AI Agent Toolbox
4 posts in this series
Continuum and Choosing an Agent Runtime: 7 Capabilities to Check From Notebook to Production
Your AI agent runs fine in a notebook but falls apart in production? Choosing an agent runtime is really choosing the layer missing between toy and production. Using ShyftLabs' Continuum, this post breaks down the 7 dimensions to evaluate: orchestration, model routing, memory, tools, durability, observability, and governance.
guizang-social-card-skill: A Reusable Pipeline for Rednote Cards and WeChat Covers
Learn what guizang-social-card-skill does, how to install it in Claude Code or Codex, and how to check layouts, themes, rendering, licensing, and QA before publishing.
female-portrait-director: Turn Portrait Prompts into a Reusable Skill
female-portrait-director is an open-source project (V1.4.1, MIT) that turns AI portrait prompts into a reusable Skill. This article breaks down its parameter locking, on-demand routing, director-style expansion, and safety boundaries, plus a 5-step method to turn your own prompts into a Skill.
ADHD: Fixing Premature Convergence in Coding Agents with Parallel Divergent Reasoning
ADHD is an open-source skill built on the Claude and Codex Agent SDK that fights premature convergence in coding agents using isolated parallel divergence plus a separate critic pass. Here is how it differs structurally from Chain-of-Thought and Tree-of-Thought, when to reach for it, how to install and trigger it, and what a single run costs.
