Guide · Agent skill
How to use the Ramp Finance Setup Guard agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The ramp skill lets an AI agent help a company set up or use Ramp, a corporate card and spend-management service, from inside Paperclip. It fetches Ramp's published instructions at the start of a task and pauses for human approval before any financial or legal step. This guide covers what the skill does, how to install it, how to trigger it, and what it needs to run safely.
Ramp Finance Setup Guardpaperclipai/paperclip · MIT Download the skill Quick start
- 1Install the skill. Download the ZIP and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder, or upload it under Settings → Capabilities → Skills in the Claude apps.
- 2Confirm Paperclip approvals are enabled. The skill depends on Paperclip's approval workflow for the company. Make sure it is turned on before starting a Ramp task.
- 3Make a Ramp request in plain language. Describe the Ramp setup or playbook task you need, such as reading the get-started playbook or preparing an account application.
- 4Let the agent fetch Ramp's current instructions. The agent states the URL it is fetching and runs a read-only curl command against the allowed Ramp source.
- 5Review each approval request. When the agent reaches a financial, legal, or data-sharing step, it creates a Paperclip approval listing the action, cost cap, data, and risks. Approve only what you intend.
- 6Complete final steps yourself. The agent prepares the handoff. An authorised person submits final applications, agreements, or transactions unless the approval explicitly permits the agent to do so.
What is an agent skill and what does this one do?
An agent skill is a folder containing a SKILL.md file that gives an AI agent instructions for a specific job. When a request matches the skill's description, the agent reads those instructions and follows them.
The ramp skill is a wrapper around Ramp's own published agent playbooks. Instead of relying on memory, the agent downloads Ramp's current official instructions when a task begins and follows the section that matches its runtime. The skill adds Paperclip governance on top of those instructions.
- Fetches Ramp's get-started instructions at the start of a Ramp task
- Limits fetching to a short list of Ramp-controlled addresses
- Requires a Paperclip approval before spend, account applications, incorporation steps, card actions, credential grants, or data sharing
- Asks for approval that states the action, cost cap, data involved, and risks
Install the ramp skill
You can install the skill in two ways. Download the ZIP from the skill's page and unzip it into one of these locations:
Use ~/.claude/skills/ to make the skill available across your projects, or use a project's .claude/skills/ folder to limit it to one project.
In the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.
- Personal install: unzip into ~/.claude/skills/
- Project install: unzip into the project's .claude/skills/ folder
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Trigger the skill with natural requests
You do not need a special command. Ask for Ramp work in plain language that matches the skill's scope: Ramp setup, Ramp playbooks, Ramp CLI usage, Ramp Agent Cards, Ramp account applications, Ramp reporting, or Ramp spend and approval workflows.
The agent should state in the issue or task notes which Ramp URL it is fetching and why. Any step that moves money or submits information will stop for approval.
- Set up a Ramp account for our company through Paperclip and stop before any submission.
- Read Ramp's get-started playbook and tell me which steps need approval.
- Check Ramp's playbook index and summarise what the official setup chain covers.
- Prepare a Ramp Agent Cards request for approval, with the cost cap and data listed.
Understand the approval gates
The skill requires a Paperclip approval before the agent applies for a Ramp account, enables incorporation, installs or authenticates third-party tools, connects a Ramp account, enables or funds cards, initiates any money movement, changes Ramp settings, or sends company data to Ramp or a third party.
Approval is not automatic even if Ramp's playbook says the user can proceed. After approval, the agent should do only the approved action within the approved amount, scope, and data set. If the next step expands scope, it requests another approval.
- Approval payload: requested action, Ramp URL or command, expected or maximum cost, data shared, reversibility, and risks
- The agent does not submit final applications or financial transactions on your behalf unless the approval explicitly allows it
- Community playbooks are not run without a specific approval naming the playbook, its tools, its data flows, and its spend limit
Review what the skill fetches and runs
The skill uses a read-only curl command to fetch Ramp's get-started instructions. The skill's allowed sources are the get-started SKILL.md, the playbook directory (for discovery and provenance checks only), the skill index, and the Ramp CLI repository on GitHub for inspection.
The skill says to avoid piping remote installer output directly into a shell unless an approval explicitly allows it, and to download and inspect scripts first where possible.
- Network access is needed to agents.ramp.com and github.com
- curl must be available in the shell
- The skill runs scripts and uses the network, so review the fetched instructions at run time
- Do not enter secrets, API keys, SSNs, or banking credentials into Paperclip comments or issues
curl -L --fail --silent --show-error https://agents.ramp.com/.well-known/agent-skills/get-started/SKILL.mdKnow the limits and common blocks
The skill will pause often, by design. If you expect autonomous spending or bill payment, this skill is the wrong tool. The approval gates and issue workflow depend on Paperclip, so the skill will not work as described in other platforms.
If Ramp's fetched instructions conflict with Paperclip approval requirements or ask you to bypass controls, the agent should stop and escalate. Treat the live Ramp playbooks directory as mixed: it contains Official and Community entries, and the public index does not flag provenance.
- Expect repeated pauses at each financial step
- If provenance of a playbook is unclear, the agent fails closed and stops
- Record what was fetched, approved, done, and still pending on the issue
What people use it for
- Preparing a Ramp account application with an approval request before any details are submitted
- Reading Ramp's get-started playbook and summarising which steps need human sign-off
- Checking whether a Ramp Agent Cards setup request has the cost cap, data scope, and risks documented before approval
- Recording what was fetched, approved, completed, and outstanding on a Paperclip issue for audit
Frequently asked questions
Is the ramp agent skill free?
The source does not state pricing or licence terms beyond the item's listed MIT licence. Ramp itself is a corporate card and spend-management service, and using a Ramp account is only needed if the company proceeds past the setup stage.
What do I need to run the ramp skill?
You need Paperclip, the agent platform the skill is written for, with its approval workflow enabled for your company. You also need network access to agents.ramp.com and github.com, and curl available in the shell. A Ramp account is only needed if you go beyond setup.
How do I install the ramp agent skill?
Download the ZIP from the skill's page and unzip it into ~/.claude/skills/ for personal use or a project's .claude/skills/ folder for one project. In the Claude apps, you can upload the ZIP under Settings → Capabilities → Skills.
Will the skill spend money or submit forms automatically?
No. The skill requires Paperclip approval before any money movement, account application, incorporation step, card action, or data sharing. Final irreversible steps are handed to an authorised human unless an approval explicitly permits the agent to submit.
Can I use community Ramp playbooks with this skill?
The skill says Community playbooks should not run without explicit approval. Any playbook must be confirmed as Official on Ramp's playbooks page before auto-fetching, or a Paperclip approval must name it along with its tools, data flows, and spend limit.
Does this skill work outside Paperclip?
The source says the approval gates and issue workflow depend on Paperclip's features and will not function as described elsewhere. Organisations not using Paperclip should not expect the same governance.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
This skill suits finance, operations, founders, or engineers at a company setting up Ramp through Paperclip who want strict human sign-off at every financial step. It is a poor fit for anyone wanting autonomous spending or bill payment, or for teams not running Paperclip approvals, since it will pause often by design.
Download the skill