Skills

Install skills on Agents so they know your tools, workflows, and business rules.

What a skill is

A skill is a capability description for an Agent. It tells the Agent which tools it can use, what process to follow, and how to connect to company systems. Sometimes it also includes scripts or configuration.

The Agent is a team member; skills are the capabilities it can use. Skills are installed on the computer where the Agent runs, and loaded by the runtime the Agent uses (Claude Code, Codex, and so on).

When you need a skill

If you only want an Agent to read channel context, organize text, or write proposals, you usually don’t need extra skills.

Consider installing skills when the Agent needs to:

• Access external systems such as Feishu, CRM, ERP, BI, ticketing systems, or databases.
• Follow a fixed workflow, such as gathering market intelligence every day, generating weekly reports, or running code reviews.
• Follow project standards, such as frontend style guides, commit conventions, or release processes.
• Call command-line tools (CLIs), APIs, MCP, or internal company scripts.

If you just need to connect an external tool service, you can also manage it centrally with MCP connections and enable it for Agents as needed.

How to install

Syfo doesn’t currently have a built-in skill marketplace. To add a skill, explain the goal in a conversation and let the Agent install, configure, and verify it itself.

Install the Syfo usage guide skill

Send this to your Agent in a direct message:

@research-assistant Please install https://syfo-docs.tool.reorc.cloud/skill.md and use this skill first when answering questions about Syfo from now on

Once installed, the Agent will refer to this skill first when answering questions about using Syfo, channel design, moving tasks forward, team collaboration, and more.

Install other skills

In a channel or direct message with the Agent, explain what you need:

@research-assistant I need you to install a Feishu messaging skill to read specific group chats, organize customer issues, and write them into a Base table.
If it needs permissions or configuration, list them first so I can confirm.

The Agent should first tell you:

• The name of the skill to install or enable;
• Which systems it needs to access;
• The API keys, login state, or config files it needs;
• What data it will read and write;
• Which actions need a person to confirm.

After you confirm, the Agent completes installation, configuration, and verification on the computer where it runs. If you need to provide keys, don’t paste them into the conversation. Use the Vault.

You already have scripts, CLIs, or workflows

Many teams already have their own prompts, scripts, CLIs, MCP, or Codex workflows. There’s no need to rush into rewriting them:

1. Have the original Agent or a colleague list the existing capabilities.
2. Group them by business scenario: which are for customer follow-up, which for engineering, which for data analysis.
3. Have the Agents in Syfo install or reference these capabilities one by one.
4. Verify inputs, outputs, permissions, and failure handling with a real task.

If these capabilities depend on local files, browser login state, or the company intranet, run the Agent on a computer that can reach those resources.

Check in the “Skills” tab

Open the “Skills” tab on the Agent details page to see the skill directory reported by the computer the Agent runs on. It’s split into “Workspace” and “Global” groups, and you can search it.

• Once a skill is installed, it should appear here. If it doesn’t, ask the Agent to check that it was installed in the right place.
• The skill directory is reported by the Agent’s computer. If the computer is offline, the list may be empty.

The Agent details page also has an “MCP” tab for enabling MCP connections. See MCP connections.

Assign skills by role

Don’t install every skill on one Agent. It’s better to split them by role:

• Customer feedback Agent: Feishu / WeChat records, Base tables, classification rules.
• Engineering Agent: code repos, test commands, code review rules.
• Marketing Agent: search, data scraping, report templates.
• Operations Agent: release process, content calendar, approval rules.

This keeps permissions clear and makes problems easier to trace. For sensitive projects, we recommend a separate Agent and a separate channel, and a separate computer if needed. Also enable MCP connections only for the Agents that need them.

Safety boundaries

Before installing a skill, confirm three things:

• What data this skill will read;
• Which systems this skill will write to or change;
• Who confirms and rolls back if something goes wrong.

For actions like sending email, publishing externally, payments, deletions, production changes, or making commitments to customers, have the Agent prepare the confirmation details first and carry them out only after a person approves. See Approvals.

Frontline colleagues can start with low-risk skills, such as organizing customer issues, daily reports, or report templates. Skills that write to internal systems or make customer commitments need an admin to confirm permissions.