AI-Assisted Authoring

AI agents can author .twin.yaml models and .miravo.yaml templates from a natural-language prompt. The model-author skill teaches any compatible agent the full modeling workflow.

Install

Agent Skills CLI:

npx skills add amine-amaach/miravo/skills/model-author

Claude Code:

ln -s /path/to/miravo/skills/model-author ~/.claude/skills/model-author

Codex:

ln -s /path/to/miravo/skills/model-author ~/.codex/skills/model-author

No dependencies. Works with any agent that supports Agent Skills.

Usage

Create a twin model for a variable frequency drive controlling an industrial fan

Model a centrifugal pump with hydraulic response and cavitation faults

Write a plant template for a water treatment facility with 3 pumping stages

The agent follows an 8-step workflow: research the physical asset, define parameters, order members by dependency, compose generator chains, design lifecycle and faults, then self-validate against a 25-point checklist.

Phenomenon-to-Generator Mapping

The skill maps real-world physics to the correct generator type:

Physical Phenomenon	Generator	Example
Setpoint / schedule	timetable or step	Daily pump speed profile
Motor acceleration	slew-rate-limiter	200 rpm/s up, 320 rpm/s down
Thermal inertia	first-order-lag	Bearing temp follows speed (tau 120s)
Process coupling	correlated	Fuel rate = 0.19 * load + 8
Stochastic load	mean-reverting-noise	Load factor drifts around 0.72 pu
On/off with deadband	hysteresis-relay	Compressor load/unload
Multi-signal combination	weighted-sum	Current = speed*k1 + load*k2
Accumulation	integrator	kWh totalizer, tank level
Wear-dependent	health-dependent	Vibration increases with lifecycle

Example Output

A speed-control chain the agent produces for a pump:

speed_setpoint:
  kind: variable
  data_type: Double
  access: readwrite
  unit: rpm
  eu_range: [0, 3600]
  generator:
    type: timetable
    repeat: true
    interpolation: step
    points:
      - { time: 0, value: 0 }
      - { time: 21600, value: 1200 }
      - { time: 64800, value: 900 }
      - { time: 79200, value: 0 }

speed_reference:
  kind: variable
  data_type: Double
  access: read
  unit: rpm
  generator:
    type: gate
    enabled_member: run_command
    input: { member: speed_setpoint }
    disabled_value: 0

speed_feedback:
  kind: variable
  data_type: Double
  access: read
  unit: rpm
  eu_range: [0, 1850]
  generator:
    type: slew-rate-limiter
    input: { member: speed_reference }
    rate_up: 220
    rate_down: 320
    min: 0
    max: 1850

Motors do not reach full speed instantly (slew-rate-limiter). Process signals are zero when the asset is off (gate).

Validation Checklist

The agent checks every model against these rules:

• schema_version: "2" present, members reference only previously declared members
• No instant temperatures — must use first-order-lag (tau 60-600s)
• No instant speed changes — must use slew-rate-limiter
• Lifecycle effects observable above noise (multiplier >= 1.3)
• gain * rated_input + bias produces expected output
• factor * source_range stays within eu_range

Limitations

Generators use linear math. Non-linear physics (pump power cube law, valve flow square root) are approximated with linear factors tuned for the primary operating range. The agent documents these simplifications with YAML comments.