Skip to main content
Custom Skills let you encode your own domain expertise, workflow preferences, or house style into a reusable configuration that MIRA applies to any session.

​
Creating a Skill

1

Open Skills Studio

Click the Skills icon in the left activity bar (⌘2), or navigate to the Skills tab.
2

Click New Skill

Click + New Skill at the top of the Skills list.
3

Fill in the basics

FieldRequiredDescription
Name✅Short descriptive name shown in the Skill Picker
DescriptionOptionalOne-line summary displayed below the name
CategoryOptionalGroups skills in the picker (Domain Expert / Coding / Research / Custom)
IconOptionalEmoji displayed on the skill badge
ColourOptionalHex colour for the badge background
4

Write your instructions

The Instructions field is the core of the Skill. Write in plain language — this text is injected into the system prompt whenever the skill is active.Effective skill instructions typically cover:
  • Domain focus — what areas the AI should prioritise and how to reason in them
  • Preferred tools — which Python libraries, frameworks, or APIs to use
  • Output format — how results should be structured (e.g. tables, scientific format, severity ratings)
  • REPL execution guidance — when to run code vs. describe code
  • Persona boundary — how to handle queries outside the skill’s domain
You can use Markdown formatting in instructions. The AI interprets it correctly.
5

Configure model overrides (optional)

OverrideTypeDefaultDescription
temperaturefloat 0.0–1.0Provider defaultControls response randomness. Lower = more deterministic.
maxIterationsinteger30 (RLM) / N/A (NAE)Maximum reasoning iterations before the engine returns its best answer.
maxTokensinteger4096 (RLM) / 8000 (NAE)Maximum tokens the model may generate per response.
temperatureEnabledbooleantrueWhether the temperature override is active for this skill.
maxIterationsEnabledbooleantrueWhether the max-iterations override is active for this skill.
maxTokensEnabledbooleanfalseWhether the max-tokens override is active. Disabled by default — model uses its own default.
Overrides apply only when the skill is active in a session. They revert to engine defaults when the skill is detached.Toggle each override individually. Disabled overrides inherit the engine’s global setting.
6

Configure MCP tool access (optional)

The MCP Allowlist controls which connected MCP servers are available when this skill is active. Leave empty to allow all connected servers, or specify a list to restrict access.
7

Save

Click Save. The skill appears immediately in the Skills list and in the Skill Picker on new chat sessions.

​
Skill instructions template

Here is a starting template you can adapt: 
## Active Skill Mode: [Your Skill Name]

This session has the [Skill Name] skill active.

## Domain Focus
You specialise in:
- [Specific area 1]
- [Specific area 2]
- [Specific area 3]

Preferred libraries/tools: [list them].
Always [specific formatting or citation requirement].

## REPL Execution
For [specific task types], execute Python code in the REPL rather than describing
what the code would do. Use [relevant libraries] to compute and verify results.
Always print and interpret actual output.

## Behaviour
- When asked about your capabilities: explain that you are in [Skill Name] mode.
- For topics outside [domain]: note your current focus and offer to address it
  from a [domain] angle if possible, or ask the user to remove the skill.

​
Editing a Skill

Click any skill in the Skills list to open it in the editor. Make your changes and click Save. Every save creates a new version — see Version History for how to restore a prior version. Built-in skills cannot be edited directly. Duplicate a built-in skill (⋮ → Duplicate) to create an editable copy.

​
Duplicating a Skill

Right-click a skill (or click ⋮) and select Duplicate. A copy is created with the same configuration, named “[Original Name] (Copy)”. Edit the copy freely — the original is unchanged.