Blog
Field notes from the Maximus workhorse engine — skill releases, AI agent architecture, research, and AI SEO.
9 posts
- ·4 min read
Maximus grows to 57 — your local model isn't reading the manual
Five skills for running Maximus on open-source local models: OpenClaw on Ollama, Ollama operations, reliable tool calling from small models, evaluation on your own tasks, and local fine-tuning. The finding behind them: on most consumer machines, the default setup silently cut off most of Maximus's own prompt. Suite now at 57 skills.
Read the post → - ·6 min read
Maximus grows to 51 — hallucination is usually an architecture bug
Two skills for small local models: maximus-slm-local-stack for hardware, KV-cache VRAM math, quantization and runtime, and maximus-local-research-assistant for grounded local RAG with a refusal contract, span citations, and verification. The premise: “it makes things up” is usually an architecture bug, not a model bug. Suite now at 51 skills.
Read the post → - ·4 min read
Maximus grows to 49 — the Film Pack, for working filmmakers
Five opt-in skills for people who make films with cameras, cast, and a call sheet: Fountain screenplay formatting to industry-standard PDF and .fdx, scene breakdown to stripboard and day-out-of-days, AICP account-code budgeting, post pipeline, and Filmhub delivery. Feature-length only, no AI slop. Suite now at 49 skills.
Read the post → - ·5 min read
Maximus grows to 44 — the approval treadmill
mtt-claude-cursor — the operating model for development driven by Claude Code inside Cursor. Pre-authorizes routine reversible in-repo work, keeps hard stops where they belong, recovers interrupted sessions from git state, and fixes command shape instead of weakening protections. Suite now at 44 skills.
Read the post → - ·4 min read
Maximus grows to 43 — adding the Transaction Analyst
maximus-transaction-analyst — turn a dense deal folder (emails, term sheets, closing docs, wires) into a two-page executive transaction memo. Reconstructs chronology, reconciles numbers, separates facts from allegations, and names what's still unresolved. Suite now at 43 skills.
Read the post → - ·5 min read
Three skills that make Maximus research like a person
maximus-chain-of-verification, maximus-investigative-research, and maximus-literature-review — the reasoning-quality layer, the reporter's discipline, and the researcher's PRISMA workflow. Suite now at 42 skills.
Read the post → - ·3 min read
Two Kimi K3 skills join Maximus
maximus-k3-model-selection and maximus-k3-self-hosting — pick the right frontier model when K3 is on the shortlist, and run K3 on your own GPUs (or honestly decide not to).
Read the post → - ·3 min read
Maximus grows to 37 — adding Contact Intelligence
A new skill for finding professional business emails from a LinkedIn URL — with pattern discovery, verification, and honest confidence scoring. Never labels a guess as verified.
Read the post → - ·5 min read
Introducing the Maximus Suite — 36 skills for AI-native engineers
36 skills. 5 pillars. One workhorse. The full suite for engineers, founders, and scientists building with AI — from cognitive OS to AI SEO.
Read the post →