config.yaml Reference

The config.yaml file is an optional configuration placed alongside your template. It controls the interview experience without modifying the template itself. The Analyzer merges it into the manifest; the Orchestrator reads only the manifest.

Complete example

meta:
  display_name: "Consulting Agreement"
  description: "Standard consulting engagement agreement"

variables:
  client_name:
    label: "Client's Legal Name"
    question: "What is the client's full legal name?"
    default: ""
    required: true
    format_hint: "Full legal name as registered"
  payment_method:
    type: choice
    choices: [bank_transfer, cheque, crypto]
    default: "bank_transfer"
  include_ip_assignment:
    type: boolean
    question: "Does this engagement involve IP assignment?"
    default: false

groups:
  - name: "Parties"
    variables: [client_name, client_address]
  - name: "IP Assignment"
    condition: include_ip_assignment
    variables: [ip_ownership_entity, ip_assignment_date]
  - name: "Milestones"
    loop: milestones
    variables: [description, date, amount]

validation:
  - rule: "end_date > effective_date"
    message: "The end date must be after the effective date"

`meta`

Optional. Metadata about the template.

Field	Type	Required	Description
`display_name`	string	No	Human-readable name shown to the user when Claude confirms the matched template. Defaults to the directory name.
`description`	string	No	Short description of the template’s purpose. Used internally and may appear in future template directory features.

meta:
  display_name: "Consulting Agreement"
  description: "Standard consulting engagement agreement"

`variables`

Optional. A map of variable names to override settings. Only variables that need customization need to be listed here — unlisted variables use defaults inferred from the template.

Each key is a variable name that appears in the template. The value is an object with any combination of the fields below.

Variable fields

Field	Type	Required	Description
`label`	string	No	Short label for the variable, shown as a heading for the question.
`question`	string	No	The full question text Claude asks the user. Overrides the auto-generated question.
`default`	any	No	Default value pre-filled during the interview. Use `"today"` for date types to resolve to the current date.
`required`	boolean	No	Whether the variable must be provided. Default: `true`.
`format_hint`	string	No	Additional formatting guidance shown alongside the question.
`type`	string	No	Explicit variable type. See Variable types below. If omitted, the type is inferred from the variable name suffix.
`choices`	list	Conditional	List of valid values. Required when `type` is `choice`. Ignored otherwise.

Variable types

The type field accepts the following values:

Type	Description
`text`	Free text input. The default for most variables.
`date`	A calendar date. The special default `"today"` resolves to the current date.
`number`	A numeric value — currency, quantity, percentage.
`email`	An email address. Validated for email format.
`phone`	A phone number. Accepts international formats.
`boolean`	A yes/no question. Used as the gate for `{% if %}` conditional sections.
`choice`	One of a fixed set of options. Requires the `choices` field.

If type is omitted, it is inferred from the variable name suffix (e.g., *_date becomes date, *_email becomes email). See Variable Type Reference for the full inference table.

Example

variables:
  client_name:
    label: "Client's Legal Name"
    question: "What is the client's full legal name?"
    required: true
    format_hint: "Full legal name as registered"

  effective_date:
    type: date
    default: "today"

  payment_method:
    type: choice
    choices: [bank_transfer, cheque, crypto]
    default: bank_transfer

  include_ip_assignment:
    type: boolean
    default: false
    question: "Does this engagement involve IP assignment?"

`groups`

Optional. Controls interview grouping and ordering. A list of group objects that batch related variables under a heading and present them together.

Without groups, Claude presents variables one at a time in template order. With groups, related variables are collected together.

Group fields

Field	Type	Required	Description
`name`	string	Yes	Display name for the group, shown as a heading during the interview.
`variables`	list	Yes	List of variable names to collect in this group.
`condition`	string	No	Name of a boolean variable. The group is only shown when that variable is `true`. The gate variable must be collected before this group.
`loop`	string	No	Name of a loop collection. The group collects items for that `{% for %}` loop. Must match the collection name in the template (e.g., `milestones` for `{% for milestone in milestones %}`).

Basic group

groups:
  - name: "Parties"
    variables: [client_name, client_address, consultant_name, consultant_address]

  - name: "Engagement Terms"
    variables: [effective_date, end_date, governing_law]

Conditional group

A group gated on a boolean variable. Claude asks the gate question first; if the answer is no, the entire group is skipped.

groups:
  - name: "IP Assignment"
    condition: include_ip_assignment
    variables: [ip_ownership_entity, ip_assignment_date]

Loop group

A group that maps to a {% for %} loop in the template. Claude collects items one at a time, prompting for the sub-fields on each pass.

groups:
  - name: "Milestones"
    loop: milestones
    variables: [description, date, amount]

`validation`

Optional. A list of cross-field validation rules evaluated after all values are collected but before rendering. If a rule fails, Claude reports the error message and prompts the user to re-enter the offending values.

Validation fields

Field	Type	Required	Description
`rule`	string	Yes	A Python-style boolean expression using variable names. Evaluated against all collected values.
`message`	string	Yes	Error message shown to the user when the rule evaluates to `false`.

Example

validation:
  - rule: "end_date > effective_date"
    message: "The end date must be after the effective date."

  - rule: "service_fee > 0"
    message: "The service fee must be a positive number."

Rules have access to all collected variable values and use Python expression syntax.