Skills
A skill is a short document that tells your AI client how your business does one job: the steps, the rules, and the reference material it should read. Qualifying a lead, chasing an invoice, writing a handover note. The AI loads it when the job comes up and follows it, so the work is done the same way every time.
Connectors give your AI the ability to act. Skills tell it how you want it done.
Knowledge, collections, skills
Three levels, and they nest:
- Knowledge is the section in the sidebar where all of this lives.
- A collection is a folder of skills. It is the thing you attach to a profile and the thing you share.
- A skill is one item inside a collection.
There are no loose skills. Every skill belongs to a collection, because a collection is what a profile actually carries.
How a skill reaches your AI
Attach a collection to a profile and every skill in it becomes a tool on that
profile's endpoint, named skill_<collection>_<skill>. Your AI sees the skill's
description in its tool list and loads the full text when the job comes up.
That description is the most important line in a skill. It is all your AI sees before deciding whether to open it, so it should say when to use the skill, not what is inside it. "Read this before replying to a new enquiry" works. "Our lead process" does not.
Changing a skill's text reaches every connected client on its next call, with no reconnect. Adding a new skill, archiving one, or changing a skill's description changes the tool list itself, and clients only re-read that when they reconnect. If a new skill does not show up in your AI, disconnect and reconnect the connection in your client.
Skills are read only over MCP
An attached collection gives your AI the skills and nothing else. There is no way for a connected client to create, edit or publish a skill through a collection, so attaching one is a decision with a single meaning: this profile can read these procedures.
Authoring happens in the dashboard, or through the FloConnector connector if you want your AI to help you write them.
Draft and published
A skill is a draft until you publish it, and nothing reaches an AI client until you do. Every save keeps a revision with its author and note, so you can see what changed and restore an earlier version. A restore adds a new revision equal to the old one; it never rewinds history.
Where skills come from
Three ways to get one:
- Write it in the dashboard.
- Import it as an Agent Skills zip, if you already have skills built for another client.
- Install it from the Library, a set of skills written and maintained by FloConnector, free to use and change.
See Install a skill for the Library, and what the two install options mean.
What goes in a skill
A skill is a directory, not a single file:
my-skill/
SKILL.md the instructions themselves
references/ detail loaded only when SKILL.md points at it
scripts/ code for your AI client to run
assets/ templates, schemas, lookup tables
Those three folder names are fixed, and files sit one level deep. The body is capped at 20KB, each file at 32KB, and a skill can carry 25 files.
The cap that matters is the body one. A skill is read in full every time it is
used, so keep the procedure short and move the detail into references/. A
reference file costs nothing until your AI actually follows the link to it.
Text only: Markdown, code, JSON, CSV, YAML. Images and other binaries are rejected.
"You are Molly, you triage enquiries like this" is a skill, not a separate kind of thing. So is a tone of voice guide, a pricing policy, or a checklist.
