Systematic debugging for the Ollama Agent Harness — reproduce, localize, reduce, fix, guard
Git workflow for the Ollama Agent Harness — atomic commits, small changes, safe history
Security hardening for the Ollama Agent Harness — input validation, command injection, secrets, deny-first review
Build a new MCP (Model Context Protocol) server from scratch — a stdio program that exposes tools to AI agents. Use when the user wants to wrap an API, CLI, or internal service as an MCP server the harness (or Claude Desktop, Cursor, etc.) can load. Adapted…
Project conventions for the Ollama Agent Harness — architectural patterns borrowed from the Claude Code paper
Passively get to know the user over time — distill durable facts about who they are, what they work on, and how they like to be helped from conversation history into USER.md and memory. The home-user analog of Microsoft Work IQ's implicit-memory layer. Use…
Build and maintain a Karpathy-style LLM-maintained personal wiki (a "Second Brain") as plain markdown the user owns. Use when the user wants to set up a second brain, capture/ingest a source into their notes, ask their knowledge base a question, file…
Evidence-first research — gathers sources on a topic, synthesises findings, and cites every claim. Marks what could not be verified instead of guessing.