How to use the Paperclip Distill agent skill

What is an agent skill?

An agent skill is a set of instructions packaged so an AI agent can load them on demand. The skill's SKILL.md file starts with a short description that tells the agent when to use it. When a task matches that description, the agent reads the rest of the file and follows it.

What does the paperclip-distill skill make the agent do?

The skill turns Paperclip activity (issue titles, comments and document bodies) into wiki-style project pages. It asks the agent to explain what a project is, what was decided, what is at risk and where it stands now, instead of producing a dated list of issues. Pages are written as patches to existing pages where they exist, so accepted decisions are kept and superseded ones are marked rather than erased.

The skill is strict about what it reads. Attachments, work products and linked URLs are not read. Writes go only to the default wiki space.

• Writes or updates wiki/projects/<slug>/standup.md first, the executive standup.
• Writes or updates wiki/projects/<slug>/index.md, the project overview with Current Direction, Workstreams, Decisions and References.
• Optionally writes decisions.md and history.md when there is enough material.
• Keeps dated audit entries in wiki/log.md and Paperclip comments rather than in project pages.

How do you install the paperclip-distill skill?

Download the skill as a ZIP from the Paperclip agent-skills page. Then choose one of two install routes, depending on how you run Claude.

For Claude Code, unzip the archive into your personal skills folder ~/.claude/skills/, or into a project's .claude/skills/ folder if you want the skill limited to one repository. The folder should contain SKILL.md.

In the Claude apps, open Settings, go to Capabilities, then Skills, and upload the ZIP there.

• Personal (all projects): ~/.claude/skills/
• Single project: .claude/skills/ inside the project folder
• Claude apps: Settings → Capabilities → Skills → upload the ZIP

What do you need before the skill will work?

This skill only makes sense inside a Paperclip and LLM Wiki setup. The source lists three requirements: a Paperclip operation issue with spaceSlug set to default, LLM Wiki tools with a wikiId and space access, and the target space's AGENTS.md page conventions.

If the operation issue carries any other spaceSlug, the skill tells the agent to stop and report the mismatch in a comment rather than write the pages.

• A Paperclip operation issue with spaceSlug set to default
• LLM Wiki tools with a wikiId and access to the space
• The target space's AGENTS.md, which sets filename, frontmatter, link and voice conventions

How do you trigger the skill with a natural request?

You normally do not call the skill by name. The agent loads it when the task matches the description: an operation issue that is a Paperclip cursor-window, distill or backfill. The source names three situations where it applies: a routine feeding a bounded bundle of recent activity for one project or root issue, a backfill that seeds the wiki with a project's history, and a manual distill-paperclip-now request from the UI.

If you are asking the agent in plain language, name the operation and the project. Requests such as the following match the skill's scope:

• Run a cursor-window distill for the Paperclip project and update its standup.
• Backfill the wiki with the history of root issue PAP-1234 from the default space.
• Distill the recent Paperclip activity for this project into the project overview and decisions.

What the skill will not do

Several tasks belong to other skills. If the operation issue is an ingest of a raw file or a query, the agent should use wiki-ingest or wiki-query instead. The skill also blocks extracting content from attachments or work products until a separate approved extraction policy exists. If you ask for that, the agent should point you to the Phase 5 asset and work-product security gate instead of improvising.

• Raw file distillation: use wiki-ingest.
• Questions against the wiki: use wiki-query.
• Attachment or work-product content: blocked until a separate policy exists.

Safety notes: scripts and network access

The skill ships as a single file, SKILL.md. It does not run scripts and does not use the network. Its safety rules are written into the instructions: the agent must never fetch /api/assets/:id/content, and must never open a work-product url, preview URL, artifact URL or other linked destination. Read the operation's own issue descriptions, comment bodies and document bodies only.

Because the skill writes wiki pages, review the output of a first run before relying on it, especially the Decisions sections, which the agent is told to preserve and mark rather than rewrite.

• Runs scripts: no
• Uses network: no
• Reads: issue descriptions, comment bodies, document bodies
• Does not read: assets, attachments, work products, linked URLs

Tips for better results

Give the agent a bounded window. The skill is designed around a cursor window or a root issue, and it asks the agent to read every issue title, comment and document in the bundle rather than sampling. Narrow bundles produce more accurate standups.

Keep the existing project pages in place. The skill reads the current standup, index, decisions and history before writing, so it writes a patch rather than a rewrite. Deleting them forces a rewrite of accepted decisions.

• Use a bounded source window for routine runs and reserve wide windows for backfills.
• Check that the operation issue shows spaceSlug set to default before starting.
• Make sure the target space's AGENTS.md is present so page conventions are followed.