Font Size:
Ask Joget AI

Skills & Central Instructions

Skills & Central Instructions are reusable instruction sets and platform-wide system prompt rules that apply across every agent and AI composer in your Joget app.

Both features live in Settings > AI Central Config > Skills and require the Admin or App Creator role.

Central Instructions

Central Instructions are always-on rules that are automatically prepended to the system prompt of every agent and composer session. Use them to enforce house rules, tone, output formats, or domain context without editing individual agents.

Two levels

Level When it applies
Global Every agent and every AI composer in the platform, no exceptions
Per-target One specific composer (e.g., Form Builder) or the Agent Builder, applied after the Global text

At runtime, the effective system instruction is:

{Global instructions}

{Per-target instructions for this session}

If only one level is set, only that level is used. Both are optional.

Setting Central Instructions

  1. Go to Settings > AI Central Config > Skills.
  2. Expand the Global accordion item and enter rules that should apply everywhere.
  3. Expand a specific target accordion (for example, Form Builder or Agent Builder ) and enter rules that apply only in that context.
  4. Click Save. All fields are saved together.
    What to put here
    Good candidates: tone guidelines ("always respond formally"), output rules ("never include raw JSON in user-facing answers"), domain context ("we use ISO 8601 dates throughout"). Keep instructions concise the text is injected into every prompt.

Skills

A Skill is a named, reusable instruction set. Unlike Central Instructions (which are always injected), Skills are fetched on demand: the AI sees only the skill name and a short description upfront, and retrieves the full instructions only when it decides the skill is relevant.

This two-step fetch keeps token usage low for sessions where the skill is not needed.

How an agent uses a Skill at runtime

  1. The AI sees a compact list of available skills (name + description only).
  2. When the AI decides a skill is relevant, it calls the Invoke Skill tool with the skill name.
  3. The tool returns a section index (the list of ## headings) with one-line hints.
  4. The AI then fetches only the specific section it needs.
    This means a large reference skill with many sections costs almost nothing unless the AI actually needs it.

Skill fields

Field What to enter
Skill Name A short, URL-safe slug unique across the platform (e.g., weekly-status-report)
Description One to two sentences explaining when the AI should use this skill. Be specific this is all the AI sees before deciding to invoke.
Instructions Full markdown text. Use ## headings to divide content into separately fetchable sections.
Available For Checkboxes: one per registered composer plus Agent Builder. Controls which contexts can discover this skill.
Associated Tool Templates Optional. Pre-configured tools that are automatically added to the agent's live tool set whenever this skill is invoked.
Status active, draft, or deprecated (see lifecycle below).

Writing Instructions with sections

Structure your skill instructions using ## headings. Each heading becomes a separately fetchable section:

# joget-form-builder
Brief description of this skill.

## Field Types
Content about field types...

## Validators
Content about validators...

## Binders
Content about binders...

The AI calls the skill once to get the index (a list of sections with hints), then calls it again for the specific section it needs. It never loads sections it does not need.

The Nav strip

The skill editor includes a live Nav preview strip below the instructions textarea. It updates automatically as you type and shows the section list the AI will receive on the first call, useful for catching missing or misnamed headings before saving.

Clicking the Nav button (↻) regenerates the nav line and writes it directly into the instructions text. This nav line is part of what the AI reads on the first call. You can add short notes after any section name to give the AI extra orientation context:

**Nav:** Field Types · Validators (run client-side unless marked server) · Binders

Skill lifecycle

Status Discoverable by AI Executes when called
active Yes Yes
draft No No returns empty even if called by name
deprecated No Yes still runs if an agent already knows the name

Use draft while authoring. Use deprecated to phase out a skill without immediately breaking agents that may still reference it by name.

Bundled skills

Seven skills ship with the AI Agent Builder plugin and are automatically seeded on first install:

Skill Scope What it covers
Joget Form Builder Form Builder Field types, layout containers, field properties
Joget Form Validators & Binders Form Builder Validator types and data binder configuration
Joget Datalist Builder Datalist Builder Column types, formatters, filters, and binders
Joget Process Builder Process Builder Activity types, transitions, routing, swimlanes
Joget Userview Builder UI Builder Userview structure, menu items, portlets, pages
Joget Theme Builder Theme Builder Color palette, typography, CSS variables
Joget Orchestrator Orchestrator Multi-artifact build order and routing rules

These are the authoritative references used by the AI composers. They are versioned with the plugin.

Update available badge

When a new plugin version includes updated skill content, an Update available badge appears on the relevant skill card. A Sync button lets you overwrite the saved skill with the bundled version after a confirmation prompt.

Sync is destructive
Syncing a bundled skill overwrites your saved content with the plugin's built-in version. Any customisations you have made are not merged; they are replaced. The sync is always manual and always requires confirmation.

Using Skills in an agent

In the Agent Builder

Add the Invoke Skill tool to any task on the Agent Builder canvas. Open its Properties panel:

Setting What it does
Enabled Skills Multi-select list of active Agent Builder skills. Leave empty to allow all; select specific skills to restrict which ones this agent instance can call.

Every skill call is recorded in the agent's run trace and appears as a node in the Agent Node Viewer.

In AI composers (Form Builder, Datalist Builder, etc.)

Skills scoped to a composer are available automatically; no configuration required. The composer sees only the skills marked Available For that composer. If no active skills are scoped to the running composer, the invoke skill function is omitted from the LLM request entirely (zero overhead).

Invocation telemetry

The Skills list shows usage statistics on every skill card:

Metric Meaning
Invocations Total calls across all agents and runs
Successes Calls that completed without error
Failures Calls rejected or errored (e.g., skill not in enabled set, called while in draft)
Last Invoked Timestamp of the most recent call

Stats are read from the database on each page load, so they persist across server restarts and page reloads.

Related Documentation

Created by Debanraj Ravindran Last modified by Debanraj Ravindran on Sep 20, 2026