Skip to content

AI Toolkit

Brian's suite of Claude skills and supporting infrastructure

Overview

Since March 2026, Brian has been building a personal AI toolkit — a collection of Claude skills that share infrastructure and work together. The philosophy is: build shared resources once (API keys, model registry, context system), then layer specialized skills on top.

Components

Built and Configured

Context Manager — Persistent memory across conversations. Stores chat summaries, file versions, and a synthesized knowledge base. Backs up nightly to Dreamhost. Published as a browsable website at context.rootphi.net.

LLM Triangulator — Queries up to 7 LLM providers in parallel and synthesizes a best answer. Supports "quick check" (economy models) and "deep dive" (flagship models) modes. Built with shared api-keys.json vault and model-registry.json that auto-refreshes weekly.

Planned

Image/Video Critique Skill — Would use vision-capable LLMs and generative models (FLUX, Kling, Sora via Together AI) to critique and produce visual content. Will extend the existing model registry and API key vault.

Shared Infrastructure

Resource Location Purpose Maintenance
api-keys.json /mnt/.claude/ Central API key vault (7 providers) Manual (rotate as needed)
model-registry.json /mnt/.claude/ Tagged model catalog (25+ models) Auto-refresh weekly (Monday 9am)
claude-context/ ~/Documents/ Persistent memory + knowledge base Auto-backup nightly (10pm)
balance-checker.py /mnt/.claude/ Credit balance monitoring Auto-check daily (8am)

Architecture Decisions

  • Shared vault over per-skill key storage (avoids duplication)
  • Auto-refreshing registry over static model lists (stays current)
  • Local-first with cloud backup over cloud-first (works offline, fast)
  • Markdown over database (human-readable, portable, Claude-native)

Source Chats