Attio Skill Generator

A meta-skill that generates customized Attio workflow skills tailored to your workspace.

When to Use This Skill

Use this skill when you want to:

• Create a lead qualification skill for your workspace
• Generate a deal management workflow skill
• Build a customer onboarding skill
• Create custom Attio workflow skills

Available Use Cases

Use Case	Primary Object	Related Objects	Description
lead-qualification	companies	people	Qualify and score inbound leads
deal-management	deals	companies, people	Manage deals through pipeline stages
customer-onboarding	companies	people, deals	Structured onboarding workflows

Generation Process

Step 1: Gather Workspace Schema (Targeted)

You (Claude) must discover the workspace schema for the specific use-case. The Python scripts are sandboxed and cannot access APIs or MCP tools.

IMPORTANT: Gather data for the PRIMARY OBJECT only. Do not gather full schemas for all objects.

Determine what to gather based on use-case:

Use Case	Gather Attributes For	Gather Lists?
lead-qualification	companies	NO (unless requested)
deal-management	deals	NO (unless requested)
customer-onboarding	companies	NO (unless requested)

FIRST: Check for attio-workspace-schema skill

Before using MCP tools, check if the attio-workspace-schema skill is installed by looking for its resource files.

Option A: Use attio-workspace-schema skill (PREFERRED - faster, no API calls)

If attio-workspace-schema skill exists:

1. Read ONLY the primary object's resource file:
   • deal-management → read resources/deals-attributes.md
   • lead-qualification → read resources/companies-attributes.md
   • customer-onboarding → read resources/companies-attributes.md
2. Extract attributes and select/status options for that object
3. Build JSON schema structure

Option B: Query MCP tools (FALLBACK - only if no schema skill)

If attio-workspace-schema skill is NOT available:

1. Call records_discover_attributes for the PRIMARY OBJECT ONLY
   - deal-management → records_discover_attributes for "deals"
   - lead-qualification → records_discover_attributes for "companies"
   - customer-onboarding → records_discover_attributes for "companies"

2. For select/status fields on the primary object, call records_get_attribute_options

Do NOT call get-lists unless the user specifically asks for list-related functionality. Lists are organizational containers, not essential to core workflows like deal management or lead qualification.

Do NOT gather:

• Full attribute schemas for secondary/related objects (companies, people for deal-management)
• Lists (unless user explicitly requests list functionality)

Note on related objects: Deal management involves linked companies and people, but these relationships are handled through record-reference fields on the deals object. You don't need full attribute lists for related objects.

Step 2: Build Schema JSON

Structure the discovered data as JSON for the generator. This structure is CRITICAL for correct output.

⚠️ REQUIRED FIELDS for each attribute:

• api_slug - The API field name (required)
• type - Field type: text, status, select, number, date, etc. (required)
• is_required - Boolean, true if field is required (optional, defaults to false)
• is_multiselect - Boolean, true if field accepts multiple values (optional, defaults to false)
• options - Array of option objects for status/select fields (REQUIRED for status/select types!)

⚠️ OPTIONS ARE CRITICAL: For status and select type fields, you MUST include the options array with the actual option titles from the workspace. Without this, the generated skill won't show pipeline stages or dropdown values!

{
  "objects": {
    "deals": {
      "display_name": "Deals",
      "attributes": [
        {
          "api_slug": "name",
          "display_name": "Deal Name",
          "type": "text",
          "is_required": true,
          "is_multiselect": false
        },
        {
          "api_slug": "stage",
          "display_name": "Deal Stage",
          "type": "status",
          "is_required": true,
          "is_multiselect": false,
          "options": [
            { "title": "MQL" },
            { "title": "Demo Request" },
            { "title": "Discovery Call" },
            { "title": "Demo Booked" },
            { "title": "Negotiations" },
            { "title": "Won 🎉" },
            { "title": "Lost" }
          ]
        },
        {
          "api_slug": "primary_interest",
          "display_name": "Primary Interest",
          "type": "select",
          "is_multiselect": false,
          "options": [
            { "title": "GLP-1 / medical weight loss" },
            { "title": "Body contouring / aesthetics" },
            { "title": "Replacing existing device" }
          ]
        },
        {
          "api_slug": "lost_reason",
          "display_name": "Lost Reason",
          "type": "select",
          "is_multiselect": true,
          "options": [
            { "title": "Pricing/Cost" },
            { "title": "Competitor" },
            { "title": "Timing Not Right" }
          ]
        },
        {
          "api_slug": "associated_people",
          "display_name": "Associated People",
          "type": "record-reference",
          "is_multiselect": true
        },
        {
          "api_slug": "associated_company",
          "display_name": "Company",
          "type": "record-reference",
          "is_multiselect": false
        }
      ]
    }
  },
  "lists": []
}

Key points:

• lists array is empty by default. Only populate if user requests list functionality.
• For status/select fields, copy the EXACT option titles from the workspace schema
• Set is_multiselect: true for fields that accept multiple values (check the Multi column in schema)

Step 3: Run the Generator

Execute the generator script with the workspace schema.

Recommended: Use file-based input (avoids shell escaping issues with large JSON):

# Save schema to file first
echo '<JSON from Step 2>' > workspace-schema.json

# Run generator with file input
python scripts/generator.py \
  --use-case lead-qualification \
  --name acme-lead-qualification \
  --workspace-schema-file workspace-schema.json \
  --output ./generated-skills

Alternative: Inline JSON (only for small schemas):

python scripts/generator.py \
  --use-case lead-qualification \
  --name acme-lead-qualification \
  --workspace-schema '{"objects": {...}}' \
  --output ./generated-skills

Parameters:

• --use-case: One of lead-qualification, deal-management, customer-onboarding
• --name: Skill name (hyphen-case, max 64 chars)
• --workspace-schema-file: Path to JSON file with workspace data (recommended)
• --workspace-schema: JSON string with workspace data (alternative for small schemas)
• --output: Output directory (default: ./generated-skills)

Step 4: Preview Generated Skill

Show the user the generated SKILL.md content for review:

cat ./generated-skills/acme-lead-qualification/SKILL.md

Allow the user to request modifications before packaging.

Step 5: Validate and Package

Validate the generated skill:

python scripts/quick_validate.py ./generated-skills/acme-lead-qualification

Package as a .skill file:

python scripts/package_skill.py ./generated-skills/acme-lead-qualification

Step 6: Return to User

Provide the user with:

1. Preview of generated skill contents
2. Path to the .skill ZIP file
3. Instructions for importing into Claude

Scripts Reference

Script	Purpose	Input
generator.py	Generate skill from templates	JSON schema + use-case
init_skill.py	Initialize empty skill structure	Skill name
package_skill.py	Validate and create ZIP	Skill directory path
quick_validate.py	Validate SKILL.md frontmatter	Skill directory path

Example Interaction

User: "Use attio-skill-generator to create a Deal Management skill for my workspace"

Claude:

1. I'll generate a Deal Management skill. Let me check for attio-workspace-schema skill...
   • [If schema skill exists: reads resources/deals-attributes.md]
   • [If no schema skill: calls records_discover_attributes for deals only]
   • [Does NOT call get-lists - lists are not needed for deal management]
2. Building workspace schema JSON with deals attributes only...
3. Running generator:

   python scripts/generator.py --use-case deal-management --name my-deal-management --workspace-schema-file schema.json
   

4. Here's the generated skill preview: [Shows SKILL.md content with deal stages and attributes]
5. Does this look correct? I can modify it before packaging.
6. Packaging skill...

   python scripts/package_skill.py ./generated-skills/my-deal-management
   

7. Your skill is ready: ./my-deal-management.skill

Note: The skill includes record-reference fields (associated_company, contacts) that link to companies and people, but without documenting those objects' full schemas. Lists are not included unless explicitly requested.

Template Customization

Templates are in resources/templates/. You can customize:

• SKILL.template.md - Main skill metadata and structure
• workflows.template.md - Workflow step patterns
• tool-reference.template.md - MCP tool reference
• examples.template.md - Example interactions

Use-case configurations are in resources/use-cases/:

• lead-qualification.yaml
• deal-management.yaml
• customer-onboarding.yaml

Dependencies

The generator requires these Python packages:

• chevron - Handlebars-compatible templating
• pyyaml - YAML parsing

Install if needed: pip install chevron pyyaml

Validation Checklist

Before packaging, verify the generated skill:

• All object slugs exist in the user's workspace
• All list IDs are valid UUIDs from the workspace
• Attribute slugs match the workspace schema
• Select/status options use exact titles from workspace
• No hardcoded values from other workspaces
• Skill name is hyphen-case, max 64 characters
• Description is max 1024 characters

See resources/validation-checklist.md for detailed checklist.

Attribution

This skill extends anthropics/skills/skill-creator. Licensed under Apache License 2.0. See LICENSE.txt for full terms.

Modifications from original:

• Added Attio workspace schema integration
• Added use-case specific template rendering
• Added preview system for generated skills
• Adapted for MCP tool workflow