You are generating output schema files for an Apify Actor. The output schema tells Apify Console how to display run results. You will analyze the Actor's source code, create dataset_schema.json, output_schema.json, and key_value_store_schema.json (if the Actor uses key-value store), and update actor.json.
Fetches apify-generate-output-schema from apify/agent-skills and configures it for Cursor.
3
Select Cursor when prompted
The CLI shows a list of agents. Use arrow keys and space to select Cursor:
◆ Which agents do you want to install to?
│
│ ── Universal (.agents/skills) ────────────────
│ · Cline · Codex · Goose · Windsurf
│ ●Cursor(selected)
│ · Cursor · Aider · Continue
4
Verify installation
Confirm successful installation by checking the skill directory location:
.cursor/skills/apify-generate-output-schema
Restart Cursor to activate apify-generate-output-schema. Access via /apify-generate-output-schema in your agent's command palette.
⚠
Security Notice
We perform automated surface-level scans (Gen AI Scanner, Socket, Snyk) during installation. These checks detect common vulnerabilities but do not guarantee complete security. Always review skill source code and verify the publisher's reputation before production use.
Skills execute code in your environment. Always review source, verify the publisher, and test in isolation before production.
You are generating output schema files for an Apify Actor. The output schema tells Apify Console how to display run results. You will analyze the Actor's source code, create dataset_schema.json, output_schema.json, and key_value_store_schema.json (if the Actor uses key-value store), and update actor.json.
Core Principles
Analyze code first: Read the Actor's source to understand what data it actually pushes to the dataset — never guess
Every field is nullable: APIs and websites are unpredictable — always set "nullable": true
Anonymize examples: Never use real user IDs, usernames, or personal data in examples
Verify against code: If TypeScript types exist, cross-check the schema against both the type definition AND the code that produces the values
Reuse existing patterns: Before generating schemas, check if other Actors in the same repository already have output schemas — match their structure, naming conventions, description style, and formatting
Don't reinvent the wheel: Reuse existing type definitions, interfaces, and utilities from the codebase instead of creating duplicate definitions
Phase 1: Discover Actor Structure
Goal: Locate the Actor and understand its output
Initial request: $ARGUMENTS
Actions:
Create todo list with all phases
Find the .actor/ directory containing actor.json
Read actor.json to understand the Actor's configuration
Check if dataset_schema.json, output_schema.json, and key_value_store_schema.json already exist
Search for existing schemas in the repository: Look for other .actor/ directories or schema files (e.g., **/dataset_schema.json, **/output_schema.json, **/key_value_store_schema.json) to learn the repo's conventions — match their description style, field naming, example formatting, and overall structure
Find all places where data is pushed to the dataset:
JavaScript/TypeScript: Search for Actor.pushData(, dataset.pushData(, Dataset.pushData(
Python: Search for Actor.push_data(, dataset.push_data(, Dataset.push_data(
Find all places where data is stored in the key-value store:
JavaScript/TypeScript: Search for Actor.setValue(, keyValueStore.setValue(, KeyValueStore.setValue(
Python: Search for Actor.set_value(, key_value_store.set_value(, KeyValueStore.set_value(
Find output type definitions — reuse them directly instead of recreating from scratch:
TypeScript: Look for output type interfaces/types (e.g., in src/types/, src/types/output.ts). If an interface or type already defines the output shape, derive the schema fields from it — do not create a parallel definition
Python: Look for TypedDict, dataclass, or Pydantic model definitions. Use the existing field names, types, and docstrings as the source of truth
Check for existing shared schema utilities or helper functions in the codebase that handle schema generation or validation — reuse them rather than creating new logic
If inline storages.dataset or storages.keyValueStore config exists in actor.json, note it for migration
Present findings to user: list all discovered dataset output fields, key-value store keys, their types, and where they come from.
Phase 2: Generate dataset_schema.json
Goal: Create a complete dataset schema with field definitions and display views
File structure
{"actorSpecification":1,"fields":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{// ALL output fields here — every field the Actor can produce,// not just the ones shown in the overview view},"required":[],"additionalProperties":true},"views":{"overview":{"title":"Overview","description":"Most important fields at a glance","transformation":{"fields":[// 8-12 most important field names]},"display":{"component":"table","properties":{// Display config for each overview field}}}}}
Consistency with existing schemas
If existing output schemas were found in the repository during Phase 1 (step 5), follow their conventions:
Match the description writing style (sentence case vs. lowercase, period vs. no period, etc.)
Match the field naming convention (camelCase vs. snake_case) — this must also match the actual keys produced by the Actor code
Match the example value style (e.g., date formats, URL patterns, placeholder names)
Match the view structure (number of fields in overview, display format choices)
Match the JSON formatting (indentation, property ordering, spacing) — all schemas in the same repository must use identical formatting, including standalone Actors
When the Actor code already has well-defined TypeScript interfaces or Python type classes, derive fields directly from those types rather than re-analyzing pushData/push_data calls from scratch. The type definition is the canonical source.
Hard rules (no exceptions)
Rule
Detail
All fields in properties
The fields.properties object must contain every field the Actor can output, not just the fields shown in the overview view. The views section selects a subset for display — the properties section must be the complete superset
"nullable": true
On every field — APIs are unpredictable
"additionalProperties": true
On the top-level fields object AND on every nested object within properties. This is the most commonly missed rule — it must appear at both levels
"required": []
Always empty array — on the top-level fields object AND on every nested object within properties
Anonymized examples
No real user IDs, usernames, or content
"type" required with "nullable"
AJV rejects nullable without a type on the same field
Warning — most common mistakes:
Only including fields that appear in the overview view. The fields.properties must list ALL output fields, even if they are not in the views section.
Only adding "required": [] and "additionalProperties": true on nested object-type properties but forgetting them on the top-level fields object. Both levels need them.
Note: nullable is an Apify-specific extension to JSON Schema draft-07. It is intentional and correct.
Field type patterns
String field:
"title":{"type":"string","description":"Title of the scraped item","nullable":true,"example":"Example Item Title"}
Number field:
"viewCount":{"type":"number","description":"Number of views","nullable":true,"example":15000}
Boolean field:
"isVerified":{"type":"boolean","description":"Whether the account is verified","nullable":true,"example":true}
Array field:
"hashtags":{"type":"array","description":"Hashtags associated with the item","items":{"type":"string"},"nullable":true,"example":["#example","#demo"]}
Nested object field:
"authorInfo":{"type":"object","description":"Information about the author","properties":{"name":{"type":"string","nullable":true},"url":{"type":"string","nullable":true}},"required":[],"additionalProperties":true,"nullable":true,"example":{"name":"Example Author","url":"https://example.com/author"}}
Enum field:
"contentType":{"type":"string","description":"Type of content","enum":["article","video","image"],"nullable":true,"example":"article"}
Union type (e.g., TypeScript ObjectType | string):
"metadata":{"type":["object","string"],"description":"Structured metadata object, or error string if unavailable","nullable":true,"example":{"key":"value"}}
Anonymized example values
Use realistic but generic values. Follow platform ID format conventions:
Field type
Example approach
IDs
Match platform format and length (e.g., 11 chars for YouTube video IDs)
Usernames
"exampleuser", "sampleuser123"
Display names
"Example Channel", "Sample Author"
URLs
Use platform's standard URL for
Implementation Guide
Prerequisites
›Claude Desktop or compatible AI client with skill support
›Clear understanding of task or problem to solve
›Willingness to iterate and refine outputs
Time Estimate
15-45 minutes depending on use case complexity
Steps
1Install skill using provided installation command
2Test with simple use case relevant to your work
3Evaluate output quality and relevance
4Iterate on prompts to improve results
5Integrate into regular workflow if valuable
Common Pitfalls
⚠Expecting perfect results without iteration
⚠Not providing enough context in prompts
⚠Using skill for tasks outside its intended scope
⚠Accepting outputs without review and validation
Best Practices
✓ Do
+Start with clear, specific prompts
+Provide relevant context and constraints
+Review and refine all outputs before using
+Iterate to improve output quality
+Document successful prompt patterns
✗ Don't
−Don't use without understanding skill limitations
−Don't skip validation of outputs
−Don't share sensitive information in prompts
−Don't expect skill to replace human judgment
💡 Pro Tips
★Be specific about desired format and style
★Ask for multiple options to choose from
★Request explanations to understand reasoning
★Combine AI efficiency with human expertise
When to Use This
✓ Use when
Use when skill capabilities match your task, clear ROI on time saved, and you can validate outputs. Best for repetitive tasks, learning, and quality improvement.
✗ Avoid when
Avoid when task requires deep expertise you can't validate, involves sensitive decisions, or when learning process is more valuable than speed of completion.
Learning Path
1Familiarize yourself with skill capabilities and limitations
2Start with low-risk, non-critical tasks
3Progress to more complex and valuable use cases
4Build expertise through regular use and experimentation
