Night Metro
The overnight map.
25 jobs. Micro-skills under each. The bot walks the lines while you sleep. Morning you only tap amber.
World's largest Grok Bot skill repository.
Inbound — what arrives
Email / inbox triage
easy-archive · glance-mail · action-draft · VIP hold · invoice extract · unanswered nudge
Slack attention
Hourly attention list from Slack, mail, calendar, Granola · Narrow Slack/GitHub event → repro pack (no auto-post) · hourly scan · mention filter
Calendar
Pull two months of open items → Sheet Kanban → daily nudge · day-ahead pull · conflict flag · prep note
Meeting notes
Fathom poll → summary + two action lists · standup-notes SKILL.md: git log last 24h → copy-paste standup draft · poll transcript · owner per item
CS inbox
Client CoS emails Agints CS when stuck · 2–6 Bot group with one owner per stage · stuck after one retry · email Agints CS
Decide — drafts + stop lines
Research brief
Fable Forensic Skill — verify the artifact, not the performance · Claude Fable 5.1 deep-article SKILL.md (effort:high, fork research) · obra/superpowers brainstorming — design-before-code gate · Force x_search + web_search in parallel, never default-web
Voice drafts
Build a send-voice guide from outbox + Slack, then sample the thread · outbox sample · slack tone · thread draft
Night log
Solo-founder Grok Bots: interview once, state file, if-nothing-send-nothing, Outer Loop never commits · loop-runner SKILL.md: one iteration, files are the only memory, cron owns cadence · memory-core dreaming: 03:00 cron, light→REM→deep; only deep writes MEMORY.md · no commit
Watch
anthropics/skills webapp-testing — Playwright + with_server.py black box · Claude Code /aun SKILL.md — one AUN URL, AskUserQuestion, no '対応済' · Babysitting-PR SKILL.md: keep an open PR merge-ready · Grok writes Hermes Kanban cards sized to context · Hermes Kanban — decompose, don’t execute · obra/superpowers test-driven-development skill (Iron Law)
Approvals
Claude Code hooks: Prettier, protect-files, Stop prompt, git if-filter · First SKILL.md: commit-helper + slash-only prod-deploy · Project hooks.json: afterFileEdit formatter + cloud-safe command hooks · disable-model-invocation for deploy/PR/billing skills · Draft-first Auto-review + takeover; Bots are not a security boundary · HAEGONG Grok Bot profiles: Spec Writer → human approval → PR Producer; PR Verifier in a separate chat
Outbound — people
Cold waterfall
LinkedIn engager → waterfall enrich → Instantly · Five-bot outbound team that cannot send · extract engagers · waterfall 3–5
LinkedIn engager
LinkedIn engager → waterfall enrich → Instantly · engaged-only · no cold scrape · no auto-DM
Follow-ups
Build a send-voice guide from outbox + Slack, then sample the thread · due threads · voice match · draft only
Blog pulse
Blog Pulse: write the role, run once, save skill, then schedule with no-guess gaps · schedule gaps · approve live · micro 1
Social lurk
Force x_search + web_search in parallel, never default-web · Claude Fable 5.1 deep-article SKILL.md (effort:high, fork research) · lurk overnight · brief in the morning
Grow — ads + search
Google Ads BOFU
Google Ads BOFU keyword-family agent · Marketing agent access stack · BOFU families · deepest conversion
Meta Advantage+
Claude Code end-to-end Meta lead-gen (browser BM + API, never ACTIVE) · Meta Ads MCP skill — create PAUSED, never delete, never +20% budget · Facebook Advantage+ outcome ads · OpenClaw meta-ads SKILL.md — daily ops + 8-stage /meta-campaign, ads paused
LinkedIn ads
LinkedIn remix → thought-leadership ads · remix posts · thought-leadership set · draft ads
SEO/AEO draft
AI search is SEO (best X for Y) · SEO/AEO workflow — GSC → keywords → CMS → draft for approval · BOFU SEO + monthly refresh · GSC pull
Better page
Replace stale links with a better page · find stale link · write better page · you deploy
Keep — SKILL.md, routines, memory, roster
Write SKILL.md
Write every agent spec like a job description · Official Agent Skills SKILL.md spec (agentskills.io) · AGENTS.md nested playbook + Cursor Cloud specific instructions · anthropics/skills mcp-builder — four-phase MCP server skill · anthropics/skills pdf — production document skill with FORMS.md split · anthropics/skills skill-creator — write, eval, iterate SKILL.md
Skill then routine
Claude Code cron: /loop, CronCreate, Desktop tasks, cloud Routines · Cursor Automations: /automate, cron, MEMORIES.md, PR/Slack tools · Start with SKILL.md, watch the browser, refine · Skill runtime — load, execute, unload + approval field · Task → 6-field skill → routine (never reverse) · Grok Bot Super: URL playbooks → official save-as-skill
Memory as files
Claude Code memory: CLAUDE.md, .claude/rules, auto MEMORY.md · Grok Bot: put changing facts in /workspace, not Bot memory · Hermes bounded memory: MEMORY.md + USER.md · OpenClaw memory is files: USER.md, MEMORY.md, memory/YYYY-MM-DD.md, DREAMS.md · Saving workspace context: file-based agent memory across chats
Named roster
Named bots manage. Heavy Lift Cloud Agents do the compute. · Named Bots, scoped memory, smallest useful roster · Install skills via plugin folders + marketplace.json · Grok Bot Templates — package skills, memory, plugins as a link · Chief of staff runs 10–20 named Grok Bots · Grok Bot Chief of Staff that reads COMPANY.md / APPROVALS.md first
Teach by demo
Cloud Agent environment.json: install vs start vs terminals + Dockerfile · One shared computer, /workspace handoffs, screens ≠ security · Teach a task — 10-minute browser demo → draft skill · ≤10 min demo
your agents work while you sleep. · 93 SKILL.md units in the warehouse