The Dec-2024 taxonomy stays chapter 1 — this release adds chapter 2 and makes the patterns runnable.
🗞️ What Changed 2025-2026
A dated, sourced map of everything that moved since Building Effective Agents: context engineering absorbed prompt engineering · the multi-agent debate (Anthropic's multi-agent research system vs Cognition's Don't Build Multi-Agents — and the 2026 synthesis) · fewer-better tools + code execution with MCP · the de-facto config stack (AGENTS.md · SKILL.md/Agent Skills open spec · MCP stateless core · hooks) · harness engineering · sandboxing & budgets · evals. → foundations/what-changed-2026.md
📜 Patterns as Code
The five workflow patterns as real, runnable .nika.yaml files — offline mock model, no API key, each one statically checked before a single token would be spent. They are now proven in CI on every push (the badge on the README is live). Writing them even caught a genuine DAG bug pre-run — which is rather the point. → patterns-as-code/
🐛 Repairs
The MIT badge pointed at a LICENSE that didn't exist (it does now) · CONTRIBUTING.md was linked but absent (written) · 12 dead links to the old concepts/ tree remapped · glossary +5 entries (AGENTS.md · Context Engineering · Harness · MCP · SKILL.md) · editor noise untracked.
Every claim in the 2026 page carries its date and primary source. Disclosure: the runnable files use Nika, built by this repo's author — alternatives are named in the folder README.