ATLAS · Leadership
Chief of Staff
Runs the day and routes everything: inbox, calendar, Slack, meetings, accountability, logistics. The operator's right hand and the top-level router.
MINT Onboards an autonomous agent the way you'd onboard a new hire — gives it a defined identity, the memory and tools it needs, and its own bounded ability to pay — so it can transact without a human approving every step. Produces an agent-onboarding spec, a payment-rails plan (non-custodial agent wallets, agent-to-agent micropayment standards, on-chain identity registries), and the guardrails (spend limits, approval boundaries) that keep autonomy safe. Triggers on: 'give my agent its own identity', 'how do I let an agent pay for things', 'agent wallet', 'onboard an autonomous agent', 'agent that can transact', 'agentic commerce', 'agent payments without a human in the loop', 'agent-to-agent payments', 'set spend limits for my agent'.
SENTRY Adds a supervising watcher agent to any multi-step AI pipeline so broken steps get caught and corrected before they cascade. Each worker gets explicit success criteria; the watcher audits every output against them, sends failures back for self-correction, and learns repeat failure modes to improve worker instructions over time. Framework-free — one review agent dropped onto an existing chain, no LangGraph/CrewAI/AutoGen adoption required. Triggers on "my agent pipeline fails silently", "one agent's bad output breaks everything downstream", "add a reviewer agent", "watcher agent", "orchestration agent", "quality gate for my agents", "who's checking the agents", "my automation works in testing but breaks at scale", "I have to manually trigger every step", "review my agents the way I would". Use for content pipelines (draft to review to publish), data pipelines (extract to validate to load), sales engines, or any chained-agent workflow that runs unattended.
MERIDIAN Installs a verified-time-first anchor block into global instructions and any time-sensitive or scheduled skill, so agents never reason about deadlines, 'today', or 'this week' from a drifted or wrong internal clock. Hardens standups, briefings, cron bots, and any schedule-aware workflow against silent date errors that cascade. Triggers on: 'my agent has the wrong date', 'the AI thinks it is the wrong day', 'time anchor', 'ground-truth time check', 'fix the date drift', 'add a date check to my instructions', 'my scheduled agent is off by a day', 'make my cron skill time-aware', 'verify the date before reasoning', 'anchor my agent to the real time'. Use when an agent runs on a schedule, reasons about due dates, or any 'what is next / what is due' logic must be correct.
KEYSTONE Install the owner's Company Brain Atlas Folder — the home for their AI company brain — as a plain local folder Claude Code reads and writes directly, so every other AOS skill can read it. Local-first by default: no cloud account, no connector, nothing to authorize — it works instantly in Claude Code. Then a progression you choose, never required: point Obsidian at the same folder to see the brain as linked notes; graduate to Google Drive, Dropbox, or OneDrive later when sharing is needed. Takes them from 'I have nothing in there' to a clean Company Brain Atlas Folder (soul.md, company-brain.md, Company/, AI-Team/) the AI reads front-door-first. Phased and gated: confirm the path, build the folders, seed the brain, verify by listing it back, then say what to fill first. Triggers on: 'set up my atlas folder', 'build my company brain as a local folder', 'create my AI company brain', 'start my second brain locally', 'set up my atlas folder in obsidian', 'I have nothing in my context fol...
The Company Brain's calls librarian: sweeps every new call recording into durable, filed context. Detects whatever recorder the member runs (Fathom, Granola, Plaud, Zoom, phone captures), identifies WHO was on the call, its NATURE, DATE, and LINK, then files by nature — 1:1s to the person's desk, location calls to the location's folder, cross-team calls to the Transcript Vault, sensitive calls to the manager's own desk only — links each into today's Daily-Notes, and harvests answers into the FAQ. Uncertain speaker attribution is tagged INFERRED, never stated as fact; nothing is ever deleted. Runs on a cadence (hourly or daily) and on demand. Triggers on: 'file my calls', 'sweep recordings', 'process today's calls', 'log this call', 'where did that call go', 'call filing sweep', 'transcript sweep', or any request to file, index, or route call recordings and transcripts into the company brain. Built for the operator whose calls evaporate before they become context.
Install an AI staff engineer that guards the default branch - it fires on every pull request, runs deterministic checks first (a failed check can never be argued past), then reviews the diff adversarially: bugs, regressions, security holes, convention breaks, and diff-vs-intent mismatches block; style never does. Approved PRs squash-merge under a human merge identity with the deploy verified live (failed deploy auto-reverts); rejections return with reasons for up to two fix rounds; protected paths, oversized diffs, and repeat strikes escalate to a human. Every decision reported to the ops channel. Triggers on: 'code review agent', 'staff engineer review', 'PR review bot', 'merge gate', 'auto-merge with review', 'who approves PRs', 'block bad merges', 'deploy rollback on failure', 'adversarial code review', or any request to put a review-and-ship gate on a repo. Built for the team whose agents open PRs faster than anyone can review them.
VEIL The redact-first safety guard that runs BEFORE any sensitive text touches an AI prompt or a shared artifact. Classifies the input (financial/admin vs clinical/personal vs operational), strips the 18 HIPAA Safe-Harbor identifiers plus your custom rules, and swaps each for a stable PROXY ID — so a dashboard, report, or draft stays usable but de-identified. Returns the redacted text, the proxy map (the member keeps it; it never enters our systems), and a one-line safety note. When the work truly needs protected data, it HALTs and says exactly what's required — it never fabricates a record. Reminds of BAA scope (Enterprise-only) and the rule: redact-first, records hold the truth, Claude needs the shape. Callable as a step inside any other skill. Triggers on: 'redact this', 'make this HIPAA-safe', 'scrub PII before I prompt', 'is this safe to put in Claude', 'check this for PHI', 'de-identify this', or any request to remove protected data before it reaches an AI tool.
LANTERN Onboard a new Agentic Society member from member link to connected MCP, first useful win, curated starter stack, and company brain setup. Guides Claude web/desktop, Claude Code, Codex, Cursor, or another MCP harness; confirms the catalog loaded; teaches agentic_os_aicos as librarian, curator, and implementation architect; runs one zero-setup first win; then stands up the three company-brain foundations: Goals, Org Chart, and Strategic Brief. Tech-stack aware and non-technical-friendly. Triggers on: 'connect to agentic society', 'set up my MCP', 'connect the agentic os', 'I just got my member link', 'help me get started', 'add the connector', 'get my first win', 'curate my starter stack', 'what AI employees should I use', 'set up my company brain', 'how do I use this', 'onboard me', or any request to connect to or get started with Agentic Society.
SPLICE The friendly, non-technical helper that connects any tool to Claude — safely — for anything WITHOUT its own dedicated wizard (Gmail/Drive/Docs/Sheets/Calendar go to multiple-gmail-account-rabbit; YouTube goes to youtube-connector). Detects the connection kind and routes it: a native connector (Calendar, Slack, and other first-party connectors), a custom/remote MCP server (URL + optional token), or an API-key connection for tools without a native option (CRM, Stripe, ClickUp). Enforces safe defaults: start READ-ONLY, grant least-privilege scopes, route anything sensitive through the redaction guard first. Verifies with one tiny test; for tools with no connector yet, gives the bridge pattern (structured export until an API exists). Triggers on: 'help me connect a tool', 'set up a connector', 'connect my CRM to Claude', 'how do I add an MCP server', 'connect Stripe / ClickUp', 'add a custom connector', or any request to wire in a tool that isn't Gmail/Google Workspace or YouTube.
SIFT Turn the operator's inbox into a 30-second morning briefing — and draft the replies in their own voice. Connects to the mail they already use (Gmail by default), reads the last day's messages, and produces a scannable briefing: what needs a reply today, what's time-sensitive, what's FYI, and the top 3 priorities. For messages that need a response, it drafts replies in the member's brand voice (from their Brand Voice Guide) — drafts, never auto-sent. Built to run every morning and save as a reusable 'Morning Briefing' skill. Tech-stack adaptive (Gmail / Workspace / Outlook) and voice-aware (uses the documented voice, or asks once). Triggers on: 'daily briefing', 'morning briefing', 'brief me on my inbox', 'what needs a reply today', 'who's waiting on me', 'catch me up on my email', 'draft replies in my voice', 'inbox briefing', or any request to summarize the inbox and draft responses. Built for the operator who starts the day knowing where to spend attention.
The end-of-day close — the evening bookend to agentic-os-daily-briefing (the morning open). Sweeps today's calls first via agentic-os-call-filing-sweep, reviews the day's sessions and changed brain files, then writes the dated Daily-Notes journal with exactly four sections — what happened, what we learned, what shipped, open questions — and copies every unanswered question into the FAQ inbox. Flags contradictions with the brain instead of fixing them silently, keeps the ledger append-only (one note per day per desk; re-runs append timestamped addenda), routes teammates to their own People desks, and on Fridays suggests promoting repeated notes into standing rules. Triggers on: 'log my day', 'daily close', 'daily journal', 'wrap up today', 'end of day close', 'update my journal', 'close out the day', 'evening review', or the scheduled end-of-business task. Built so the company brain learns something every working day.
VEIL Stand up a lightweight, practical AI data-governance model for a small regulated team — defensible if audited, without a compliance department. Built for practices handling sensitive data (healthcare/HIPAA, legal, financial, HR) that want to use AI safely now. Produces six artifacts from the member's actual data sources: a data-classification map; a role-based access matrix; a proxy-identifier scheme for dashboards and AI workflows; a Claude/connector compliance checklist (read-only first, redact-first, review every output); BAA guidance (Enterprise-only); and a one-page 'AI Data Governance Lite' policy doc. Phased and gated — plans first, never invents the member's data surface. Composes with the redaction guard as per-prompt enforcement. Not legal advice — counsel reviews before adoption. Triggers on: 'set up HIPAA-safe AI guardrails', 'AI data governance for my practice', 'who can see what in our dashboards', 'make our AI use compliant', 'governance lite'.
LANTERN The live, guided delivery of the Agentic OS 7-Day Quick Start (Chat → Code). A coach that walks a member through the crash course one block at a time — Start Here → Day 1..7 — adapting to where they are: if the foundations exist it skips setup and hands a SEED PROMPT for the next real step. Each block states the Topic, Outcome, Do-this steps, and resource links, then WAITS before advancing. Frames the climb in the Agentic OS voice, points to the foundation (CLAUDE.md + Company Brain + Atlas Folder) and the team — you're the CEO; HARPER recruits your AI employees; ATLAS routes the work; 10 leaders with 140+ specialists behind them — holds the compliance rails (redact-first; review every output) while bending each day toward the member's goal. The script lives in supporting/7-day-quickstart.md. Triggers on: 'walk me through the quick start', 'start the 7 day crash course', 'getting started with Claude', 'guide me day by day', 'where do I start with Agentic OS', 'I'm on day 3, what now'.
OUTRIDER Scout — the reconnaissance AI employee that primes a domain BEFORE you research or build. Triggers: 'I don't know where to start with…', 'prime me on…', 'what do I need to know about…', 'scout this…', 'I'm not a subject-matter expert in…', 'get me up to speed on…', 'before I build…', 'what are my unknown unknowns on…'. It maps the vocabulary, frameworks, methodologies, tools, and standards of a field, and surfaces the unknown-unknowns — so whatever you build next starts from expertise, not a blank page. Run it as step zero.
Install an AI chief of staff that lives in the team's Slack - the knowledge layer that turns transcripts, recordings, and channel chatter into a structured Company Brain (session recaps, member and company profiles, use cases, decisions, open questions), posts the narrative Week Ahead / Week in Review digests, owns the company's running ship log, and hands every action item to the PM agent through the handoff queue. Triggers on: 'chief of staff agent', 'AI chief of staff', 'process this transcript', 'meeting recap bot', 'company brain', 'knowledge base from meetings', 'weekly digest', 'week in review', 'who was at the session', 'member profiles', 'capture decisions from slack', or any request to turn raw team activity into durable knowledge and a team-facing weekly story. Built for the team whose meetings evaporate and whose wins never get written down.
Install an AI project manager that lives in the team's Slack - posts the daily standup, scans the team channel through the day, captures action items, blockers, open loops, and decisions into one Tracker (a Slack List), sends end-of-day check-ins, and archives the week every Monday. Watchdog jobs verify every scheduled post actually ran. Nothing gets an owner unless that person committed in their own words, and the agent never claims a write it did not verify. Triggers on: 'slack pm', 'project manager agent', 'AI project manager', 'track action items in slack', 'daily standup bot', 'who owns what', 'standup automation', 'action item tracker', 'keep the team unblocked', 'what's blocked', or any request to run standups and track team commitments in Slack. Built for the team whose action items die in the scrollback.
ABACUS Turn any structured spreadsheet into a real, shareable, deployed dashboard in one Claude Code session — zero BI licensing cost. Takes a Google Sheet, Airtable base, or CSV the operator already keeps and stands up a live visual dashboard: define the one-to-three operating questions that must be answerable at a glance, map columns to metrics, build locally starting with ONE visible number, then push to GitHub Pages (or Vercel) for a shareable URL. The spreadsheet stays the source of truth; the dashboard is the view on top. Stack-adaptive, role-aware (owner / operator / practitioner each see their own cut), compliance-aware (sensitive data routes through proxy IDs). The question matters before the platform. Triggers on: 'turn my spreadsheet into a dashboard', 'build a dashboard from my Google Sheet', 'visualize this data', 'make this CSV into a dashboard', 'deploy a dashboard I can share', or any request to see structured data at a glance on a shareable page.
BATON Use when running multi-AI workflows across Claude, Codex, Cursor, or any agent tools and you keep losing context when an agent hits its limit or you switch tools mid-task. Triggers on "set up agent handoff", "stop losing context between agents", "handoff.md", "multi-agent continuity", "my agent stopped mid-task", "switch AI tools without starting over". Installs a tool-agnostic handoff.md protocol in your repo so any agent picks up exactly where the last one left off.
TEMPO Use when the member says "restructure my week", "optimize my calendar", "time block my schedule", "I'm drowning in meetings", "protect my focus time", or "plan my week" — takes a natural-language dump of their current calendar, priorities, and constraints and returns a restructured week with time blocks matched to task type, energy level, and deadline urgency. No calendar API needed.
BASTION Turns any technical problem you don't fully understand (stability, uptime, security, architecture) into a hired AI specialist that works from professional judgment, not shallow patches — and accumulates memory of how YOUR systems work. Built on the "Who Not How" pattern: instead of telling the AI what to do, you hire it for the role. Also a meta-pattern for converting ANY professional role into a self-improving skill file. Triggers on: "fix this bug but properly", "act as my SRE", "be my head of stability and security", "my site keeps going down", "diagnose this outage", "review my architecture for reliability", "hire an AI for this role", "turn this job description into a skill", "make this prompt remember my stack", "stop giving me shallow patches".
RECKON Document-driven tax and accounting return engine that ingests client financial documents, extracts fields, flags deductions and credits, runs compliance checks, and produces a review-ready return plus an advisory summary — shifting the professional from data entry to advisory work. Built for tax pros, bookkeepers, and accounting firms; the pattern generalizes to any document-heavy financial-prep workflow. Triggers on: 'prep this return', 'parse these tax documents', 'extract fields from these W-2s and 1099s', 'flag deductions and credits', 'run a compliance check on this return', 'build the advisory summary', 'turn tax prep into advisory work', 'ai tax prep', 'speed up return prep'. Keeps regulated PII in compliant systems — points at records, never copies them into the brain.
VAULT Give the company brain a second home it can't lose. The Atlas Folder holds the operator's most sensitive context, yet it lives on one machine, one account, no browsable map, no copy — one crash from gone. Extends `agentic-os-atlas-folder-setup`: confirms the Atlas structure, then mirrors it into an Obsidian vault so the brain is linkable, searchable, and recoverable, on a backup cadence you set. Read-then-sync only — additive and versioned, never overwriting the source of truth. Because the mirror arrived as a third-party plugin touching the most sensitive folder you own, it vets that plugin before anything installs. Triggers on: 'back up my atlas folder', 'mirror my company brain to obsidian', 'sync atlas to obsidian', 'make my brain browsable', 'obsidian vault for my second brain', 'protect my atlas folder from a crash', 'set up atlas backup cadence', 'is this obsidian plugin safe', 'vet this brain plugin', 'recover my atlas folder'.
GLIDE Founder away-mode orchestrator. When activated, auto-pilot-mode takes over operational management by coordinating inbox-triage, calendar-manager, slack-triage, team-accountability, crm-sync, and contact-manager according to founder-defined rules. Produces daily operational summaries, escalates exceptions, and maintains a full audit trail. VIP tier only — EA SUPREME expansion pack. Triggers on: 'turn on auto-pilot', 'activate auto-pilot', 'I'm going on vacation', 'going offline', 'away mode', 'disconnect mode', 'deep work sprint', 'conference mode', 'family time mode', 'auto-pilot status', 'how's auto-pilot doing', 'turn off auto-pilot', 'deactivate auto-pilot', 'I'm back', 'change auto-pilot rules', 'adjust auto-pilot', 'auto-pilot daily summary', or any request to delegate operational management during a founder absence. Does NOT trigger for single-skill delegation — route to the specific skill. Does NOT trigger for scheduling recurring tasks — route to schedule.
TEMPO Manages {{OWNER_NAME}}'s Google Calendar — daily overviews, time-blocking, conflict detection, buffer enforcement, meeting prep blocks, rescheduling, free-time finding, and weekly calendar audits. This is the EA managing your time. Triggers on: 'what's on my calendar', 'calendar overview', 'block time for', 'schedule focus time', 'do I have conflicts', 'add a buffer', 'move my meeting', 'reschedule', 'when am I free', 'find me a slot', 'calendar audit', 'meeting load', 'prep block', 'time block', 'deep work block', or any request involving viewing, creating, modifying, or analyzing calendar events. Routes FROM chief-of-staff for calendar-related requests. Does NOT handle cron/scheduled tasks (use schedule) or meeting transcript processing (use meeting-to-action).
HELM Master orchestrator for ALL operations — the top-level router for the entire AICOS system. Assesses your current state, determines what needs to happen, and routes to the right specialist skill. This is the DEFAULT entry point for all operational work. Triggers on: 'what should I be doing', 'what's the status', 'chief of staff mode', 'COS mode', 'run the machine', or any ambiguous operational request. Routes to event-chief-of-staff for event/show operations, and directly to specialist skills for everything else. When in doubt about which skill to use, this one activates first and routes appropriately.
TONIC Use when a registered dietitian, functional medicine practitioner, or allied health provider says "prep me for my client call", "summarize this session", "draft a follow-up plan", "fill in the protocol", "build a meal plan", or "help me interpret these labs". An AI copilot that synthesizes client history into pre-call briefs, turns call notes into post-call deliverables and protocol drafts, builds constraint-aware meal plans, and produces lab-discussion briefs — all under HIPAA-safe prompting discipline, with the practitioner keeping clinical judgment.
RELAY Orchestrate your outbound communication rhythm across email, Slack, iMessage, and social. Ensure the right people hear from you at the right frequency through the right channel. Triggers on: 'comms cadence', 'communication rhythm', 'who should I message', 'outreach schedule', 'stay ahead of comms', 'communication plan', 'who haven't I talked to', 'outbound rhythm', 'message queue', 'comms check', or any request to manage, plan, or optimize your outbound communication frequency and channel strategy. Owns general outbound communication across ALL contacts — if the user specifically mentions 'partners' or 'partnerships', defer to strategic-partner-manager.
VELLUM Use this skill whenever the user wants to create, read, edit, or manipulate Word documents (.docx files). Triggers include: any mention of 'Word doc', 'word document', '.docx', or requests to produce professional documents with formatting like tables of contents, headings, page numbers, or letterheads. Also use when extracting or reorganizing content from .docx files, inserting or replacing images in documents, performing find-and-replace in Word files, working with tracked changes or comments, or converting content into a polished Word document. If the user asks for a 'report', 'memo', 'letter', 'template', or similar deliverable as a Word or .docx file, use this skill. Do NOT use for PDFs, spreadsheets, Google Docs, or general coding tasks unrelated to document generation.
STEWARD Guided setup wizard for the EA SUPREME expansion pack. Walks through connecting additional MCPs (Slack, HubSpot, Granola, Firecrawl, Webflow), activating all 10 EA SUPREME skills (calendar-manager, slack-triage, contact-manager, project-ops, personal-logistics, crm-sync, content-publish-pipeline, travel-logistics, auto-pilot-mode, memory-command-center), testing each system, and configuring the founder's operational preferences. Requires base AICOS setup to be complete first. Triggers on: 'run EA SUPREME setup', 'set up EA SUPREME', 'activate EA SUPREME', 'install EA SUPREME', 'run-ea-supreme-setup', 'expand my EA', 'upgrade to EA SUPREME', 'I bought EA SUPREME', or any request to set up, configure, or activate the EA SUPREME executive assistant expansion pack.
Connects a member's Google Workspace to Claude Code — Gmail, Calendar, Drive, Docs, Sheets, Slides, Forms, Tasks, Chat, and Contacts, all ten services as one MCP server per account, multi-account capable (personal + work + client accounts side by side, each isolated). Walks through the one-time Google Cloud OAuth setup, then registers one MCP server instance per account with exactly the APIs and scopes that account needs. Trigger phrases: "connect my Google Workspace", "set up Google Workspace for Claude Code", "connect my Google account to Claude", "add my work Gmail and Calendar", "connect Drive and Docs to Claude", "connect another Google Workspace account", "set up Google Sheets/Slides/Forms/Tasks/Chat/Contacts in Claude Code".
POSTMARK Scan, prioritize, and draft responses for your email inbox. Connects to Gmail to surface urgent messages, categorize by action type, draft replies, and flag follow-ups for the Chief of Staff. Triggers on: 'triage my inbox', 'check my email', 'inbox triage', 'what emails need attention', 'email priority scan', 'draft replies', 'inbox zero', 'scan my inbox', 'what did I miss', or any request to process, prioritize, or respond to email. Also triggers when the morning-brief detects unprocessed urgent messages. When someone says 'what did I miss' — use this skill if the context implies email or messages; route to meeting-to-action if about meetings, or comms-cadence if about relationships.
SPLICE 🐰 Gmail Rabbit — Helps multiply the number of Gmail accounts you can easily connect to with Claude Code. Walks you through connecting multiple Gmail/Google Workspace accounts, each as its own MCP server instance. Invoke with: "Hey Gmail Rabbit" or "Gmail Rabbit, help me connect another account" Packaged and strategized by Big Jason. Former professional basketball player with 30+ years of email marketing/copywriting/strategy/product launch experience, over 1 billion dollars generated in email revenue and 18+ years in email deliverability. Text 702-901-9770 or email bigjason@emailresponsewarrior.com for your email needs.
VELLUM Use this skill whenever the user wants to do anything with PDF files. This includes reading or extracting text/tables from PDFs, combining or merging multiple PDFs into one, splitting PDFs apart, rotating pages, adding watermarks, creating new PDFs, filling PDF forms, encrypting/decrypting PDFs, extracting images, and OCR on scanned PDFs to make them searchable. If the user mentions a .pdf file or asks to produce one, use this skill.
VOYAGE Manages {{OWNER_NAME}}'s personal life logistics — birthdays, anniversaries, gift planning, personal appointments, maintenance schedules, vendor contacts, and errand tracking. The EA who remembers everything personal so {{OWNER_NAME}} never drops the ball on the people and things that matter most. Triggers on: 'birthday', 'anniversary', 'gift idea', 'what should I get', 'dentist', 'doctor', 'car service', 'oil change', 'personal appointment', 'schedule a haircut', 'errand', 'personal list', 'personal stuff coming up', 'vendor contact', 'who is my accountant', 'what's my dentist number', 'personal briefing', 'maintenance due', or any request involving personal life tracking, reminders, gift planning, or non-business logistics. Routes FROM chief-of-staff for personal logistics requests. Does NOT handle business calendar operations (use calendar-manager), business task tracking (use team-accountability), or business contact management (use contact-manager).
EASEL Use this skill any time a .pptx file is involved in any way — as input, output, or both. This includes: creating slide decks, pitch decks, or presentations; reading, parsing, or extracting text from any .pptx file (even if the extracted content will be used elsewhere, like in an email or summary); editing, modifying, or updating existing presentations; combining or splitting slide files; working with templates, layouts, speaker notes, or comments. Trigger whenever the user mentions "deck," "slides," "presentation," or references a .pptx filename, regardless of what they plan to do with the content afterward. If a .pptx file needs to be opened, created, or touched, use this skill.
GANTT Team project management through whatever tool you use: ClickUp (recommended), Asana, Notion, Monday.com, Linear, Trello, Jira, or a spreadsheet. Creates and tracks tasks, surfaces overdue and blocked work, generates sprint and status summaries, drafts nudges for slipped items, and balances workload. Stack-adaptive: it routes to your connected project tool. Confirms before creating, and never sends a nudge without approval. Triggers on: 'project ops', 'project status', 'how are my projects', 'create a task', 'add a task for [Name]', 'sprint summary', 'what got done this week', 'what is overdue', 'nudge [Name]', 'set up a project', 'my tasks', 'who is overloaded', 'workload balance', 'project dashboard'. Does NOT trigger for informal commitment tracking (route to team-accountability) or calendar management (route to calendar-manager).
GIRDER Hardens any AI-generated integration against external failures by adding retry, reset, warm-up, and unit-test layers to every external touchpoint, then ships a defensive-integration review checklist so brittle happy-path code becomes fault-tolerant. Triggers on: 'my scraper keeps breaking', 'make this integration resilient', 'add retry logic', 'my automation breaks when the site changes', 'this API call fails silently', 'harden my integration', 'why does my workflow go down when an upstream changes', 'add error handling to this', 'defensive coding prompt', 'review my integration for failure points', 'my browser automation is flaky'. Works for browser automation, web scrapers, REST APIs, MCP connectors, and webhook handlers.
KEYSTONE The ATLAS Cascade — guided install of an AI Chief of Staff onto the canonical Company Brain Atlas Folder. Walks the user through 5 sequenced phases (Atlas Folder → Tune → Link → Activate → Specialize) and ends with ATLAS (their Chief of Staff) plus one goal-matched department head — ACE, NOVA, SUNNY, IRIS, OLLIE, or SAGE — hired and working inside a live Company Brain Atlas Folder (soul.md, company-brain.md, root CLAUDE.md, Company/, AI-Team/, Daily-Notes). This is THE first-run entry point — the one thing a new member runs to get set up. Triggers on: 'set up my company brain', 'build my company brain', 'set me up', 'get me started', 'onboard me', 'run AICOS setup', 'install my chief of staff', 'set up AICOS', 'run aicos-setup', 'build my Atlas Folder', 'hire my AI chief of staff', 'train my AI agents', or any request to set up, build, or scaffold the company brain / AI Chief of Staff system from scratch. Also triggers when ops-brain detects no existing configuration.
METRONOME Create a scheduled task that can be run on demand or automatically on an interval. Triggers on: 'schedule this', 'run this every morning', 'remind me tomorrow at 3pm', 'set up a recurring task', 'make this a cron job', 'automate this on an interval'.
PRIMER Interactive wizard that walks you through creating your CLAUDE.md global instructions file — the foundation layer that configures Claude's behavior for every session. Collects your identity, operating rules, communication preferences, skill routing, decision stack, and Life-First values through a guided conversation, then generates and writes a customized CLAUDE.md to ~/.claude/CLAUDE.md. This is the first step of AICOS installation — without it, the COS persona is not active. Triggers on: 'set up my global instructions', 'create my CLAUDE.md', 'configure global instructions', 'setup global instructions', 'I need a CLAUDE.md', 'customize my Claude config', 'global instructions wizard', or when run-aicos-setup detects no CLAUDE.md exists at ~/.claude/CLAUDE.md.
CHIME Priority routing and triage for Slack messages. Scans channels and DMs, categorizes by urgency, generates daily digests, drafts responses, and searches conversations. Triggers on: 'slack triage', 'slack digest', 'what's happening on slack', 'summarize slack', 'summarize #channel', 'what did I miss in slack', 'any urgent slack messages', 'catch me up on this thread', 'find the slack message about', 'draft a slack reply', 'send a slack message at', 'schedule slack message', 'slack priority scan', 'urgent slack', or any request to scan, prioritize, summarize, search, or respond to Slack messages. Does NOT trigger for outbound communication cadence — route to comms-cadence for that. Does NOT trigger for general message drafting without Slack context — route to comms-cadence.
TALLY Track commitments, deadlines, and deliverables across your team. Surface who owes what by when, flag overdue items, and generate accountability reports. Triggers on: 'team accountability', 'who owes what', 'overdue items', 'commitment tracker', 'team check-in', 'accountability report', 'delegation tracker', 'what's outstanding', 'follow up on delegated tasks', 'team status', or any request to track, follow up on, or report on commitments made by or to team members. Also triggers when meeting-to-action creates action items assigned to others.
SEMAPHORE Keep one master skill in sync across a whole team with no re-installs and no version drift — the installed skill is a lightweight pointer that reads the master file from a shared cloud drive at runtime, so the owner edits one file and everyone gets the update on their next run. Use when several people run the same skill and updates keep going stale on copies. Triggers on: 'keep my skill in sync across my team', 'stop version drift on a shared skill', 'update a skill for everyone at once', 'team skill sync', 'shared skill on Google Drive', or any request to maintain one source of truth for a skill across multiple people.
VOYAGE Manages {{OWNER_NAME}}'s trip planning and travel logistics — flight research, hotel research, itinerary building, travel-day calendar blocks, timezone prep, packing checklists, and pre-trip briefings. The EA who has your boarding pass printed, your hotel confirmation ready, and a restaurant recommendation for dinner before you even ask. Triggers on: 'I'm going to [city]', 'plan my trip', 'find flights to', 'find hotels in', 'build my itinerary', 'block my calendar for travel', 'what time is it in [city]', 'timezone for [city]', 'what should I pack', 'packing list', 'brief me on my trip', 'pre-trip briefing', 'travel prep', or any request involving trip planning, travel research, itinerary creation, or travel-day logistics. Routes FROM chief-of-staff for travel-related requests. Does NOT book flights, hotels, or any paid service — researches, recommends, and organizes only. Does NOT handle personal appointment scheduling (use personal-logistics).
BEACON The visibility audit that shows a local or service business exactly how findable they are online — and what the gap is costing them in real dollars. Scans the Google Business Profile, reviews, the website (speed, schema, citation consistency), and presence across AI answer engines, then estimates the revenue leaking from the gap and hands back a prioritized 90-day plan in the owner's language — leads, bookings, did-the-phone-ring — not keyword jargon. Stack-adaptive: uses a web-scan connector (Firecrawl) where available, degrades to guided manual checks. Read-only and evidence-backed — every claim cites what it found, every dollar figure shows its math and is marked an estimate. Triggers on: 'visibility scan', 'invisible business', 'am I showing up', 'local SEO audit', 'google business profile audit', 'AI visibility', 'missed revenue audit', 'why aren't customers finding us', or any request to audit how findable a local or service business is.
LATTICE Suite of tools for creating elaborate, multi-component claude.ai HTML artifacts using modern frontend web technologies (React, Tailwind CSS, shadcn/ui). Use for complex artifacts requiring state management, routing, or shadcn/ui components - not for simple single-file HTML/JSX artifacts.
ABACUS Use this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved.
Connect YouTube research tools (search, video/channel stats, trending, transcripts, comments) to Claude Code — API key only, no OAuth flow, reuses your existing Google Cloud project if you already set one up for Gmail/Drive. Invoke with: "connect YouTube", "set up YouTube MCP", "pull YouTube data into Claude"