How to use the Paperclip Board Manager agent skill

What the Paperclip Board Manager skill makes the agent do

An agent skill is a folder with a SKILL.md file that gives an AI agent specific instructions for a task. This skill teaches the agent to act as a board member for a company run through Paperclip. You talk to it conversationally and it turns your requests into calls to the Paperclip web API, then summarizes the results instead of showing raw data.

According to the skill's instructions, the agent can help with onboarding, company and agent management, approvals, task monitoring, cost oversight, and reviewing work product.

• Check the company dashboard: agents, open and blocked tasks, monthly budget use, and pending approvals
• Create a company and set a monthly budget
• Hire a CEO agent and guide you through its name, icon, working directory, adapter and budget
• Review and approve pending requests
• Keep a Decision Log issue so later sessions can pick up prior context

How do you install the paperclip-board skill?

Download the ZIP of the skill from this page. It contains SKILL.md. You can install it in one of two ways.

For Claude Code, unzip the skill folder into ~/.claude/skills/ to make it available everywhere, or into a project's .claude/skills/ folder to limit it to that project.

In the Claude apps, upload the ZIP under Settings → Capabilities → Skills.

• Personal install: ~/.claude/skills/
• Project install: the .claude/skills/ folder inside your project
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

What do you need before the skill works?

The skill manages an existing Paperclip server. It cannot provide one. Set up the server with the paperclipai CLI using the setup command below. That command sets the environment variables the skill reads.

The skill expects curl to be available in the agent's bash environment. It also reads two environment variables: PAPERCLIP_API_URL, the base URL of the server, and PAPERCLIP_COMPANY_ID, the active company (this can be empty if no company exists yet). If the server is not in local_trusted mode, you also need PAPERCLIP_API_KEY. The CEO agent runs on the Claude Code adapter (claude_local) when you use the default setup.

• A Paperclip server running locally or remotely
• curl available in the agent's bash environment
• PAPERCLIP_API_URL and PAPERCLIP_COMPANY_ID set
• PAPERCLIP_API_KEY if the server is not in local_trusted mode
• The Claude Code adapter (claude_local) if the CEO agent is to run locally

How do you trigger the skill with natural requests?

You do not need to name the skill or know any API details. Describe what you want in ordinary language, such as checking the dashboard, starting a company, or reviewing approvals. The skill's description covers onboarding, company or agent management, approvals, task monitoring, cost oversight and work product review.

At the start of a new conversation, the skill checks whether the environment variables are set. If PAPERCLIP_COMPANY_ID is set, it fetches the dashboard and gives you a brief status summary. If it is not set, it lists existing companies or guides you through creating one.

• Start with the goal, for example: check on my company
• Name the outcome you want, for example: hire a CEO agent with a budget of about $100 a month
• Ask for a decision record, for example: add that to the decision log

Example requests and what happens

Here are typical requests and what the agent does with them. The agent reads the state first, makes the change if you asked for one, and reports back in plain language with links to the web UI.

Before modifying a document or config, the skill re-reads it from the API so it does not overwrite newer changes. Because the company and agent settings can change, review what the agent proposes when the request has side effects.

• 'Show me the dashboard.' The agent presents agent counts, task counts, budget use this month and pending approvals.
• 'Set up a new company called Acme for a web agency with a $500 monthly budget.' The agent asks for missing details, creates the company and reports the company ID and issue prefix.
• 'Hire a CEO.' The agent asks for name, icon, working directory, adapter and budget, then submits the hire.
• 'What is waiting for my approval?' The agent lists pending approvals and what each one asks for.
• 'Help me draft a hiring plan.' The agent discusses goals and roles with you, then stores the plan as a document and writes a local copy.

Safety notes: scripts and network access

This skill runs scripts and uses the network. It sends curl requests to the URL in PAPERCLIP_API_URL, and it writes local files such as ./artifacts/decision-log.md and ./artifacts/hiring-plan.md. Its actions are real changes to the Paperclip server, so treat it as an operator with write access.

The skill's own onboarding flow auto-approves the CEO hire when company governance requires approval, because the user has just asked for the hire. Do not point this skill at a production server until you have read what it can change.

• It can create agents, approve hires and modify company settings
• It sets requireBoardApprovalForNewAgents to true on new companies so hires go through approval
• It can start an agent's heartbeat, which runs the agent
• Use a local or test server first, and review requests before approving them

Tips for getting good results

Give the agent the context it cannot see, such as the mission of the company and the budget you are comfortable with. The skill stores the company description in the company record and records decisions in the Decision Log, so later sessions can rebuild context from it.

If you are unsure what a change will do, ask the agent to explain it first. Ask for a summary rather than raw data; the skill is designed to summarize and include web UI links in its replies.

• State the mission and monthly budget up front
• Ask for a summary of pending approvals before approving anything
• Check blocked tasks on the dashboard when the summary mentions them
• Keep your Board Operations issue in the Paperclip UI so the decision log stays visible