JTBD Generator Skill

This skill generates complete JTBD (Jobs-to-be-Done) markdown files from user-provided step descriptions. It discovers the correct API operations, maps data flow between steps, and produces validated, hybrid-format JTBD files.

How to Use

When the user wants to create a JTBD, they will provide:

1. Prerequisites: Conditions that must be met (e.g., "user logged in", "API asset exists in Exchange")
2. Steps: List of operations in natural language (e.g., "Get GAV from Exchange", "Create API in API Manager")

You will guide them through discovery, validation, and generation.

Workflow

Step 1: Parse User Input

Extract the user's intent:

• Job purpose/title
• Prerequisites
• Step descriptions (in natural language)

If anything is unclear, ask clarifying questions:

• What's the goal of this workflow?
• What should happen in each step?
• Are there specific APIs you want to use?

Step 2: Discover Operations

For each step description:

1. Search for matching operations:

   import sys
   from pathlib import Path
   
   # Add JTBD generator library to path
   sys.path.insert(0, '.claude/skills/jtbd-generator')
   from lib import api_discovery
   
   # Search for operations matching the description
   results = api_discovery.search_operations(
       "create api",  # User's description keywords
       None,  # Search all APIs, or specify URN like "urn:api:api-manager"
       repo_root=Path(".")
   )
   
   # Show top 3-5 matches
   for op in results[:5]:
       print(f"{op['score']:.2f} - {op['operationId']} ({op['api']})")
       print(f"  {op['method']} {op['path']}")
       print(f"  {op['summary']}")

2. Present options to user:

   • Show the top matches with scores
   • Include operation summary and API
   • Ask user to confirm or choose

3. Get operation details:

   # Once user confirms, get full details
   details = api_discovery.get_operation_details(
       api_urn="urn:api:api-manager",
       operation_id="createOrganizationsEnvironmentsApis",
       repo_root=Path(".")
   )
   
   print(f"Parameters: {len(details['parameters'])}")
   print(f"Method: {details['method']} {details['path']}")

4. Store selection:

   • Step number
   • Step name (user-provided)
   • API URN
   • Operation ID
   • Operation details

Step 3: Analyze Parameters & Data Flow

For each step (in order from 1 to N):

1. Analyze parameters:

   from lib import parameter_analyzer
   
   # Build inputs considering previous steps
   inputs = parameter_analyzer.build_all_inputs(
       api_urn="urn:api:api-manager",
       operation_id="createOrganizationsEnvironmentsApis",
       repo_root=Path("."),
       previous_steps=previous_steps  # List of already-processed steps
   )
   
   # Review detected sources
   for param_name, input_def in inputs.items():
       if 'from' in input_def:
           source = input_def['from']
           print(f"{param_name}: from {source.get('api', source.get('step'))}")
       elif input_def.get('userProvided'):
           print(f"{param_name}: user-provided")

2. Suggest outputs:

   from lib import response_analyzer
   
   # Suggest outputs considering next steps
   outputs = response_analyzer.analyze_response_for_operation(
       api_urn="urn:api:api-manager",
       operation_id="createOrganizationsEnvironmentsApis",
       repo_root=Path("."),
       next_steps=remaining_steps  # Steps not yet processed
   )
   
   # Show suggestions
   for output in outputs:
       print(f"{output['name']}: {output['path']}")
       if 'used_by' in output:
           print(f"  → Used by: {', '.join(output['used_by'])}")

3. Confirm with user:

   • Show detected data flow
   • Highlight any user-provided parameters
   • Ask if adjustments are needed

Step 4: Generate JTBD Structure

1. Create kebab-case name:

   from lib.utils import kebab_case
   
   name = kebab_case("Deploy API with Omni Gateway")
   # Result: "deploy-api-with-flex-gateway"

2. Build YAML blocks for each step:

   from lib import jtbd_builder
   
   yaml_block = jtbd_builder.build_step_yaml(
       api_urn="urn:api:api-manager",
       operation_id="createOrganizationsEnvironmentsApis",
       inputs=inputs,
       outputs=outputs
   )

3. Generate prose sections:

   Use your intelligence to write:

   • Overview: Action-oriented description (start with verb: "Deploys...", "Creates...", "Configures...")
   • What you'll build: Clear outcome statement
   • Step descriptions: Explain what each step does and why
   • What you'll need: Prerequisites for each step
   • What happens next: Outcomes and connections to next steps

   Quality guidelines:

   • Be conversational and helpful
   • Explain the "why", not just the "what"
   • Connect steps logically
   • Anticipate common questions

4. Build complete markdown:

   from lib import jtbd_builder
   
   # Prepare step definitions
   steps = []
   for i, step_info in enumerate(step_definitions, 1):
       step_md = jtbd_builder.build_step_markdown(
           step_number=i,
           step_name=step_info['name'],
           step_description="[Your prose explanation]",
           operation_summary=step_info['operation_summary'],
           yaml_block=step_info['yaml_block'],
           what_you_need=step_info.get('what_you_need'),
           what_happens_next="[Your explanation of outcomes]"
       )
       steps.append({'markdown': step_md})
   
   # Assemble complete JTBD
   jtbd_content = jtbd_builder.build_complete_jtbd(
       name="deploy-api-with-flex-gateway",
       description="Deploy API instance to Omni Gateway. Use when deploying APIs to Omni Gateway, setting up API instances, or connecting Exchange assets to gateways.",
       title="Deploy API with Omni Gateway",
       overview="Deploys an API instance to a Omni Gateway by retrieving asset details from Exchange, discovering available gateway targets, and creating the API instance in API Manager with the proper configuration.",
       what_youll_build="An API instance connected to your Omni Gateway target",
       prerequisites=[
           "**Authentication** - Valid Bearer token for Anypoint Platform with API Manager and Exchange permissions",
           "**Resources** - API asset published in Exchange and Omni Gateway deployed"
       ],
       steps=steps,
       completion_items=[
           "Asset retrieved from Exchange",
           "Gateway target identified",
           "API instance created in API Manager"
       ],
       what_youve_built="✅ **API Deployment** - Connected Exchange asset to API Manager, configured Omni Gateway as target, API ready for policy configuration",
       next_steps=[
           "**Apply Policies** - Add security policies (OAuth2, IP allowlist) and configure rate limiting",
           "**Test API** - Verify endpoint is accessible and test through gateway"
       ]
   )

Step 5: Validate

Run the validator to ensure correctness:

python3 .claude/skills/jtbd-generator/scripts/validate_jtbd.py /tmp/generated-job.md .

If validation fails:

• Explain the error in plain English
• Auto-fix when possible:
  • Add missing outputs (common IDs)
  • Correct step references
  • Generate basic descriptions
• Regenerate and validate again

If validation succeeds:

• Show success message
• Proceed to save

Step 6: Save & Confirm

1. Determine save location:

   • Default: skills/{name}/SKILL.md
   • Ask user if unsure

2. Save the file:

   from pathlib import Path
   
   output_dir = Path("skills/deploy-api-with-flex-gateway")
   output_dir.mkdir(parents=True, exist_ok=True)
   output_path = output_dir / "SKILL.md"
   output_path.write_text(jtbd_content)

3. Confirm with user:

   • Show file path
   • Mention validation status
   • Suggest next actions

4. Suggest related workflows:

   • Based on the APIs used
   • Based on similar existing JTBDs
   • Based on common next steps

Example Interaction

User: Create a JTBD with these prerequisites: user logged in, API asset in Exchange.
Steps: 1) Get GAV from Exchange, 2) Get omni gateway targets, 3) Create API in api-manager.

You: I'll create a workflow for deploying an API with a Omni Gateway target.

Let me find the right operations for each step...

[Calls search_operations for each step]

I found these operations:

Step 1: Get GAV from Exchange
  → getAssetsByGroupidByAssetid (urn:api:exchange-experience)
     GET /assets/{groupId}/{assetId}
     Score: 0.85

Step 2: Get omni gateway targets
  → getOrganizationsByOrganizationidEnvironmentsByEnvironmentidGateways (urn:api:flex-gateway-manager)
     GET /organizations/{organizationId}/environments/{environmentId}/gateways
     Score: 0.78

Step 3: Create API in api-manager
  → createOrganizationsEnvironmentsApis (urn:api:api-manager)
     POST /organizations/{organizationId}/environments/{environmentId}/apis
     Score: 0.92

Does this match your intent?

User: Yes

You: Building the workflow...

[Analyzes parameters and data flow]

Data flow detected:
✓ Step 1 outputs: groupId, assetId, version
✓ Step 2 outputs: gatewayId
✓ Step 3 uses: groupId, assetId, version (from Step 1), gatewayId (from Step 2)

[Generates complete JTBD with prose]

Generated JTBD:
- Name: deploy-api-with-flex-gateway
- Steps: 3
- Prerequisites: User authentication, Exchange asset, Omni Gateway

[Validates]

Validation: ✅ PASSED

Save to skills/deploy-api-with-flex-gateway/SKILL.md?

User: Yes

You: ✅ Saved successfully!

File: skills/deploy-api-with-flex-gateway/SKILL.md

Next steps:
- Review the generated workflow
- Test with: python3 .claude/skills/jtbd-generator/scripts/validate_jtbd.py skills/deploy-api-with-flex-gateway/SKILL.md .
- Related workflows: deploy-api-with-rate-limiting, setup-multi-upstream-routing

Data Flow Intelligence

Apply these rules when detecting parameter sources:

Rule 1: Output → Input Matching (High Confidence)

If Step N outputs 'environmentApiId' AND Step N+1 needs 'environmentApiId':
  → Auto-link with { from: { step: "Step N", output: "environmentApiId" } }

Rule 2: Input Reuse (High Confidence)

If Step N uses 'organizationId' AND Step N+1 needs 'organizationId':
  → Reuse with { from: { step: "Step N", input: "organizationId" } }
  (Not from original source - reuse the previous step's input)

Rule 3: x-origin Suggestions (High Confidence)

If parameter has x-origin annotation:
  → Use that as the default source
  → Add alternatives if provided

Rule 4: Common Patterns (Medium Confidence)

If parameter is 'organizationId' and no previous source:
  → Default to access-management#getOrganizations

If parameter is 'environmentId' and no previous source:
  → Default to access-management#listEnvironments

Rule 5: User-Provided Fallback (Low Confidence)

If no source detected:
  → Mark as userProvided: true
  → Add example if parameter has example in schema
  → Mention in conversation that this needs user input

Error Handling

Operation Not Found

"I couldn't find an operation matching '{description}' in {api_urn}.

Let me search more broadly..."

[Search all APIs, show top matches]

"Here are some operations that might work:
1. {operationId} ({api}) - {summary}
2. {operationId} ({api}) - {summary}

Which one fits best, or would you like to search differently?"

Validation Fails

"The generated JTBD has validation errors:

❌ Step 2 references unknown output 'apiId' from Step 1

I'll fix this by:
- Checking Step 1's actual outputs
- Updating Step 2's input to use the correct output name

Regenerating..."

Ambiguous Step Description

"Step 2: 'Get gateway targets' is ambiguous.

I found operations in two APIs:
1. Omni Gateway Manager - List omni gateways
2. Runtime Manager - List CloudHub workers

Which one should I use?"

Tips for Quality Output

1. Write conversational prose:

   • "Start by retrieving..." not "This step retrieves..."
   • "You'll need the organization ID..." not "organizationId is required"

2. Explain the why:

   • Not just "Create an API instance"
   • But "Create an API instance to register your Exchange asset with API Manager, enabling policy application and gateway routing"

3. Connect steps logically:

   • End each step with "What happens next" that leads into the next step
   • Show how outputs become inputs

4. Anticipate issues:

   • Add "Common issues" for known problems
   • Mention typical error responses

5. Be specific:

   • "Organization Business Group GUID" not just "organization ID"
   • "Target environment ID (e.g., Production, Sandbox)" not just "environment ID"

Available Python Utilities

You have access to these utilities in .claude/skills/jtbd-generator/lib/:

api_discovery.py

• list_available_apis(repo_root) - List all APIs with metadata
• search_operations(query, api_urn, repo_root) - Fuzzy search for operations
• get_operation_details(api_urn, operation_id, repo_root) - Get full operation details
• get_operations_by_api(api_urn, repo_root) - List all operations in an API

parameter_analyzer.py

• analyze_parameters(spec, operation_id, spec_path) - Extract all parameters with metadata
• detect_parameter_source(param_name, param_def, previous_steps) - Detect where parameter should come from
• build_input_definition(param_name, param_def, source) - Build JTBD input definition
• build_all_inputs(api_urn, operation_id, repo_root, previous_steps) - Build complete inputs section

response_analyzer.py

• suggest_outputs(spec, operation_id, spec_path, next_steps) - Suggest which fields to capture
• generate_jsonpath(field_path) - Generate JSONPath expressions
• analyze_response_for_operation(api_urn, operation_id, repo_root, next_steps) - Main entry point

jtbd_builder.py

• build_frontmatter(name, description) - Generate YAML frontmatter
• build_step_yaml(api_urn, operation_id, inputs, outputs) - Generate YAML block for step
• build_step_markdown(step_number, step_name, ...) - Generate complete step section
• build_complete_jtbd(name, description, title, ...) - Assemble complete JTBD

utils.py

• load_openapi_spec(api_path) - Load and parse OpenAPI spec
• urn_to_path(urn, repo_root) - Convert URN to filesystem path
• path_to_urn(api_path) - Convert filesystem path to URN
• kebab_case(text) - Convert text to kebab-case
• resolve_ref(ref, spec, spec_path) - Resolve $ref references
• find_api_dirs(repo_root) - Find all API directories

common_patterns.py

• match_common_pattern(param_name) - Check for known patterns
• is_likely_user_provided(param_name) - Check if likely user-provided
• get_example_for_param(param_name, param_schema) - Get example values

Success Criteria

A successful JTBD generation:

1. ✅ Passes validation (validate_jtbd.py)
2. ✅ All operations exist in referenced API specs
3. ✅ Data flow correctly maps outputs to inputs
4. ✅ Prose sections are clear and helpful (not template-like)
5. ✅ User completed generation in <5 minutes of interaction
6. ✅ User understands what the workflow does and how to use it

Notes

• Always run the workflow in order: Parse → Discover → Analyze → Generate → Validate → Save
• Use the Python utilities for spec parsing and structure building
• Use your intelligence for prose generation and user interaction
• Be conversational and helpful
• When in doubt, ask the user for clarification
• Default to skills/{name}/SKILL.md for output location