claude_onboarding_agent

Set up Claude to build and maintain a structured knowledge base using the Karpathy LLM Wiki pattern, conformant to Google's Open Knowledge Format (OKF v0.1) — works with codebases, personal notes, or both. Integrates with Obsidian via the official Obsidian CLI (token-efficient, dispatched through a dedicated subagent — no always-on MCP overhead).

Guided onboarding orchestrator — scans your repo, infers your use case, and dispatches to the right setup skill. Run this if you're new and want a personalized Claude Code setup.

Read-only audit of your current Claude setup — prints a prioritized HIGH/MEDIUM/LOW findings list covering package manager hygiene, CLAUDE.md quality, permissions, git hygiene, and MCP servers, without changing any files. Power-user stage of /checkup — invoke directly when you want the raw findings report without a verdict or follow-up actions.

The single user-facing maintenance entrypoint for an existing Claude setup. Health-checks your configuration, decides whether it is fine as-is, needs an upgrade, or needs a rebuild, and takes that action — handing off to /upgrade-setup or /onboarding --rebuild as needed. Use when you want your setup checked and fixed, not just a findings report.

Set up Claude for the input side of academic research — literature reviews, paper screening and summaries, reference management with Zotero, annotated bibliographies, and reading notes for your research domain. For drafting the manuscript itself, use academic-writing-setup.

Set up Claude for academic writing — thesis, paper, or dissertation. Configures LaTeX/Typst stack, bibliography (Zotero + Better BibTeX), citation style, and strict no-invented-citations rules so Claude helps you write without hallucinating references.

Refresh the anchor-derived marker sections in this project's CLAUDE.md and AGENTS.md by re-fetching the latest anchors, diffing against on-disk content, and applying confirmed changes. Supports --dry-run.

Set up Claude for software development — installs Superpowers, configures an iterative coding workflow (brainstorm → plan → subagents → review → commit), and generates CLAUDE.md, AGENTS.md, and settings.json tailored to your stack.

Set up Claude for on-brand content voice. Generates per-platform voice rules keyed on your brand voice and target audience. Does NOT cover publishing, analytics, or media pipelines — this skill is scoped to voice only.

Set up Claude for data science and ML engineering — configures notebook workflow, experiment tracking, reproducibility conventions, and data layout so Claude supports you from exploration to productive modeling.

Set up Claude for UI/UX design tooling — wires up your design tool (Figma, Sketch, XD, Penpot) and the Figma MCP handoff. Pair with web-development-setup for the frontend stack, testing, and accessibility rules.

Set up Claude for DevOps and cloud engineering — configures your cloud provider, IaC tool, and CI/CD platform so Claude helps with infrastructure planning, safe applies, and security reviews.

Set up Claude for business writing — emails, memos, reports, proposals. Configures writing style, document-type preferences, and company context so Claude produces on-brand text. Does NOT cover presentations or slide decks.

Power-user tool — normally invoked via /checkup. Re-apply current best practices to an existing onboarding-agent setup. Detects the original setup, computes per-section diffs against the latest templates, asks for per-change confirmation with a unified-diff preview, backs up everything it will touch, and never touches content outside the plugin's delimited sections. Supports --dry-run.

Set up Claude for web development — frontend, backend, or full-stack. Configures framework-aware tool permissions, styling/linter conventions, env-var hygiene, and deploy-target pointers so Claude ships production-ready web apps from day one.

Install and wire up Graphify — an open-source knowledge-graph tool for Claude Code. Registers a /graphify slash command and a PreToolUse hook that consults a local graph before file-search tool calls, dramatically reducing token cost on large codebases, docs, and mixed-media corpora. Handles code (25 languages via tree-sitter), Markdown, PDFs, diagrams, images, audio/video.