Skills

A skill is a reusable set of instructions — how your company writes a renewal brief, what belongs in an incident retro. Workflows load a skill when relevant, instead of copying the same instructions into every workflow that needs them.

How a skill loads

Skills load in stages:

1. Metadata — name and description, always loaded (~100 tokens)
2. Instructions — the body, loaded when the skill is triggered
3. Resources — extra files (templates, examples), loaded if needed

The description is the only part Alma sees when deciding whether to use a skill. “Renewal brief” is a weak description. “How to write the pre-call brief for an account renewal — structure, tone, and what to check” gives Alma something to match against a task.

Creating a skill

Click New skill and fill in:

Field	Notes
Name	The slug derives from it — lowercase, hyphenated
Description	Max 500 characters. This is what triggers the skill.
Instructions	The body — the actual how-to

Write instructions the way you’d brief a new colleague: what good looks like, what to check, what to avoid.

What the Console can’t set yet

Caution

Coming soon. A skill supports more than the Console can author:

• Resources — extra template and example files
• Allowed tools — a whitelist restricting what the skill may call
• Model — pinning the skill to Sonnet, Opus, or Haiku
• Context mode — running the skill in isolated context

These display on the detail page if a skill has them, but there’s no way to set them from the Console — and skill detail is read-only, so you can’t edit a skill after creating it either. Use chat or the API instead.

Where skills get used

A workflow step loads a skill when the task calls for it. There’s no explicit wiring — the description is the matching surface, which is why it’s worth writing carefully.