Skills

A skill is a reusable piece of know-how for Cooper's AI agents. It holds written instructions (the steps, rules and domain knowledge for a job) plus optional files the agent can read. You write a skill once on the Skills page and then give it to as many agents as need it. When you improve the skill, every agent that has it gets the new version.

Admins and team leads use skills to capture how the company works: how to reconcile a credit card statement, how to review a subcontractor quote, how to write a daily report. Cooper keeps a version history of every skill, so you can see who changed what and go back to an earlier version.

Key conceptsCopy Anchor Link

How skills and agents work togetherCopy Anchor Link

An agent is a persona with its own instructions, such as a procurement assistant. A skill is a capability you plug into agents. One agent can have many skills, and one skill can be shared by many agents.

• You attach skills on the agent. Open the agent on the Agents page and pick skills in its Skills section. The Skills page shows the result in the Used By column, but you can't attach a skill from the Skills page itself.
• Agents get the skill's instructions. When an agent runs in Chat or on the AI Team, Cooper adds the instructions of each of its skills to the agent's own instructions.
• Agents can load a skill on demand. When an AI assistant such as Claude or ChatGPT switches to one of your agents through the CooperBuild connection, Cooper gives it a list of the agent's skills with each skill's name and description. The assistant loads a skill's Instructions and Agent Files when the description says it applies. This is why a clear Description matters.
• Media Documents are for people. Cooper does not hand Media Documents to agents automatically. Put anything an agent must read into Instructions or Agent Files.
• Connector skills appear only where they work. If a skill lists Required Connectors, agents only see it in runs that can reach every one of those systems. Everywhere else, the skill is left out completely, so the agent is never taught a procedure it has no tool for.
• Changes apply everywhere. Editing a skill updates it for every agent that has it. There is no per-agent copy.

Open SkillsCopy Anchor Link

1. In the sidebar, open Brain.
2. Click Skills.

The page opens at /connect/skills with two tabs: Skills (the list) and Updates (a feed of recent changes).

You need the Skills permission in the Brain section of your role to see the page. Without it, the sidebar item is hidden and opening the address shows an access denied page. If your company's plan does not include Skills, Cooper shows a locked screen that says it isn't in your plan. See Permissions.

Read the Skills listCopy Anchor Link

The Skills tab lists your workspace's skills and Cooper's system skills, most recently updated first. More skills load as you scroll. The footer shows how many are loaded, for example "Showing 25 Out of 40".

What you can do with the list:

• Open a skill. Click a row. Cooper opens Update Skill, or the read-only View Skill for a system skill or if you can't edit skills.
• Sort. Click Skill, Category, Version or Activity in the header and choose Sort Ascending or Sort Descending. Choose Remove Sort to go back to the default order.
• Group. Drag the Category or Used By column header onto the bar that says Drag a column header here to group rows. When you group by Used By, a skill that several agents share appears once under each agent, and skills no agent uses are grouped under —.
• Show or hide columns. Click Toggle Columns (the columns icon next to Add Skill) and tick the columns you want.
• Open an agent. Click an avatar in Used By. Cooper opens that agent in an Update Agent panel, so you can change it without leaving the page.

There is no search box on the Skills tab. To find a skill quickly, sort by Skill or group by Category.

Add a skillCopy Anchor Link

A new skill does nothing until you give it to an agent. See Give a skill to an agent.

If you close the panel with unsaved changes, Cooper asks "You have unsaved changes that will be lost if you close this panel. Are you sure?". Click Discard to close or Cancel to keep editing. An orange asterisk next to the panel title means you have unsaved changes.

Attach files to a skillCopy Anchor Link

A skill can carry two kinds of files. Pick the kind by who needs to read the file.

Add Agent FilesCopy Anchor Link

1. In the Agent Files section, click the Add text files area (or its browse link).
2. Choose one or more text files.
3. Each file appears as a card with its type, name and size, marked "stored inline".

Cooper reads the file's text into the skill when you save. It skips files it can't use and tells you why:

• "Only text-based files (.md, .json, .txt, .csv, .yaml, .xml, etc.) can be stored as documents." when you pick a file of another type.
• "The following files exceed the 8 MB limit and were skipped: …" for files over 8 MB.
• "Empty files were skipped: …" for empty files.

Add Media DocumentsCopy Anchor Link

1. In the Media Documents section, click the Upload binary files area (or its browse link).
2. Choose one or more files. Cooper shows Uploading… while it uploads them.
3. Each file appears as a card with its name and type.

Remove a fileCopy Anchor Link

Hover over the file card and click the × button. Then save the skill.

The number next to Agent Files or Media Documents shows how many files the section holds.

Require a connectorCopy Anchor Link

Use Required Connectors when the skill only works with an outside system that your workspace connects through Settings → Connectors, such as a sales database.

1. In the Required Connectors section, click each system the skill needs. Selected systems are highlighted. Click again to deselect.
2. In Instructions, refer to the system's tool as {{tool:key}}, for example {{tool:apollo}}. When an agent uses the skill, Cooper replaces the placeholder with the real tool name for your workspace.
3. Save the skill.

How it behaves:

• Agents only see the skill in a run that can reach every selected system. That means the connector exists in your workspace, is active, and the person the agent works for has connected their account to it.
• Where a required system isn't available, the skill is hidden from the agent. Voice calls have no connector tools, so connector skills never appear there.
• The list of systems to choose from comes from the connector gallery. Seeing it needs the Connectors permission in Settings. Without it, Cooper shows "No connectors to choose from.", although systems already on the skill stay visible.

Declare the API keys a skill needsCopy Anchor Link

In the API Keys section, toggle each AI provider the skill expects a key for: OpenAI, Anthropic, Gemini or xAI (Grok). Selected providers are highlighted.

This is a declaration only. It tells agents and admins which provider keys the skill relies on. It does not store or add a key. Your workspace admin sets up the actual keys.

Edit a skillCopy Anchor Link

If nothing actually changed, Cooper shows "No skill changes to save." and does not create a new version.

Every agent that has the skill uses the new version from its next run.

View a system skillCopy Anchor Link

System skills come with Cooper and are shared by every workspace. They show a System badge.

Click a system skill's row to open the read-only View Skill panel. It shows the name, category, version, Description, Instructions, API Keys, Required Connectors, Agent Files and Media Documents. Sections without content are left out. When the skill has required connectors, the panel notes "Agents only see this skill in workspaces where these are connected."

You can give system skills to your agents like any other skill. You can't edit, revert or delete them, and their checkboxes in the list are disabled. Only Cooper super admins can change system skills.

If your role can't edit skills, clicking any skill also opens View Skill.

See a skill's version historyCopy Anchor Link

Cooper saves a version every time a skill is created, edited, reverted or deleted.

1. In the Version column, click the version badge (it has a clock icon when there is history). A skill still on v1 has no history to open.
2. The Recent Versions list shows up to eight earlier versions, newest first.

Each entry is an earlier version you can go back to. It shows:

• The version number, such as v2.
• The change summary of the save that replaced that version. If none was written, Cooper lists what the version contained, for example "Saved instructions, description, 2 files", or just "Saved version".
• Who made that save. If an AI agent did it on someone's behalf, it shows as the person plus the agent, for example "Dana Lee + Procurement Agent".
• How long ago, such as "3h ago" or "just now".

Revert a skill to an earlier versionCopy Anchor Link

1. Click the skill's version badge to open Recent Versions.
2. Hover over the version you want and click the wand button (Revert to vN).
3. Cooper shows "Skill reverted to version N." and the badge moves up by one.

Reverting does not delete history. It copies the old version's content into a new version, so you can always revert the revert. If the skill already matches that version, Cooper shows "Skill already matches version N." and saves nothing.

The revert button doesn't appear on system skills unless you are a Cooper super admin. Reverting needs the Write action on Skills.

Track changes on the Updates tabCopy Anchor Link

The Updates tab is a feed of the latest 60 skill changes in your workspace: creations, edits, reverts and deletions.

Each card shows:

• What happened, for example "Updated Reconciliation Workflow to v4", "Reverted Reconciliation Workflow to v2 (now v5)" or "Archived Reconciliation Workflow at v6".
• Up to three of the fields that changed, such as Instructions • Agent Files.
• Who made the change, with an Agent-assisted pill when an AI agent was involved.
• The action and the skill name as small pills.
• How long ago, and the exact date and time.

To find an update, type in Search by skill.... The search matches the skill name, the person or agent, the change summary and the changed fields. Click the × to clear it. If nothing matches, Cooper shows "No matching skill updates".

Click a card to open Skill Update. It lists the changed fields and a Diff Summary with a Before and After preview for each field. Long text is shortened in the preview.

If no skill has been changed yet, the tab shows "No skill updates yet".

Give a skill to an agentCopy Anchor Link

You attach skills on the agent, not on the skill.

The agent's skill picker lists your workspace's skills and Cooper's system skills. See Agents for the rest of the agent form.

Delete skillsCopy Anchor Link

1. Click the row's ... menu and choose Delete. To delete several skills, tick their checkboxes and click Delete in the footer.
2. In the Are You Sure? window, click Delete.

Cooper shows "Skill(s) deleted successfully." and the skill disappears from the list. Cooper archives the skill rather than erasing it, and records the deletion as a final version on the Updates tab ("Archived … at vN").

Deleting a skill does not remove it from the agents that have it. Before you delete a skill, check its Used By column and take the skill off those agents on the Agents page.

You can't delete system skills unless you are a Cooper super admin. Select all only ticks skills you are allowed to delete.

Fields referenceCopy Anchor Link

PermissionsCopy Anchor Link

Skills are controlled by the Skills permission in the Brain section of a role. An admin sets it in Settings → Roles. Each action includes the ones below it: Delete includes Write, and Write includes Read.

Workspace owners can do everything except change system skills.

Tips and best practicesCopy Anchor Link

• Write the description as a trigger. "Use when reconciling a credit card statement against receipts" helps an agent pick the right skill far better than "Reconciliation".
• Keep one job per skill. Small, focused skills are easier to share between agents and to improve without side effects.
• Put what agents must read in Instructions or Agent Files. Agents don't read Media Documents automatically.
• Use Agent Files for structured reference data, such as a cost code list in CSV or a JSON template, instead of pasting it into the instructions.
• Always fill in Change Summary. It makes the version history and the Updates tab readable, and it tells your team why a skill changed.
• Use Required Connectors for skills built on an outside system. Agents without that connection won't waste time on a procedure they can't follow.
• Check Used By before a big change. Every agent with the skill picks up the change on its next run.
• Revert instead of re-typing. If an edit makes an agent worse, revert to the earlier version from the version badge.

TroubleshootingCopy Anchor Link

For AI agentsCopy Anchor Link

There is no dedicated MCP tool for managing skills. Agents work with skills through the generic record tools on the Skill model, and through the agent-switching tools that hand out skill lists. Reading Skill records needs Skills Read (CRM.SKILLS.READ); writing needs Skills Write (CRM.SKILLS.WRITE).

How skill loading works:

• talk_to_agent returns a manifest for the agent's skills: id, name, loadWhen (the skill's description), requiredApiKeys, and tools for connector skills. Only skills this session can use are listed.
• To load a skill, call db_find with model: "Skill", where: {"_id": "<id from manifest>"} and select: ["instructions", "documents"]. Pass where and select as real objects and arrays, not strings.
• Where loaded instructions contain {{tool:<key>}}, call the tool the manifest's tools map gives for that key. db_find returns the instructions raw, with the placeholder still in them.

Key Skill fields:

• name (required, unique per workspace among non-archived skills), description, instructions (HTML from the rich-text editor).
• category: the id of a SkillCategory.
• requiredApiKeys: any of openai, anthropic, gemini, xai.
• requiresConnectors: a list of { key, hosts }. key is lowercase letters, digits or underscores, starting with a letter, such as apollo.
• documents: Agent Files, each { fileName, fileType, content, size }. content is the text, up to 8 MB.
• attachments: ids of Media Documents.
• isSystemLevel: true for Cooper's system skills. Only super admins can create or change them.
• status: Active, Draft or Archived. The Skills page shows Active and Draft skills.
• version and revisionHistory: maintained by the app.

Agents are linked to skills through the agent's skills list (model Agent). To give an agent a skill, update the agent; see Agents.

Things to know:

• Edits made with db_create and db_update write the record directly. They don't go through the app's versioning, so they don't add a version to the history or a card to the Updates tab. When the user wants a tracked change, ask them to make it on the Skills page.
• Don't try to change a system skill (isSystemLevel: true). Copy its content into a new workspace skill instead.

Common errors:

To connect Claude or ChatGPT to Cooper, see AI agents.

RelatedCopy Anchor Link

• Agents: create agents and attach skills to them.
• AI Team: where agents run with their skills.
• Brain overview: how Agents, Skills, the Org Wiki and the Knowledge Graph fit together.
• Org Wiki: company knowledge that agents look up, as opposed to procedures they follow.
• Automations: run agents on a schedule or when something happens.
• Connect settings: manage the skill categories used on this page.
• AI agents: connect Claude or ChatGPT to Cooper through MCP.