Skip to content

Agent Skills

Agent Workbench is installed as an Agent Skill. This page is for the person configuring the skill for a user or project.

Use the skill when you want an agent to manage long-running repository work with a ledger instead of relying only on chat history.

Install with GitHub CLI

Preview the skill:

gh skill preview MuNeNiCK/agent-workbench agent-workbench

Install for a user:

gh skill install MuNeNiCK/agent-workbench agent-workbench \
  --scope user \
  --agent <target-agent>

Install for a project:

gh skill install MuNeNiCK/agent-workbench agent-workbench \
  --scope project \
  --agent <target-agent>

Use gh skill install --help to choose the correct target agent value.

What agents get

The installed skill gives agents:

  • the Agent Workbench operating rules
  • focused workflow references
  • close-ready and interruption troubleshooting guidance
  • a Linux x86_64 wrapper that fetches and executes the CLI

The wrapper is part of the installed skill. Humans normally do not call it directly; they ask the agent to use $agent-workbench.

What humans should ask

After installing the skill, ask the agent to use it explicitly:

Use $agent-workbench for this project.

For a new project:

Use $agent-workbench and initialize this project.

For existing work:

Use $agent-workbench and report the current status, next action, applicable
rules, corrections, and command profiles before planning.

For completion:

Use $agent-workbench and report close readiness, validation evidence, review
state, and any blockers before closing this work.

What not to copy

Do not copy installed project-scope skill directories back into the source skill package. The source package is skills/agent-workbench/ in this repository. Installed copies are generated by gh skill install.