Skills

Agent skill files that teach your AI coding assistant how to install, configure, and use workbench.

What Are Skills?

Skills are instruction files that teach your AI coding agent how to work with workbench. When installed, your agent can automatically install workbench, configure agents and profiles, and write execution plans — all from natural language prompts.

Workbench ships three skills:

Installing Skills from the Download

If you don't have workbench installed yet, you can bootstrap the process by downloading the skills and placing them manually. Your agent will then use the install-workbench skill to guide you through the rest.

1. Download the skills zip using the button above

2. Unzip into your project's .agents/skills/ directory:

mkdir -p .agents/skills
unzip workbench-skills.zip -d .agents/skills/

This creates:

.agents/skills/
├── install-workbench/
│   └── SKILL.md
├── configure-workbench/
│   └── SKILL.md
└── use-workbench/
    └── SKILL.md

3. For Claude Code users, also copy to .claude/skills/:

mkdir -p .claude/skills
cp -r .agents/skills/install-workbench .claude/skills/
cp -r .agents/skills/configure-workbench .claude/skills/
cp -r .agents/skills/use-workbench .claude/skills/

4. Ask your agent to install workbench — it will now have the install-workbench skill and can walk you through the full setup:

"Install workbench and set up this repo"

Installing Skills via wb setup

If you already have workbench installed, wb setup handles skill installation automatically:

wb setup                    # auto-detect your agent platform and install skills
wb setup --agent claude     # explicitly install for Claude Code
wb setup --update           # force-update skills to the latest version
wb setup --symlink          # symlink instead of copy (stays in sync with package updates)

Using Skills

Once installed, skills activate automatically based on what you ask your agent to do:

install-workbench

Ask your agent to install workbench or check prerequisites:

"Install wbcli and set up this repo for workbench"

"Check if I have all the prerequisites for workbench"

The agent will check your Python version, git, tmux, and agent CLI, then run pip install wbcli and wb setup.

configure-workbench

Ask your agent to configure agents or profiles:

"Set up Gemini as the reviewer and Claude as the implementor"

"Add a custom agent adapter for my-cli"

"Create a profile for fast iteration with codex"

The agent will use wb agents and wb profile commands to configure your setup.

use-workbench

From an idea: Ask your agent to write a plan from scratch:

"Write a workbench plan to add authentication to this app"

/use-workbench

The agent will survey your codebase, design a task graph with proper dependencies and file ownership, and produce a ready-to-run plan.

From an existing plan: You can also point the skill at an existing plan or spec to restructure it for workbench execution. This is useful when you have a rough plan, a design doc, or an implementation spec that isn't yet formatted for parallel agent dispatch:

"Convert this plan into a workbench plan: .workbench/draft.md"

"Restructure docs/spec.md as a workbench plan with proper task isolation"

The agent reads the existing document, identifies independent units of work, determines file ownership to avoid merge conflicts, sets up dependency ordering across waves, and fills in the detail each task needs to be self-contained — function signatures, interfaces from earlier tasks, test commands, and edge cases. The result is a plan at the right level of specificity for agents that can only see their own task description.

Running the plan: Once a plan is ready, the skill can also generate the wb run command with the right flags for your situation — TDD mode, specific base branches, named sessions, agent selection, directive overrides, and more:

"Run this plan with Gemini for implementation and Claude for review"

"Run plan.md in TDD mode on the feature-auth branch"

See Plan Format for details on the structure it generates.