Skills

Use when a Claude Code session needs a capability it doesn't have locally (Google Calendar/Gmail/Drive, Slack DM, Discord, WhatsApp, etc.) and the user's AutoHub server has the tool wired in. Delegates one natural-language request to autojack via the local chat API and returns the result.

Recall, store, and associate memories via the AutoMem MCP server (mcp__memory__*). Invoke on user corrections ("actually", "no, I prefer", "stop doing X", "we decided X already", "I told you before"), decisions that stabilize after a round of pushback ("let's go with X", "ship it", "yeah that's the plan", "do it that way"), or articulated patterns ("I always X", "every time", "my thing is"). Also invoke when storing any fact that needs to outlive the session, or when a recalled memory needs to be updated, invalidated, or associated. Covers storage discipline, the atomic recall-store-verify-associate ritual, temporal validity, the never-store list, and the silent-fail verification workaround.

Understand AutoVault-managed skills. AutoVault syncs skills into the agent's normal skill directory, so loaded skills can be used directly without an AutoVault MCP server.

Control Jack's two Ulanzi TC001 LED matrix displays (32×8 pixels each) running AWTRIX 3 over the LAN — push notifications (scrolling text + color + icon + optional sound), set brightness, read battery / temperature / humidity, or identify which physical unit is which. Device-direct via the bundled `bin/awtrix` helper (raw HTTP, no Home Assistant, no secrets). Because AWTRIX runs its own app loop on-device and notifications overlay then auto-dismiss, there is NO takeover gate to manage (unlike the Divoom Pixoo). Trigger when Jack says "notify/show X on the awtrix", "put X on the office/bedroom display", "what's the battery/temp on the clock", or "locate the awtrix displays".

Design and harden compact, useful CLI installers, updaters, setup wizards, and command output surfaces without leaking debug logs or machine JSON into human UX.

Deploy and operate commerce apps on Cloudflare when a user wants Pages or Workers deployment, DNS/custom domains, storage bindings, secrets, logs, per-environment env vars, or optional Registrar domain registration handled by an agent.

Use when deploying an Astro + EmDash CMS site to Cloudflare, when a site serves raw source / downloads wrangler.jsonc / 404s at the root after an Astro migration, when every page redirects to /_emdash/admin/setup even after setup is complete, when CMS posts render with an empty title/body, when an admin preview link 404s, when adding a custom API route/contact form to an EmDash/Astro worker, when adding a second/new content collection to an already-live EmDash site (schema create/add-field), or when standing up EmDash with D1 + R2 and an admin login. EmDash/Astro deploys to Cloudflare Workers, NOT Pages.

Compatibility alias for babysit. Use when the user says "address Codex comments", "fix Codex feedback", or passes a PR number/URL with Codex review context; route to the unified babysit PR workflow.

Draft a conventional-commit-style message from the repository's staged changes. Uses git only; no external services or API keys required.

Capture, package, optimize, verify, and place documentation screenshots for GitHub READMEs, repo docs, and marketing/docs sites. Use for release screenshot passes, visual docs updates, screenshot QA, and reusable screenshot workflows across repos.

Research a person, business, or concept with primary sources and persist a durable AutoMem dossier through AutoHub's context-module write path (scripts/build-context-module.js). Use when asked to "build a dossier on X", "research X and remember them", or for pre-meeting/collab background that should survive the conversation. The stored dossier is the deliverable — verify it with a recall before reporting done.

Operate FlashSpace virtual workspaces on Jack's Mac via the installed `flashspace` CLI (v1.0.0 from FlashSpace.app). Switch profiles and workspaces, assign/unassign apps, manage floating apps, focus windows, inventory displays, and clean unassigned apps. Use when Jack says "switch workspace/space", "go to Code/Web/Chat/Terminal", "FlashSpace", "assign this app", "float/unfloat", "list workspaces", "hide unassigned", or wants a dual-display workspace change. Prefer this over reinventing Mission Control / yabai / AeroSpace scripts.

Bootstrap, inventory, verify, and maintain AutoHub Home Assistant MCP access as a reusable AutoVault skill.

Render fixed-size HTML/CSS design exports into production raster assets when users need social cards, favicons, docs images, app-store screenshots, or other web-rendered artwork captured, dimension-checked, and prepared for merge.

Triage Jack's Gmail unread inbox into action buckets (!Action / !AI Handle / !Waiting On / archive) using the google_workspace MCP, then post one digest. Use this whenever Jack asks to triage, clear, sort, process, or "do" his email or inbox — "inbox triage", "clear the unread pile", "go through my gmail", "sort my inbox", "process my email", "deal with my unread" — even if he never says the word "skill". It enumerates all unread, runs a small parallel read-only classifier fan-out, shows the buckets for approval, then applies Gmail labels SEQUENTIALLY and posts a "Worth your attention" + "Security roundup" digest. It mutates the live inbox, so it always gets approval before writing.

Guide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK).

Maintain AutoHub MCP server configuration, tool registry scans, health checks, and host config repair through a standardized AutoVault skill.

Add Pirsch Analytics to a website, create or inspect the domain with explicit authorization, install and verify the tracking snippet, configure optional DNS verification and custom events, and wire a scoped MCP client without exposing credentials.

Operate and partially configure Raycast on Jack's Mac via deeplinks, Script Commands, file-based AI providers/skills, and AI Command import packs. Use when Jack says "open in Raycast", "run the X Raycast command", "deeplink", "Script Command", "AI Command/Agent/Skill", "providers.yaml", "Import AI Commands", "PromptLab", clipboard history, Superwhisper toggle, or wants Autohub Raycast scripts installed (delegate install/sync to raycast-autojack). Prefer this over inventing osascript launchers. Not a full settings API — see Supported config surfaces.

Use when preparing to tag or cut a release of a WordPress plugin (or any repo that carries its version in readme.txt "Stable tag" and tags releases as bare semver). Triggers on "is this ready to tag", "prep the release", "release readiness", "get the release ready", "release preflight", "what's left before we ship X.Y.Z", or any moment just before a release tag. Audits everything merged since the last git tag, surfaces the gaps (undocumented features, missing changelog entries, dirty working tree, stale branches/worktrees, open PRs, version not bumped, failing tests), then — after one approval — closes the in-repo gaps it can (drafts the changelog, bumps the version, dispatches the repo's docs/verification agents, runs the suite behind a safety guard) and stops at a green "ready to tag" state. It NEVER pushes, tags, or publishes anything: every action that leaves the repo stays the human's decision.

Operate the Resend platform from the terminal — send emails (including React Email .tsx templates via --react-email), manage domains, contacts, broadcasts, templates, webhooks, API keys, logs, automations, and events via the `resend` CLI. Use when the user wants to run Resend commands in the shell, scripts, or CI/CD pipelines, or send/preview React Email templates. Always load this skill before running `resend` commands — it contains the non-interactive flag contract and gotchas that prevent silent failures.

Diagnose recent AutoHub runtime incidents from logs and live health evidence, separating first causes from retry cascades, process-local database poison, credential-copy drift, platform recipient or permission failures, and intentional fallbacks. Use when AutoHub logs show repeated errors, stuck writes, webhook retry storms, MCP authentication failures, or regressions after a restart.