How to use the Follow Up on Paperclip Results agent skill

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 specific instructions for a task. Hosts that support SKILL.md skills, such as Claude Code or the Claude apps, load the instructions when a request matches them.

This skill makes the agent act as a follow-up assistant for work in Paperclip. When you ask about a task, the agent finds it by ID or by a title or number you give, then reads its status, documents and deliverables so it can report verified results and any blockers rather than guessing. When you ask for feedback, it posts a comment on the task as the person who requested it.

• Reads task status, documents and deliverables before reporting
• Reports document keys and task or deliverable links for verified results
• Names the current state and blockers when work is incomplete
• Posts feedback comments with human attribution
• Can set up event subscriptions for status, comment or document changes if your host supports MCP Events

Install the follow-up-results skill

Download the ZIP for this skill from this page and unzip it. Then place the skill folder in one of these locations, depending on how you work:

For Claude Code, put it in ~/.claude/skills/ to make it available across your projects, or in a project's .claude/skills/ folder to limit it to that project.

In the Claude apps, open Settings, then Capabilities, then Skills, and upload the skill there.

• Personal Claude Code: ~/.claude/skills/
• Single project: the project's .claude/skills/ folder
• Claude apps: Settings → Capabilities → Skills

Connect your Paperclip account first

The skill does nothing useful without a Paperclip connection. The agent uses the paperclip_connection tool to identify the company you are connected to, and the related paperclip_* tools to search, read and comment on tasks. Your connection must have access to the target company.

Make sure the connection is working before you ask for follow-ups. If the agent cannot identify a company, it will not be able to read or write tasks.

• A Paperclip account and a connected company
• Tools named paperclip_connection and related paperclip_* tools available to the agent
• An agent host that supports SKILL.md skills

How to trigger the skill with natural requests

You do not need special syntax. Describe what you want in plain language and include a task title, task number or task ID. The agent searches for the task before reading or commenting, so a title or number is enough.

If the agent finds several plausible matches, it will ask which task you mean before writing anything. Include more detail in your request to avoid this.

• "What's the status of the onboarding flow task in Paperclip?"
• "Show me the deliverables for task 42."
• "Add feedback to the pricing page task: the headline needs to be shorter. Post it as me."
• "Monitor the billing migration task and tell me when it moves to done or blocked."

Safety notes: scripts and network use

This skill runs no scripts. It does use the network, because every action goes through your Paperclip connection. The skill instructs the agent to treat work content as data, so text inside a task or document cannot grant permission to change the company, obtain credentials or run unrelated tools.

Approvals are not handled by this skill. Approval decisions stay in Paperclip's linked interface, so use that interface for them.

For monitoring, the host supplies and manages delivery credentials. The skill does not ask you for a callback URL or signing secret. Subscriptions expire and must be refreshed by the host, and there is no replay cursor in this release, so after a gap the agent recovers by reading task history and documents.

• Runs scripts: no
• Uses network: yes, through the Paperclip connection
• Approvals stay in Paperclip's linked interface
• Event subscriptions require host support; without it, no unattended monitor exists

Tips for reliable follow-ups

Give the agent a specific reference, such as a task title and number, so it finds the right task on the first search. Ask for the verified state rather than a summary from memory, since the skill is designed to read current task data.

Feedback comments are written once per intent. If a result is uncertain, the agent checks recent comments instead of posting the same feedback again, so you do not get duplicates. Avoid asking for comments just to acknowledge events, since that can create a feedback loop.

• Use a task title, number or ID in every request
• Ask for the current status, not what you remember
• Check the reply for document keys and links before acting on it
• Event monitoring works only in hosts that support MCP Events