existing markdown OS. Doesn't change runtime; unlocks graph view, backlinks, mobile reading, vault portability.…
0/7 done
Always-on cadence engine on Hetzner VPS
ActiveCadence engine must fire regardless of laptop state — going fully nomadic 2026, deregistering from Germany, no permanent home for a Mac mini.
0/12 done
System health
all green · 3 tasks · checked 2026-05-24 16:30:11morning-brief
0.0h ago · daily 07:00
granola-digest
0.4h ago · Thursday 18:00
weekly-review
0.1h ago · Sunday 16:00
Webflow & graphic design agency
magier.
Profit
€90.081 / €1.000.000
Margin
10.5% / 30%
€7.206
Bonus YTD
8% tier · proj. €21.619
OKRs — Q2 2026
from Notion · 4 key resultsOTP NPS
94%
85 / 90
Implementing our first 3 internal agents or tools
0%
0 / 3
New OTP Structure creates healthy margins
102%
66 / 65
magier Studios booked new €250k OTP Revenue
54%
284k / 522k
Dream ideas
Pre-churn health signal from call transcripts
Pain pattern
Elephant and Neuroflash churn was confirmed at the weekly with no advance warning. Meanwhile, Moonwards is genuinely happy but that satisfaction only surfaced because Slint happened to run a feedback call. The gap between these two outcomes is a per-client sentiment tracker that does not exist today. Seen in: magier - Weekly, Feedback Call Magier x Moonwards
Suggestion
Extend /granola-digest to output a weekly per-client health score alongside the OKR insights. For every active client mentioned in transcripts, extract satisfaction signals, unresolved asks, and friction indicators, then assign a simple green/yellow/red status. Surface yellow and red clients on the magier dashboard tab.
Implementation hint
The existing /granola-digest already scans all transcripts and classifies by client. Add a second pass that filters for client-work calls, extracts sentiment per client, and writes to ~/Claude/magier/client-health.md. The magier dashboard tab already exists and can render an additional card. One-day build on top of the current digest infrastructure.
Why now
Last week's dream was a post-churn autopsy. This is the proactive version: catch the yellow before it turns red. Two clients leaving in the same month without warning means the signal exists in the calls and nobody is reading it.
AI initiative registry across magier
Pain pattern
Multiple team members are building Claude Code skills, Slack bots, and AI automations independently. Willi discovered overlap only by accident during conversation. Max created a FigJam for mapping but it has no automated feed and will go stale within two weeks. Seen in: magier - Weekly, Daily Stand Up
Suggestion
A shared registry where each AI tool, skill, or automation built by any team member is logged with owner, status, and what it does. Weekly rollup surfaced in the magier dashboard or Slack channel. Does not need to be fancy: a Google Sheet or Notion table with five columns is enough.
Implementation hint
Willi's existing /skill-usage skill already scans ~/.claude/skills/ and reports which are active. The simplest extension is a shared Google Sheet that each person updates after building something. The smarter version runs a cron job that diffs each person's skill directory and writes changes to one central file. Half-day build for the manual version, one-day build for the automated version.
Why now
OKR 3 targets three internal agents or tools and sits at zero. The team is building things that could count toward this OKR but nobody has visibility into what exists across the org. The coordination problem is cheap to solve and expensive to ignore.
Initiatives
0 open · click Pick up to copy a resume prompt for Claude CodeNo magier initiatives in
~/Claude/magier/initiatives.md yet.Rank-and-rent SEO sites
LeadBridge Local.
Active initiatives and their next moves. Pick up where you left to resume work in Claude Code.
Initiatives
4 open · click Pick up to copy a resume promptSite build queue
Activequeue of sites to build, in order. Source: memory `next-niches-queue`.
0/4 done
EasyBell address provisioning — Rostock and Ulm
ActiveEasyBell needs verified local addresses in each city to provision local phone numbers. Blocks per-site numbers on new builds. See memory `address-verification-rostock-ulm`.
0/2 done
Novel infrastructure / new skills
Activerolls up `/level-up` candidates as they emerge from Sunday review.
0/3 done
Backlog — deferred LB work
Activesee `~/Claude/leadbridge-local/BACKLOG.md` for the canonical list. Top item: EasyBell voicemail-to-email setup. This section mirrors only what `/task` adds via the markdown layer.
capture pad
Wax-to-silver ring kit
TrueToYou.
Active initiatives and their next moves. Pick up where you left to resume work in Claude Code.
Initiatives
2 open · click Pick up to copy a resume promptUS validation launch
Activeprove demand with real purchases before building full infrastructure. Validation thesis: 10+ committed buyers from ~$1k ad spend . See memory `truetoyou-pricing-gtm`.
0/5 done
Brand + creative
Tendbrand designed already; ongoing visual polish for ads + site as needed.
capture pad
The bet
Erbright.
Active initiatives and their next moves. Pick up where you left to resume work in Claude Code.
Initiatives
2 open · click Pick up to copy a resume promptNo-Vorkasse-Partner validation bet
Activevalidate whether a Kanzlei will sign on as no-upfront partner before investing in product build. Signed commitment = GO; 30+ cold contacts with zero commitments = KILL.
1/5 done
Friends-first informal interviews — pre-validation phase
Pausedper `project_erbright` memory — friends-first informal interviews before pitching any lawyer.
capture pad
Life outside the businesses
Personal.
Health, fitness, taxes, life admin, nomad logistics. Anything not tied to a business.
Initiatives
1 open · click Pick up to copy a resume promptPersonal / life admin
Tendcaptures personal things that need doing but don't slot under a business.
capture pad
Parking lot
Ideas.
Sparks waiting for Sunday triage. Grouped by business so they're browsable without bleeding into today.
Operating System
11 ideas2026-05-24 14:49
Morning-brief refactor + voice-override dashboard button — extract /morning-brief to morning_brief.py (launchd 6:55am, top card on dashboard), slim skill to /override-plan, then add Override button with Web Speech API → 3 sequential voice questions (displacement / reason / reschedule) → headless `claude -p` → verdict back to dashboard. MVP non-streaming; streaming only if override frequency >5x/week. Full design walkthrough in chat 2026-05-24.
2026-05-24 11:33
Joy as personal health/coaching assistant — pull Apple Health data + current Alma tracking (food, weight, calories burned, steps) into a health dashboard. Feed in training goals, current routine (reps, mobility, handstand training, etc.) so Joy becomes not just executive assistant but also coach — can suggest exercise improvements based on what's been done and what the goals are.
2026-05-24 10:25
Hermes Agent (Nous Research) as future EA-substrate when leaving laptop entirely — open-source self-hosted agent framework. Self-improving 3-layer memory, 18+ messaging platforms (Telegram/SMS/iMessage/Slack/Email/etc.). Reliability when laptop closed comes from running on $5/mo VPS, NOT from Hermes itself. BYOK LLM — must route via free-tier (DeepSeek/Llama/OpenRouter free models) or local Ollama to honor [[no-separate-anthropic-keys]]. **Don't build now** — overkill for today's wake-up problem (phone alarm or Google Calendar event solves it free + instantly). Worth real eval at Phase 4+ when EA features (inbox-digest, Notion ritual, Granola digest) have proven they change behavior on the Mac and the friction of "must be at laptop" becomes the actual bottleneck. NotebookLM source: c85eb3c1-0279-428f-ad94-3375832f30f6.
2026-05-23 21:35
In-dashboard chat interface — END-STATE VISION. Today: dashboard is read-only HTML, all interaction happens in Claude Code CLI. Vision: open dashboard, see daily plan, respond inline (multiple-choice picks, free-text edits, "block this in calendar?" prompts) — full morning ritual + decision-flow lives in the dashboard, no CLI required. Web app territory, not static HTML. Big build. Build LAST, after JTBD + Granola + inbox-digest + team-checkin all stabilize. Trigger: when the friction of "switch to terminal" itself becomes the bottleneck.
2026-05-23 21:05
"What needs your brain today" combined card — collapse emails-to-reply + LinkedIn drafts + team blockers + calendar conflicts into one glanceable surface on EA tab. End-state goal, only meaningful once the three feeders (inbox, LinkedIn, team) exist. Build LAST.
2026-05-23 21:05
Cross-business operations rollup — LeadBridge has STATUS.md + /audit-portfolio. Magier and TrueToYou don't have equivalents. Unify: one health line per business in EA hero strip. Reuses team-check-in plumbing once that lands.
2026-05-23 15:00
iPhone microphone records the entire day → transcripts → fed into Claude Code as ambient context
2026-05-23 13:30
Extend `~/Claude/connections.md` with monthly cost per tool + total burn rollup + per-business cost breakdown (inspired by Jack's connected-tools overview from his video). Possibly auto-flag tools that cost €X but haven't been touched in 6+ weeks via skill-usage cross-reference.
2026-05-23 12:10
Personal daily 10–15 min "morning podcast" — digests podcasts/YouTube/news across my interests (health, travel/adventure, lifestyle business + indie hackers, health tracking devices, legal tech) and serves a custom audio brief each morning, capped with a Seneca-style philosophical/motivational quote. Build for myself first; productize as an app later. Explicitly NOT covering Claude Code (Intel Weekly already does that).
2026-05-23 09:25
Personal CRM — one markdown page per interesting person I meet, surfaces relevant contacts when starting new initiatives (e.g. legal tech folks for Erbright). Maybe weekly review prompt to add anyone met that week. Open question: plain markdown vs. vector DB for semantic surfacing. Better idea welcome.
2026-05-23 09:17
Personal finance dashboard pulling bank/brokerage data via GoCardless API + realistic magier exit valuation model for accurate net worth tracking
magier
8 ideas2026-05-24 11:05
Magier: agent that listens to client meeting recordings (theirs or ours) and auto-creates briefing tickets in our backend — detects needs mentioned in meetings (e.g. "we need LinkedIn banners"), asks follow-up questions for missing brief details, and optionally proactively suggests Magier services based on detected problems. Key open question: data ownership/privacy — do clients want us to have their meeting data, or do they want the agent in their workspace and only the ticket gets pushed to us?
2026-05-23 21:35
Scope adjustments to team-checkin spark (2026-05-23 21:05): DROP project budget warnings (Harvest covers it), DROP Many Requests subscription health (existing Looker dashboard, skip until rebuild later), KEEP Many Requests time tracking only (combined with Harvest/Forecast). "Who's away this/next week" moves from EA → magier dashboard only. Birthdays today stays on EA; birthdays-this-week moves to magier. Reason: don't duplicate existing dashboards, don't clutter EA with magier-internal items.
2026-05-23 21:35
LinkedIn pipeline = critical for magier per Willi 2026-05-23. Existing `linkedin-content-magic` skill is the substrate. Vision (already captured above): agentic agency, 3 options per posting day, zero ideation friction. Tied to Granola digester (voice/topics come from transcripts) — Granola must land first or in parallel.
2026-05-23 21:35
**Granola continuous digestion pipeline — foundational, not feature.** Every Granola transcript → digester that updates memory (new people/decisions/context), project files at `~/Claude/magier/clients/<x>/notes.md`, voice samples for LinkedIn agency, decisions/log.md (on commitments), and surfaces "things you said in calls that need follow-through" to EA. Current `/sales-followup` is a narrow special-case of this. This is the 10× lever — Willi's words. Need: Granola export/API path investigation as first step. Could obviate `voice_examples.md` and continuous-memory-update problem in one move.
2026-05-23 21:35
**JTBD mapping for magier — DO THIS FIRST before more EA builds.** 45-min structured session, outputs `~/Claude/magier/jtbd.md`. Per role-category (sales/delivery/HR/finance/ops/comms/leadership) → discrete jobs (verb+object+outcome) → frequency / time / cognitive load / drain-energy / automate-delegate-keep-kill. North star for exit strategy AND for prioritizing every EA-build that comes after. Best done in a Sat freestyle slot.
2026-05-23 21:05
Team check-in digest for magier — sources are Harvest + Forecast (one-time projects time + budgets), Many Requests (subscription clients + their time tracking, also viewed via Looker Studio), Slack (immediate-attention-only filter, not feed), Deel (PTO + birthdays + payroll), Google Sheet (PTO active/passive planning — Deel↔Sheet congruence check is a sub-agent). NOT ClickUp (Willi corrected — magier doesn't use it). Notion has SOPs + task board for smaller magier ops todos. Big build, multi-stage. Open: which source first.
2026-05-23 21:05
LinkedIn agentic agency — END-STATE VISION (not next build). Replace the "I sit down and write posts" workflow entirely. A multi-agent team (social-media strategist + LinkedIn expert + copywriter + visual director) runs continuously off Willi's voice/wiki/recent intel and proposes 3 fully-drafted post options per posting day. Willi picks one, suggests edits in chat, agency revises, posts. Friction target = zero ideation, only taste calls. Visuals included. Pillar balance handled invisibly by the agency, never surfaced as a balance number. Build comes AFTER inbox-digest + team-checkin land — this is the high bar, not the MVP. Existing `linkedin-content-magic` v1.2.0 is the starting substrate.
2026-05-23 09:21
Shared AI personal assistant for Anna + me at magier — own email identity (e.g. jack@magier.com), sends onboarding follow-ups + handles routine customer requests as if it were us. Stretch: full persona that could even attend Google Meet calls. Also cross-references designer time-tracking vs. vacation calendar and surfaces "X is out next week" heads-ups.
LeadBridge Local
2 ideas2026-05-24 20:37
OpenRouter as spare-tire infrastructure — prepaid $10 capped account, two jobs: (1) bulk engine for LeadBridge token-heavy German SEO grunt work (keyword clustering, lead qualification) via cheap models like DeepSeek instead of burning Claude Code rate limits, (2) failover for /three-brain when Codex/Gemini CLI rate-limit mid-flow. Not scouting, not production, not default. Research at ~/Claude/shared/research/2026-05-24-openrouter/. Three-brain converged: prepaid hard cap ≠ Joy-bot metered key risk.
2026-05-24 10:57
OpenClaw-style agent that finds $500k–$1.2M homes without pools, renders an AI pool into the backyard, and mails a before/after postcard to the homeowner — autopilot lead-gen for pool installers. Same pattern generalizes to any high-ticket home-improvement (solar, landscaping, driveways, extensions).
Unsorted
4 ideas2026-05-23 14:30
OS-backlog: automate /audit-portfolio monthly (1st of month, weekday). Source: same pass.
2026-05-23 14:30
OS-backlog: automate /intel-collect Mon/Wed/Fri to actually fire the intel pipeline. Currently the skill exists but Willi has to remember to run it 3x/week — anti-pattern per reduce-friction. Source: same pass.
2026-05-23 14:30
OS-backlog: automate /quarter-review on calendar-quarter boundaries (1st of Jan/Apr/Jul/Oct, 16:00). Source: same pass.
2026-05-23 14:30
OS-backlog: automate /audit-os on quarterly launchd schedule (every 3 months) — companion to /quarter-review. Source: today's automation pass deliberately stopped at 2 routines to avoid over-engineering.