using-workbench

Route development tasks into quick, medium, or large workflow lanes.

Updated Mar 18, 2026
One-click install
npx skills add https://github.com/pgoell/pgoell-claude-tools --skill using-workbench-pgoell
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: using-workbench
Source: https://github.com/pgoell/pgoell-claude-tools/tree/main/plugins/workbench/skills/using-workbench
Command: npx skills add https://github.com/pgoell/pgoell-claude-tools --skill using-workbench-pgoell

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Deciding how much process a coding task needs is hard: too little process risks expensive mistakes, too much wastes effort on trivial changes. This Skill triages every task into the right workflow lane before work begins. ## Core Features & Use Cases - Reflex Ladder: Runs a six-rung check (YAGNI, existing code, standard library, platform features, installed dependencies, one-liners) to decide whether anything should be built at all. - Three-Lane Triage: Routes tasks into Quick (no design artifacts), Medium (one combined design/plan document), or Large (full brainstorm, spec, and plan pipeline) based on blast radius and reversibility. - Always-On Disciplines: Enforces test-driven development, systematic debugging, and verification-before-completion in every lane. - Use Case: A user asks to add a feature. The Skill first checks whether existing code already covers it, then routes the work to the medium lane with a single combined design document instead of a heavyweight spec. ## Quick Start Ask the agent to load the workbench rules and triage your next coding task into the right workflow lane before writing any code.

Frequently Asked Questions about using-workbench

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I decide how much process a coding task needs?

Route the task by cost of misunderstanding, blast radius, and reversibility rather than line count. Quick lane covers unambiguous small changes, medium covers one coherent feature, and large covers new subsystems or expensive-to-reverse decisions.

How does the workbench triage decide between quick, medium, and large lanes?

Quick lane handles unambiguous, easily reversed changes with no design artifacts. Medium lane produces one combined design and plan document. Large lane runs the full pipeline of brainstorming, spec, and plan for cross-cutting or high-risk work.

Does using-workbench work with the superpowers plugin installed?

Yes, both meta-skills can fire at session start and coexist. When a slug exists in both plugins, such as brainstorming or test-driven-development, the workbench version takes precedence and bare slugs resolve to workbench skills.

Can I use workbench skills in Codex instead of Claude Code?

Yes, skills are auto-discovered in Codex and activate natively. The references/codex-tools.md file maps Claude Code tool names like Task and TodoWrite to Codex equivalents such as spawn_agent and update_plan.

When should a task skip the build step entirely?

Run the six-rung reflex ladder first: check whether the work is needed at all, already exists in the codebase, is covered by the standard library, a native platform feature, an installed dependency, or a one-liner. If any rung holds, do not build it.

What happens when user instructions conflict with workbench skill rules?

User instructions always win. CLAUDE.md, AGENTS.md, or direct requests outrank skill rules, which in turn outrank the default system prompt. If the user says skip TDD, the skill defers to that instruction.