Morzai AI Skills
This is the master gateway for the Morzai AI suite. It identifies user needs and routes them to the appropriate specialized sub-skill.
Root Responsibility
- Route the request to exactly one child skill.
- Read
api/commands.jsonbefore asking follow-up questions. - Ask only for missing required inputs, and ask at most 2 short questions in one turn.
- After routing, open the target child skill and continue the task in the same user-facing agent.
- Do not promise unsupported behavior from the root layer.
Capability Layers
Layer 1: Single-Function Editing Tools
apparel-recolorgarment-retouchclothing-adjustment
Layer 2: Composite Listing Workflow
morzai-ecommerce-product-kit
Source Of Truth
- Routing rules, required inputs, follow-up prompts, and execution mapping are defined in api/commands.json.
- Child skill details stay in:
Flow
Step 1: Detect Intent
- Read
api/commands.json. - Choose the single best matching command.
- If still ambiguous, ask one short disambiguation question.
Step 2: Fill Missing Inputs
- Use the command's
required_inputsandoptional_inputs. - Ask only for missing required inputs first.
- If the user gave a sample image for recolor, treat it as a way to derive the target color, not as permission to skip color confirmation entirely.
Step 3: Enter Child Skill
- Open the routed child skill file.
- From this point on, follow the child skill's workflow and constraints.
- The root skill stops being the domain expert after routing and missing-input collection.
Step 4: Execute
- Use the command's
executionmapping fromapi/commands.json. - Layer 1 workflows use the routed child skill's remote MCP metadata.
morzai-ecommerce-product-kituses the publicmorzaiCLI path.- Return a clean result summary to the user.
Command Map
Layer 1
apparel-recolor-> skills/apparel-recolor/SKILL.mdgarment-retouch-> skills/garment-retouch/SKILL.mdclothing-adjustment-> skills/clothing-adjustment/SKILL.md
Layer 2
morzai-ecommerce-product-kit-> skills/ecommerce-product-kit/SKILL.md
Guardrails
- The root skill must not invent new parameters outside
api/commands.json. - The root skill must not restate all child-skill knowledge.
- The root skill must not keep working at the root layer after the child skill has been selected.