Flova LogoFlova LogoDocuments

Skill System

Internal Structure of a Skill

The Core Concepts section introduced the basic role of Skills: each sub-agent's business knowledge is managed inside one shared Skill document, and edits to a given section affect the corresponding agent precisely. This section expands on the structure of a Skill and how it is used.

Section Structure

A Skill contains the following sections, each corresponding to one sub-agent's work guide:

  • planner

    Agent: Planner

    Contents: workflow stages, task dependencies, and pause-for-confirmation points

  • multimodal_analyze_tool

    Agent: analysis tool

    Contents: rules for multimodal analysis, including what to extract from uploaded materials

  • storyboard_designer

    Agent: storyboard designer

    Contents: principles for narrative structure, key-element design, shot-description language, and audio-layer design

  • media_generator

    Agent: media generator

    Contents: model choices and reference-usage strategies for keyframes, video, audio, and music generation

  • write_the_prompt

    Agent: media generator, prompt stage

    Contents: prompt-writing rules for image and video generation, cinematic language, and style description

  • video_assembler

    Agent: video assembler

    Contents: timeline assembly order, sync anchor priority, transition style, and export settings

All sections are optional. You can define only the parts you care about and leave the rest to default behavior. That keeps Skill creation lightweight.

Building Your Own Skills

You can create your own Skills in two main ways:

Create in the Skill Module

You can create a Skill directly inside the Skill module. That can mean building a new methodology from scratch, or taking an existing Skill, whether one you created earlier or one shared by another creator, and adapting it into your own version. This is the lowest-friction path: you can simply tell Flova something like "default to warm tones when generating visuals" or "I prefer faster shot transitions and less slow motion," and Flova will write those preferences into the Skill document.

Distill from a Project

As you work on real projects, you and Flova gradually settle into a workflow that fits the project. Those adjustments accumulate inside the current project Skill. You can then save that battle-tested Skill as "My Skill" and reuse it later as your own methodology.

Skill sections are maintained by Flova, not written manually by the user. You describe preferences in conversation, and Flova translates them into structured rules inside the Skill document.

Things to Watch When Creating Skills

  • Keep the description short: the description field remains in the Planner's context at all times. The more concise it is, the more focused Flova can stay.

  • The number of active Skills is limited: you can enable multiple Skills in the external Skill area, but only one Skill's detailed sections are fully loaded into a project at any given time. The names and descriptions of other enabled Skills still stay in the Planner context so it can decide whether switching makes sense.

How Flova Uses Skills

Skill loading follows a two-level mechanism. This keeps the Planner aware of high-level direction without overwhelming every sub-agent with the entire document.

Level 1: Always Loaded

The names and descriptions of all enabled Skills remain in the Planner's context at all times. The Planner knows which Skill the current project is using and which alternative Skills are available, and can use that information when deciding what to do next.

Level 2: Injected on Demand

Only the detailed sections of the one Skill that is actually loaded into the current project are injected into sub-agent contexts, and only when the relevant agent is called.

  • when the Planner starts, it reads planner

  • when the storyboard designer starts, it reads storyboard_designer

  • when the media generator starts, it reads media_generator and write_the_prompt

  • when the analysis tool runs, it reads multimodal_analyze_tool

  • when the video assembler starts, it reads video_assembler

This design gives each agent the guidance it needs at the right moment without letting different business rules interfere with one another.

Skill Writing Guide

Quick Start has been split out into a standalone file: Quick Start

Last updated on

On this page