Skills and Policies

Skills

A skill is a tool the agent can use. The Structure → Skills page is a company-wide library: a skill is defined once and assigned to any agents.

Each skill has a name, a slug, a short description, a type, Markdown content (an instruction telling the agent how and when to use it), a type-specific setting and an active/inactive status.

Skill types

Type	What it does	Setting
Built-in (builtin)	The platform's ready-made tools	The list below
MCP	Calls a tool of an MCP server (JSON-RPC over HTTP; the server must offer the …/call address)	Server address; authentication: none, API key or Bearer token
HTTP API	Calls a REST webhook or API (POST by default, GET selectable)	Address, method, fixed headers
Function	Custom Python code; needs the signature run(args: dict) -> str. Runs in a restricted environment (few built-in functions) with a 5-second timeout	Code
Database	Runs an SQL query on a connected database	Connection and schema (Settings → Databases)

For technical schemas: Skill Types.

Built-in tools

Tool	Function
Web search (web_search)	Web search with DuckDuckGo
Chart generation (text_to_chart)	A chart from JSON/CSV data. Currently a placeholder: the tool is called but does not produce a real chart
Write journal (journal_write)	Appends a Markdown entry to the session journal
Agent delegation (delegate_to_agent)	Hands the task to another agent → Delegation
Instagram post (instagram_post)	Posts to an Instagram Business account → Settings → Social media
WhatsApp message (whatsapp_send)	Sends a message through the WhatsApp Business API

Assigning to an agent

When creating or editing an agent, choose from the library in the Skills section or add a tool specific to that agent. Skill edits go to the approval process (department head + executive); until approved, the old version is in force. → Change Requests

Policies

A policy is a written rule agents must follow; it is written as a Markdown document. It has three scopes:

Scope	Applies to
Company	All agents in the company
Department	Agents in that department (automatically when linked to the department)
Agent	Only that agent (an extra policy specific to the agent)

Policies are created, edited and made active/inactive from the Structure → Policies page. Edits go to approval like skills.

How does a policy affect the agent?

• The names of the policies the agent must follow are added to its instruction.
• If the policy text contains a code block marked policy, yaml or json, the policy engine reads it as an executable rule and evaluates the agent's tool calls as allow / ask / deny. By default the engine works in trial mode (dry_run: the decision is recorded but not enforced). The mode (off, dry_run, enforce) can currently only be changed through the API: PUT /policies/config. Details: Policy Engine.

So a plain-text policy is a guide steering the agent; for hard enforcement you need to write a rule block and put the engine in enforce mode (there is no setting for this in the interface yet).