How to Share AI Agent Skills Across Your Team

TL;DR

Store shared AI skills in one GitHub repository, then install that repository as a plugin in Claude Code and Codex. This gives the team a central source of truth, automatic updates, version control, and consistent outputs, while preserving each repeatable process as a company-owned asset instead of an isolated file on one employee’s computer.

Transcript

The people that understand how to use skills, how to use plugins with codecs and claude code just have this unfair advantage. But there's so many questions around it, like how do I use it with my team or how do I post it on GitHub or what are the skills I need to know? So on today's episode, we break it down super clearly. I brought my friend Remy ... Read More

Key Insights

• An agent skill is a markdown SOP that teaches an AI the exact steps and preferences required for a repeatable process. It replaces repeated prompting with reusable instructions that the agent can consult whenever it recognizes a relevant task.
• A well-designed skill can turn a previously iterative task into a one-shot workflow. Remy demonstrates this with proposal generation, where branding, layout, pricing placement, and other preferences are already documented instead of being explained again in every new session.
• An agent uses skills like a capable employee uses a bookshelf of SOPs. The skill name and description help the agent identify the right resource, while the full document provides the detailed instructions needed to complete the selected task.
• Team skill sharing is difficult when every employee keeps separate files on a personal computer. Valuable processes remain isolated, updates do not propagate, and ongoing improvements can produce many conflicting versions without a central source of truth.
• File sharing through Slack or email creates duplicate skills rather than synchronized resources. When one person updates a copy, other copies remain unchanged, so the team gradually loses consistency even when everyone began with the same instructions.
• Cloud storage alone does not place skills in the special folder Claude reads. Google Drive, Dropbox, and Obsidian can support shared storage, but they may require symlinks, folder configuration, installations, and technical troubleshooting that become difficult across a diverse team.
• A GitHub repository installed as a plugin provides one source of truth for team skills. According to the description, this arrangement supports automatic updates, version control, rollback, consistent outputs, and use across both Claude Code and Codex.
• Skills can encode company standards such as Notion formatting, brand voice, proposal design, and email formatting. Sharing those skills helps different team members produce outputs with the same whitespace, highlights, buttons, logos, signatures, language, and structural preferences.

Questions & Answers

Q: What is an AI agent skill?

An AI agent skill is a markdown file that acts as a standard operating procedure for the agent. It records the steps, preferences, formatting rules, and context needed to perform a repeatable task. When the task arises, the agent can read the relevant skill instead of requiring the user to explain the same requirements again in a new session.

Q: How do AI skills improve the quality of repeatable work?

AI skills preserve the specific choices that produced an acceptable result, including workflow steps, colors, logos, layout, pricing placement, brand voice, and formatting. Once those choices are documented clearly, the agent can reuse them during future tasks. Remy says this approach lets him produce a branded proposal in one shot without repeating all of his preferences.

Q: How can a team share AI agent skills effectively?

The system described keeps team skills in one GitHub repository and installs that repository as a plugin in Claude Code and Codex. The repository serves as the central source of truth, so employees use the same maintained instructions rather than independent copies. The description says this provides automatic updates, version control, consistent output, and company ownership of the work.

Q: Why is sending skill files through Slack or email ineffective?

Sending a skill as a file or zip gives each recipient a separate copy. If the original creator later improves the skill, the distributed copies do not update automatically. Changes made by recipients also do not return to the creator. Across a team, this can produce many conflicting versions and eliminate any reliable central source of truth.

Q: Why are Google Drive, Dropbox, and Obsidian awkward for sharing skills?

Claude does not automatically read skills from an ordinary shared folder in Google Drive, Dropbox, or Obsidian. The files must be connected to Claude’s designated skills folder, which can require symlinks and other folder setup. Remy says these workarounds may be manageable for a small technical team but often break when less technical employees must configure them.

Q: What kinds of company processes can be packaged as AI skills?

The examples include proposal generation, Notion formatting, brand voice, and email formatting. A Notion skill can specify whitespace, divider lines, highlights, and readability. A brand voice skill can guide email and landing-page copy. An email skill can preserve elements such as highlights, blue buttons, a logo, a signature, and the preferred overall layout.

Q: Why should AI skills be treated as company assets?

Skills capture repeatable ways of working rather than merely storing isolated prompts. When a strong process exists only on one employee’s computer, the wider team cannot reliably use or improve it. A company-controlled repository preserves that process centrally, supports shared maintenance and version history, and keeps ownership with the company instead of tying the asset to one person’s local setup.

Q: How does version control help teams manage agent skills?

Version control records changes to centrally maintained skill files and supports rollback when an update causes problems. It also reduces confusion about which copy should be treated as authoritative. In the system described, GitHub provides this shared history while the repository acts as the team’s source of truth, allowing skills to improve without creating disconnected versions across employees’ computers.

Summary & Key Takeaways

• An agent skill is a markdown SOP that documents exactly how an AI should complete a repeatable task. Instead of explaining preferences during every new session, users can encode steps, formatting rules, brand details, and other requirements once. The agent then consults that skill whenever the relevant task arises.

• Sharing skill files through Slack, email, Google Drive, Dropbox, or Obsidian creates practical problems. Copies become outdated, team members make separate improvements, and Claude requires skills in its designated skills folder. Symlinks and other folder workarounds may function for small technical teams, but they become fragile when adopted more broadly.

• Remy’s team system keeps shared skills in a GitHub repository and installs the repository as a plugin in Claude Code and Codex. The repository becomes the central source of truth, providing automatic updates, version control, rollback, and company ownership. A separate repository can hold personal skills that should not be distributed company-wide.