feat: add Claude Code skill for Data Designer library#309
Open
johnnygreco wants to merge 3 commits intomainfrom
Open
feat: add Claude Code skill for Data Designer library#309johnnygreco wants to merge 3 commits intomainfrom
johnnygreco wants to merge 3 commits intomainfrom
Conversation
Greptile OverviewGreptile Summary
|
| Filename | Overview |
|---|---|
| packages/data-designer-config/src/data_designer/config/init.py | Exports ConstraintType and InequalityOperator via TYPE_CHECKING and lazy imports; small API-surface fix. |
| skill/README.md | Adds setup instructions for installing and using the Claude Code skill; includes download commands for the skill folder. |
| skill/data-designer/SKILL.md | Introduces core skill definition and workflow docs; currently instructs capturing 'chain-of-thought' via extract_reasoning_content. |
| skill/data-designer/hooks/check_data_designer.sh | Adds session-start hook to detect data_designer installation and print version/path context. |
| skill/data-designer/references/advanced_patterns.md | Adds advanced usage guide; currently recommends storing chain-of-thought via extract_reasoning_content. |
| skill/data-designer/references/api_reference.md | Adds large API reference; includes wording that extract_reasoning_content captures 'chain-of-thought'. |
| skill/data-designer/scripts/helpers/pydantic_info_utils.py | Adds shared helper utilities for printing Pydantic model schemas/types, including nested model and enum expansion. |
| skill/test_info_scripts.py | Adds comprehensive unit/integration tests for the introspection scripts using subprocess uv runs. |
Sequence Diagram
sequenceDiagram
participant User
participant ClaudeCode as Claude Code
participant HookStart as check_data_designer.sh
participant InfoScripts as get_*_info.py
participant DD as data_designer (DataDesigner)
participant MCP as MCP Provider/Tools
participant HooksPost as ruff_lint.sh & ty_check.sh
User->>ClaudeCode: Activate data-designer skill
ClaudeCode->>HookStart: SessionStart hook (startup)
HookStart-->>ClaudeCode: Prints install/version/library path
User->>ClaudeCode: Describe dataset + requirements
ClaudeCode->>InfoScripts: (Optional) uv run get_column/sampler/processor/validator info
InfoScripts-->>ClaudeCode: Prints schema/field references
ClaudeCode->>DD: Build DataDesignerConfigBuilder + add columns
alt Tool-augmented generation
ClaudeCode->>DD: Configure ToolConfig(tool_alias, providers, allow_tools)
DD->>MCP: Call tool(s) via MCP during LLM column execution
MCP-->>DD: Tool results
end
ClaudeCode->>DD: validate(config_builder)
ClaudeCode->>DD: preview(config_builder, num_records=...)
ClaudeCode->>DD: create(config_builder, num_records=..., dataset_name=...)
ClaudeCode->>HooksPost: PostToolUse hook on Write/Edit
HooksPost-->>ClaudeCode: Lint/type-check edited file
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
Comment on lines
+247
to
+259
| ``` | ||
|
|
||
| ### Trace & Reasoning Capture | ||
|
|
||
| ```python | ||
| dd.LLMTextColumnConfig( | ||
| name="answer", | ||
| prompt="...", | ||
| model_alias="nvidia-reasoning", | ||
| with_trace=dd.TraceType.ALL_MESSAGES, # -> answer__trace column | ||
| extract_reasoning_content=True, # -> answer__reasoning_content column | ||
| ) | ||
| ``` |
There was a problem hiding this comment.
Encourages chain-of-thought capture
This skill docs recommend extract_reasoning_content=True and describe it as “capture chain-of-thought”. Persisting model reasoning content is typically disallowed/undesired (privacy/safety + provider ToS) and can also break runs if the underlying provider/model doesn’t return a separate reasoning field. Consider removing this recommendation (or reframing to capture tool traces only) and documenting safe alternatives (e.g., with_trace=TraceType.ALL_MESSAGES without extracting reasoning).
Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!
Prompt To Fix With AI
This is a comment left during a code review.
Path: skill/data-designer/SKILL.md
Line: 247:259
Comment:
**Encourages chain-of-thought capture**
This skill docs recommend `extract_reasoning_content=True` and describe it as “capture chain-of-thought”. Persisting model reasoning content is typically disallowed/undesired (privacy/safety + provider ToS) and can also break runs if the underlying provider/model doesn’t return a separate reasoning field. Consider removing this recommendation (or reframing to capture tool traces only) and documenting safe alternatives (e.g., `with_trace=TraceType.ALL_MESSAGES` without extracting reasoning).
<sub>Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!</sub>
How can I resolve this? If you propose a fix, please make it concise.
Comment on lines
+372
to
+390
| ## 8. Trace & Reasoning Extraction | ||
|
|
||
| ### Trace Types | ||
|
|
||
| - `TraceType.NONE` (default): No trace | ||
| - `TraceType.LAST_MESSAGE`: Only final response -> `{name}__trace` | ||
| - `TraceType.ALL_MESSAGES`: Full conversation -> `{name}__trace` | ||
|
|
||
| ### Reasoning Content | ||
|
|
||
| `extract_reasoning_content=True` creates `{name}__reasoning_content` with chain-of-thought. | ||
|
|
||
| Available on all LLM column types. | ||
|
|
||
| ### Use Cases | ||
|
|
||
| - **Debugging**: `ALL_MESSAGES` to see full conversation including tool calls | ||
| - **Fine-tuning data**: `extract_reasoning_content=True` for clean reasoning extraction | ||
| - **Tool-use training**: `ALL_MESSAGES` to capture tool call patterns |
There was a problem hiding this comment.
Avoid persisting model reasoning
This section documents extract_reasoning_content=True as producing “chain-of-thought” and suggests using it for fine-tuning. Persisting model reasoning content is generally not appropriate (privacy/safety + provider policies), and it may not be supported consistently across providers/models. Recommend removing/rewriting this guidance to focus on safer debugging artifacts (e.g., tool traces / final outputs) rather than storing reasoning content.
Prompt To Fix With AI
This is a comment left during a code review.
Path: skill/data-designer/references/advanced_patterns.md
Line: 372:390
Comment:
**Avoid persisting model reasoning**
This section documents `extract_reasoning_content=True` as producing “chain-of-thought” and suggests using it for fine-tuning. Persisting model reasoning content is generally not appropriate (privacy/safety + provider policies), and it may not be supported consistently across providers/models. Recommend removing/rewriting this guidance to focus on safer debugging artifacts (e.g., tool traces / final outputs) rather than storing reasoning content.
How can I resolve this? If you propose a fix, please make it concise.
Comment on lines
+103
to
+113
| ```python | ||
| dd.LLMTextColumnConfig( | ||
| name: str, | ||
| prompt: str, # Jinja2 template | ||
| model_alias: str, | ||
| system_prompt: str | None = None, | ||
| tool_alias: str | None = None, # MCP tool reference | ||
| with_trace: dd.TraceType = TraceType.NONE, # NONE, LAST_MESSAGE, ALL_MESSAGES | ||
| extract_reasoning_content: bool = False, # -> {name}__reasoning_content | ||
| multi_modal_context: list[dd.ImageContext] | None = None, | ||
| drop: bool = False, |
There was a problem hiding this comment.
Reasoning field documentation
The API reference describes extract_reasoning_content as “capture chain-of-thought”. If the library intentionally supports extracting a separate reasoning field, this should be documented more carefully (what exactly is captured, which providers support it, and whether it’s safe/allowed to store). Otherwise, consider removing or rewording to avoid encouraging collection of chain-of-thought.
Prompt To Fix With AI
This is a comment left during a code review.
Path: skill/data-designer/references/api_reference.md
Line: 103:113
Comment:
**Reasoning field documentation**
The API reference describes `extract_reasoning_content` as “capture chain-of-thought”. If the library intentionally supports extracting a separate reasoning field, this should be documented more carefully (what exactly is captured, which providers support it, and whether it’s safe/allowed to store). Otherwise, consider removing or rewording to avoid encouraging collection of chain-of-thought.
How can I resolve this? If you propose a fix, please make it concise.
johnnygreco
commented
Feb 6, 2026
Comment on lines
+63
to
+64
| ConstraintType, | ||
| InequalityOperator, |
Contributor
Author
There was a problem hiding this comment.
ugh, these leaked over from a different PR 🤦♂️
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
📋 Summary
Adds a comprehensive Claude Code skill that teaches Claude how to generate synthetic datasets using NVIDIA NeMo Data Designer. When activated, Claude can design and build complete data generation pipelines — choosing the right column types, writing prompts, wiring up dependencies, and iterating on previews — all from a natural language description of the dataset.
🔄 Changes
✨ Added
skill/data-designer/SKILL.md— Core skill definition with step-by-step workflow, column type decision tree, key patterns (Jinja2 templating, seed datasets, constraints, processors, validators), and best practicesskill/data-designer/examples/— 5 runnable pattern-reference scripts:basic_text_generation.py— Sampler + expression + LLM text columnsstructured_and_code.py— Pydantic structured output + -seed_dataset_with_judge.py— Seed data + LLM judge scoringcustom_column_with_llm.py—@custom_column_generatorwith multi-model orchestrationmcp_tool_use.py— MCP tool calling with trace captureskill/data-designer/references/— Complete API reference (api_reference.md) and advanced patterns guide (advanced_patterns.md)skill/data-designer/scripts/— Discovery tools for API introspection (get_column_info.py,get_sampler_info.py,get_processor_info.py,get_validator_info.py, and sharedpydantic_info_utils.py)skill/data-designer/hooks/— Session hooks for environment checks (check_data_designer.sh), ruff linting (ruff_lint.sh), and ty type checking (ty_check.sh)skill/README.md— Setup instructions and quick start guideskill/test_info_scripts.py— 900-line test suite for the discovery scripts🔧 Changed
AGENTS.md— AddedEmbeddingColumnConfigandCustomColumnConfigto the column configuratiopackages/data-designer-config/src/data_designer/config/__init__.py** — Added missingConstraintTypeandInequalityOperatorexports to lazy imports🔍 Attention Areas
skill/data-designer/SKILL.md— Core skill definition that drives Claude's behavior; verify the workflow, decision tree, and patterns are accurate and completeskill/data-designer/references/api_reference.md— Complete API documentation (557 lines); ensure accuracy against current codebasepackages/data-designer-config/src/data_designer/config/__init__.py— Nets (ConstraintType,InequalityOperator) that affect the user-facing API surface🤖 Generated with AI