jkrumm

Read and write the second brain — a git-backed Obsidian vault at ~/SourceRoot/brain (PARA structure + a top-level wiki/ knowledge tree). Use whenever the user mentions "brain", "second brain", "knowledge base", "note", "remember", "look up", "find note", "capture", "wiki", or "okf". Also use for any request to save, retrieve, or organize personal knowledge that belongs in the brain rather than in a repo-specific doc.

Write, draft, and polish prose through a human-owned 7-step pipeline. Use whenever the user says "distill", "write article", "draft", "summarize into", "create content", "synthesize", or "publish". Also use for any task where the output is a piece of long-form writing — the pipeline keeps prose quality out of coding sessions.

RETIRED — manage the (now decommissioned) per-machine mlx-audio TTS/STT stack; TTS/STT moved to the cloud audio-gateway (VPS)

Operate the Mac mini remote dev host — connecting from the MacBook over mosh or herdr --remote, running and reattaching many Claude Code agents, herdr workspaces/panes and its socket API, claude --bg durable daemons, the Uptime Kuma readiness heartbeat, and the failure modes specific to a headless always-on Mac. Use when the user mentions herdr, mosh, the mini, "remote dev", "dev host", detaching or reattaching agents, "did my agent survive", sessions dying on lid-close, or asks how to reach a dev server running on the mini.

Manage the personal image stack — public CDN uploads/transform URLs and the private image-share layer. Use whenever an image needs a URL (public or private) for an article, blog post, vault note, README, OpenGraph tag, or a one-off share; or when the user says "upload this image", "share this image", "private image", "publish image", "host this screenshot", "get me a CDN link", "resize this", mentions "image-share", or asks where an image lives.

Bidirectionally sync ALL git repos across two machines — this MacBook and the always-on Mac mini — over SSH, using GitHub/GitLab as the transport. Commits and pushes uncommitted work on both sides, then rebases/fast-forwards each so they converge, with per-repo subagents resolving rebases and merge conflicts. This is a MULTI-repo, TWO-machine operation — distinct from /commit, /ship, or /git-cleanup, which act on a single repo. Use whenever the user wants to sync their machines/repos, "catch up" the laptop to the Mac mini or push laptop work back, is about to travel or just got back, has uncommitted work scattered across many repos, or wants to reconcile branches that diverged between the two machines. Trigger even on a bare "sync", "sync my machines", or "get my repos up to date".

Run validation (format, lint, tsc, test) via sideclaw MCP tool

Smart git commit with conventional commits, OpenSpec awareness, and intelligent splitting for GitHub

Hand repo work to a bounded Claude Code episode via the `hermes-cc.sh` verb dispatcher, then answer with its verdict. Use when triage needs the actual source — "why is X failing, look in the repo", "what changed in Y", "warum ist Z rot, schau ins repo", "read the code and tell me", "check the repo for", a red monitor whose cause is code-shaped, a stale GitHub issue, or any question you can only answer by guessing otherwise. Also the path for "file an issue about what you find" (author tier) and, only with Johannes's explicit confirmation, "make that change" (implement tier — isolated worktree, draft PR), then "merge it" to land that PR.

Update Hermes Agent to latest version, resolve conflicts in locally-modified source files, and restart the gateway

Test, observe, and improve Hermes Agent behavior — send test messages, read session traces, identify routing failures, and fix SOUL.md/SKILL.md

Read, search, and write Johannes's Obsidian vault (the PARA second brain at ~/SourceRoot/brain) via the Obsidian CLI (metadata-aware — backlinks, tags, Dataview) with a filesystem fallback. Capture notes to the inbox, create resource/inspiration notes with the right frontmatter, search by text/tag/backlink.

Diagnose and remediate homelab / VPS / Mac Mini infrastructure through the bounded `hermes-ops.sh` verb dispatcher — monitors down, unhealthy or restart-looping containers, cron failures, push-monitor heartbeat gaps, uptime-kuma config corruption, deploy drift. Read-only triage is free; every mutation is a named verb needing --confirm and --why. Use for "#alerts" messages, "what's down", "is everything up", "why is X red", "restart Y", "redeploy Z".

Call the argo REST API (https://argo.jkrumm.com/api) for TickTick tasks, Gmail, Calendar, Docker (homelab + VPS), UptimeKuma, Slack, weather, Garmin Health, Strength tracking, WalkingPad treadmill stats, AI usage/cost, user profile, and read-only SQL — use curl with Bearer $HOMELAB_API_KEY

TTS audio synthesis via the VPS audio-gateway (https://audio-gateway.jkrumm.com/v1) — Gemini 3.1 Flash TTS, voice "Charon", OpenAI-compatible /v1/audio/speech. Use for any spoken output: morning/evening briefings, voice memos, ad-hoc TTS.

Recommend what Johannes should read next — personal book recommender grounded in his actual taste. Use when he asks "what should I read", wants book recommendations, wants to find his next novel, wants fiction, fantasy, sci-fi, thriller, or travel/adventure picks, or wants to turn his reading history into picks. Pulls taste from the Argo Reading API (Hardcover shelf — ratings, genres, read / want-to-read) + a durable taste profile in Obsidian, researches real matching books on the web, ranks them, and captures picks back to Obsidian.

Add per-person staff logins to a vibe-stack app's website using Cloudflare Access — the agent sets it up ITSELF via the Cloudflare API (works on the free workers.dev URL, no domain). Use only when SEVERAL people each need their OWN login (their own email); a single owner is already private via the access key. Triggers: "mein Team soll sich einloggen", "Mitarbeiter-Logins", "jeder mit eigener E-Mail", "staff logins".

Turn on AI inside a vibe-stack app's server so the WEBSITE (and autonomous tasks) can summarise, translate, or parse free text into saved data — using Cloudflare Workers AI, no API key and no extra account. Use when the owner wants the app itself to be "smart": "fass das zusammen", "übersetze", "erkenne automatisch", "aus dem Text rausziehen", "ich tippe einfach 'handvoll Erdbeeren mit Skyr' und es trägt sich ein", a free-text field that fills in fields by itself, or an AI step in a server task.

Manage Cloudflare resources for a vibe-stack app via wrangler — create the D1 database, create an R2 storage bucket, add a secret/API key, query D1 from the CLI, and regenerate types after editing wrangler.jsonc. Use when the owner mentions database, storage, bucket, secret, api key, binding, Cloudflare, Datenbank, or Speicher.

Connect a vibe-stack app to Claude as a custom connector so the owner can use it from anywhere — phone, desktop/web Chat, Projects, and routines, not just the Code tab. Provisions the OAUTH_KV namespace and the OWNER_SECRET access key, deploys, walks the owner (in German) through adding the custom connector, and generates the claude-setup/ paste bundle. Use when the owner wants to use/manage the app from their phone or outside the Code tab, "von überall", "auf dem Handy", "mit Claude verbinden", "Connector einrichten", or right after the first deploy.

Publish a vibe-stack app live, read its public URL, view logs, and roll back. Use when the owner says deploy / publish / go live / roll back / veröffentlichen / online stellen / rückgängig.

RARELY needed — vibe-stack is ONE growing app, so a new thing is normally a new page (use add-page / add-data), NOT a new app. Only scaffold a fully separate app (its own DB + R2 + URL) when the owner explicitly insists on an isolated environment — confirm they don't just mean a new feature first. Then clone the kit, copy the boilerplate to ~/vibe-apps/<name>, rename, npm install, connect Cloudflare, deploy. Triggers: "wirklich eine eigene, getrennte App", "komplett separate app", "neu anfangen".

Set up a vibe-stack app as a BUSINESS / operations tool the owner runs by talking — define the things they manage (reservations, employees, orders, inventory, …) and the rules, then wire the data, a domain operating-manual, and a plain-language guide so the agent can run the live data. Use when the owner wants to manage a business, a team, bookings, or a shop (not just track personal data), e.g. "ich will Reservierungen und Mitarbeiter verwalten", "für mein Restaurant", "meinen Betrieb organisieren".

Read and change the LIVE data in a vibe-stack app by talking — list, find, add, update, or cancel real records (reservations, employees, orders, inventory, …). Use when the owner wants to SEE or CHANGE actual data in their running app (not build a feature), e.g. "zeig mir die Reservierungen heute", "trag Müller für 4 um 19 Uhr ein", "wer arbeitet Freitag?", "storniere Tisch 5", "ändere die Telefonnummer von Anna".

Full audit of the claude-remote project — fix OpenAPI schemas (t.Any → typed), align CLAUDE.md with reality, trim duplication from watchdog instructions, validate and deploy

Add Telegram as a channel. Can replace WhatsApp entirely or run alongside it. Also configurable as a control-only channel (triggers actions) or passive channel (receives notifications only).

Add new capabilities or modify NanoClaw behavior. Use when user wants to add channels (Telegram, Slack, email input), change triggers, add integrations, modify the router, or make any other customizations. This is an interactive skill that asks questions to understand what the user wants.

Query GitHub — list repos, check notifications, view open PRs and issues, get recent activity. Calls the GitHub API via the claude-remote-api passthrough proxy (PAT is injected server-side). Use when the user asks about GitHub repos, PRs, issues, notifications, or code activity.

Query homelab infrastructure status — check UptimeKuma monitor uptime, get monitor summaries, and read ntfy topic messages (e.g., homelab-watchdog alerts). Use proactively when the user asks about server status, uptime, alerts, or infrastructure health.

Sync documentation with code changes. Audits git diff and updates ARCHITECTURE.md, DECISIONS.md, TASKS.md, CLAUDE.md, and README.md to reflect current implementation.

Efficiently develop Hono applications using Hono CLI. Supports documentation search, API reference lookup, request testing, and bundle optimization.

Mastra v1 API patterns and examples. Reference material for agents, tools, workflows, and streaming. Based on latest @mastra/core docs.

TanStack Query v5 patterns for React. Reference for queries, mutations, infinite scroll, and cache management.

Design and restyle data visualizations and analytics UIs in a basalt-ui app (Mantine v9 + visx). The dataviz successor — restraint doctrine plus the live theme-lab tuning loop and the centralized, lint-enforced --vx-* token system. Use when building or restyling charts/dashboards, choosing a palette, tuning dark/light, or when charts look childish / inconsistent / over-colored.

Scaffold or refresh a basalt-ui app — guided wrapper around 'basalt-ui init' (first-time scaffold) and 'basalt-ui sync' (drift-managed refresh). The whole agentic layer ships in the npm package — init places the skills, rules, CLAUDE block, and DESIGN.md; sync keeps the managed half fresh. Use when setting up basalt-ui in a new/existing app, wiring the provider/shell/vite preset, or when the doctrine feels missing.

Add or extend a chart in a basalt-ui app — pick a kind vs go bespoke (Rule of Three), register a consumer series with defineSeries/groupTokens (since VX.series is app-side, not shipped), compose the visx primitives, and respect the @visx-only-in-charts + Mantine-free boundary. Use when adding a new chart, a new metric/series color, or extracting a repeated chart shape into a kind.

Upgrade assistant for ALL manually-managed Docker containers with dependency checking, shared infrastructure awareness, and breaking change analysis

HomeLab documentation maintenance — audit infrastructure changes and update README.md, CLAUDE.md, docs/, and skill files

Full health audit of the homelab server across 8 phases, then offer to fix each issue found

Full health audit of the VPS — system resources, containers, Cloudflare tunnel, Tailscale, errors, updates, backup status, and manual upgrade checks

Documentation maintenance — sync compose files against README/CLAUDE.md service tables, verify Secrets section coverage, check Makefile targets

Analyze the modelpick catalog, metrics, and probes to recommend the optimal model for a given use-case, and surface My Stack drift. Trigger when the user asks which model to use for something (scored fast/coding/orchestrator/tts/stt, the research-only embedding/vision/image categories, or a custom need), whether a better model exists for a task, to compare candidate models, or to review/refresh their decided stack.

Refresh the IU model catalog in modelpick from a check-key portal HTML export — parse, diff, sync, seed, optionally probe, and commit the snapshot. Trigger when the user mentions a new IU model list, the check-key page (ue-self-service.app.iu-it.org/check-key), an HTML/export they saved, or that the modelpick catalog looks stale or missing recent models.

HomeLab TickTick proxy API reference — endpoints, task model, auth, and code patterns for argo.jkrumm.com/api

Raycast extension development reference — command modes, UI components, hooks, APIs, and patterns for building the TickTick extension

Create releases with AI-enhanced GitHub release notes for Free Planning Poker

Audit linewatch's monitoring chain — collector, heartbeat, watchdog, alerting — then hunt for the defect classes this repo has actually shipped. Use when asked to check the watchdog, the healthcheck, whether alerting works, whether it is safe to arm, or to look for improvements.