SKILL.md
The reusable step format, spelled in capitals because that is the filename: what a Skill declares about itself, and the two sections in its body.
On this page
A Skill is a reusable step kept in your account library, and it is a folder
whose one required file is SKILL.md, in capitals.
That spelling matters more than it should. There is no skill.md on YAMO, and
there is no such thing as a "skill file" inside a Playbook either: the steps of
a Playbook are Play files, and a Skill is the
separate, reusable thing one can be imported from.
Required
name: lowercase words joined by hyphens. It is the identifier.description: what this Skill does, in a sentence. It is what you read when choosing one from the library.scope: whether it is one of yours or one that ships with YAMO.
Declared, with defaults
tools: the tools it needs. Empty by default.inputsandoutput: the shape of what it takes and what it returns, described as a schema rather than as prose. Empty by default.outputFormat: markdown by default.version: an integer, 1 by default.
Optional
model: pin a model for this Skill.search: declare that it searches the web as it works.
The body
Two sections, and only the second is required.
## Persona: who the work reads as coming from, for this step. Omitted if there is nothing to say.## Instructions: what the step does. This one cannot be empty.
Importing one
Adding a Skill to your library is an upload of its SKILL.md. A Skill that
declares a tool you have not connected is refused rather than accepted and left
broken, and its scope is recorded as yours whatever the file claims.
Importing a Skill into a Playbook creates a step from it, and that step records which Skill it came from. See The Skill library.
More in Reference
Related
playbook.md
The Playbook document itself: the fields it must declare, the ones it may, what each optional field means by its absence, and the sections in its body.Play files
The contract for one step: what every Play file has to declare, what it may add, and the rule about what a step reads when it does not say so.The agent endpoint
The one programmatic door into YAMO: what an agent can call, how a Service becomes a callable tool, and what consent and spend caps mean for the payer.Limits
The boundaries a run actually meets: how long it may take, how often a step is retried, what gets truncated, and how large an uploaded file may be.Pricing
How YAMO charges: you pay for the runs you make, and nothing for the ones you do not. What a run is, what it includes, and what publishing costs.Who it's for
The six kinds of expert YAMO is built for, and what changes when the method you already follow becomes a Service that somebody else can run.
Open a Playbook that does this
Every template in the library publishes its contract and a real example output.