Build Workflows v3.1, Expanded Agents Matrix, and Streamlined Architecture Decisions¶

Three updates that tighten the framework's Build step and expand platform coverage for agents.

Building Workflows skill upgraded to v3.1¶

The building-workflows skill now enforces a structured gate between Design and Construct. Step 3.1 (Design) opens with a plan mode prompt so you collaborate on architecture before any artifacts are generated. A spec approval checkpoint blocks progression until you explicitly approve the AI Building Block Spec. Step 3.2 (Construct) then offers a build path choice — the model generates artifacts, or you build manually from the spec with a construction guide. Integration research (web search) is deferred from Design to Construct so it doesn't interrupt planning. New reference specs for the agentskills.io skill format and Claude Code subagent format ensure generated artifacts follow proper conventions.

Agents by Platform matrix expanded¶

The Agents index page now shows one row per agent approach across all four platforms — including browser-based offerings (Claude in Chrome, Atlas Browser, Gemini in Chrome) and the Claude Agent SDK. Each row links to official docs and the corresponding platform guide.

Architecture decisions simplified¶

The Design phase's architecture checklist has been replaced with an extract-then-confirm pattern. Instead of eight sequential questions, the skill asks one question (platform), extracts tool integrations, trigger/schedule, and browser access from the Workflow Definition, then presents a single confirmation block. Deployment surface, shareability, and code comfort are deferred to Construct where they're actually needed.

Skills & Agents pages repositioned as platform-agnostic¶

The Agents & Skills section now leads with GitHub download as the primary access method across all four detail pages, with Claude Code plugin install as a secondary option. Verbose platform tabs replaced with concise "Using These Skills" sections.