Template Constructor

End-to-end skill for building, validating, and deploying Machina templates. Covers the full lifecycle: scaffold -> validate -> install -> analyze -> trace -> configure secrets.

Available Commands

Command	Description	Reference
`init-template`	Scaffold new template project from scratch	init-template.md
`create-template`	Generate individual YAML components (agent, workflow, prompt, mapping)	create-template.md
`validate-template`	Check YAML files against correct patterns before installation	validate-template.md
`install-template`	Import templates via MCP (local or Git)	install-template.md
`analyze-template`	Analyze template structure, dependencies, secrets	analyze-template.md
`trace-agent`	Trace agent execution chain with variable propagation	trace-agent.md
`configure-secrets`	Configure vault secrets for connectors	configure-secrets.md

Quick Workflow

code

1. init-template    → Scaffold project structure
2. create-template  → Add YAML components
3. validate-template → Check syntax before deploy
4. install-template  → Deploy via MCP
5. configure-secrets → Set up credentials
6. analyze-template  → Verify installation
7. trace-agent       → Debug execution flow

Command Dispatch

When the user triggers this skill, determine which command they need:

User Says	Command
"init template", "scaffold template", "new template project"	Read init-template.md
"create agent", "create workflow", "scaffold YAML"	Read create-template.md
"validate", "check YAML", "verify template"	Read validate-template.md
"install", "import", "deploy template"	Read install-template.md
"analyze", "what's in this template", "overview"	Read analyze-template.md
"trace", "execution chain", "variable flow", "debug agent"	Read trace-agent.md
"secrets", "credentials", "configure API key"	Read configure-secrets.md

Load the appropriate reference file based on the user's intent, then follow its instructions.

MCP Server Selection

Ask the user which environment to target. The MCP server prefix follows the pattern mcp__<environment>__:

Environment	MCP Server Prefix
Local dev	`mcp__docker-localhost__`

Additional environments (staging, production) depend on the project's MCP configuration. Check available MCP servers in the current session.

SDK Skill Registration

After installing a template, optionally register it as a skill in the SDK for discoverability:

python

create_skill(
    name="template-name",
    config={
        "title": "Template Title",
        "description": "What it does",
        "template_path": "agent-templates/template-name",
        "agents": ["template-name-executor"],
        "workflows": ["template-name-main-workflow"],
        "connectors": ["machina-ai"],
        "secrets": ["TEMP_CONTEXT_VARIABLE_SDK_OPENAI_API_KEY"]
    }
)

Template Repository Layout

Templates live in repositories with this structure:

code

<repo-root>/
├── agent-templates/    # Reusable agent patterns
│   └── <template-name>/
└── connectors/         # External service integrations
    └── <connector-name>/

When installing via MCP, the project_path must point to the template directory inside the container (e.g., /app/<repo-name>/agent-templates/<template-name>).

Key Constraints

•Connector .py files: No helper functions outside command functions. Each function must be self-contained.
•Expression syntax: Always use $.get('field') — never ${field} or $field
•Prompt files: Use prompts: array (not prompt:) with instruction: (not messages:)
•Connector YAML: Use filetype: (not type:) and filename: (not script:)
•Install order: connectors -> prompts -> mappings -> workflows -> agents