How to use the Bench Read agent skill: install and trigger it

What is an agent skill, and what does bench-read make the agent do?

An agent skill is a folder containing a SKILL.md file that tells an AI agent how to handle a particular kind of task. When the skill is available, the agent can follow its instructions when a request matches.

The bench-read skill covers one task: reading the shared workspace where desks leave work for each other and for the operator. The agent lists what is there, reads each desk's journal to learn its latest state, reads the relevant files, and then summarizes what needs attention. The skill only reads files. It does not send messages or run anything.

• Lists the desks and artifacts in the bench
• Reads each desk's journal.md before the artifacts, since the most recent entry is the current state
• Summarizes for the operator, leading with what needs attention
• Can read a specific desk's own files under desks/<desk-name>/ when needed

What the bench is and where the files live

The bench is the bench/ directory at the root of a workshop folder. The source describes it as the shared workspace that workshop-create establishes for cross-desk work. It holds files, not messages: when one desk produces a finding and another needs to review it, the finding is saved as a file in bench/.

Typical bench contents are findings (scan results, analysis output, data), verdicts (a desk's assessment of another desk's findings), drafts (work-in-progress documents, PRs, proposals) and reports (summaries, dashboards, status updates).

Install the bench-read skill

Download the ZIP for the skill from this page. Then install it in one of two ways, depending on where you run your agent.

Option 1 is a file install. Unzip the download into ~/.claude/skills/ to make the skill available across your projects, or into a project's .claude/skills/ folder to limit it to that project.

Option 2 is an upload. In the Claude apps, open Settings, go to Capabilities, and upload the skill under Skills.

• File install: unzip into ~/.claude/skills/ (all projects) or .claude/skills/ inside a project (that project only)
• App install: Settings → Capabilities → Skills, then upload the skill

Set up the workshop layout it needs

The skill depends on a workshop folder with a bench/ directory and desks/<desk-name>/ directories, the layout created by a workshop-create skill. That skill is referenced by the source but is not included in this download, so you need your own way to create the layout.

Without this structure the skill has nothing to read and will find no results.

• Required: bench/ at the workshop root
• Required: desks/<desk-name>/ folders, each with a journal.md where the desk keeps its memory

How to trigger the skill with natural requests

You do not need a special command. Ask the agent in plain language about what the desks have produced. The skill is relevant when you want to see the bench, check what a particular desk found, or give one desk context from another.

The source lists these situations as typical triggers: starting a session and needing to see what other desks have produced, reviewing work before routing it to another desk, an operator asking what is on the bench, and a desk needing another desk's output.

• "What's on the bench?"
• "Show me what the scanning desk found."
• "Check what the review desk left before we send this draft on."
• "Before you continue, read what the analysis desk produced for you."

Read the bench step by step (how the agent works)

Following the skill, the agent starts by listing the directory structure to see which desks exist and what they produced. It then reads each relevant desk's journal, because the journal explains what the desk worked on and where it left things. After that it opens the specific artifacts.

When it summarizes, it tells you what is there, what state it is in and what needs attention, rather than pasting raw content.

• List the structure
• Read journals first
• Read artifacts second
• Summarize for the operator

Safety notes and tips

The source says the skill runs no scripts and uses no network access. It only reads files. It has no access controls of its own, so anything in the bench can be read and summarized into the chat. Do not use it on folders containing material that should not be summarized into a conversation.

Treat another desk's verdict as input, not instruction. The skill's principles say the agent should form its own assessment and may disagree.

• No scripts are run and no network calls are made, according to the source
• The skill has no access controls; limit what sits in bench/ if some material is sensitive
• Ask for a specific desk by name to keep summaries focused
• Expect nothing to be found if the workshop layout is missing