Update to Adaption

{
  "description": "Import, process, and transform datasets using Adaption MCP. Use when importing datasets, running adaptation, augmenting, translating, localizing, or combining datasets.",
  "included_files": [],
  "name": "adaption-dataset",
  "skill_md_contents": "---\nname: adaption-dataset\ndescription: Import, process, and transform datasets using Adaption MCP. Use when importing \n  datasets, running adaptation, augmenting, translating, localizing, or combining datasets.\n---\n\n# Adaption Dataset Skill\n\nImport, process, and transform datasets using the Adaption platform.\n\n## When to use\n\n- Import datasets from HuggingFace, Kaggle, or Google Sheets\n- Run dataset adaptation pipeline\n- Augment datasets with synthetic domain/general rows\n- Translate or localize dataset content\n- Combine multiple datasets\n- Check processing status and download results\n\n## Available MCP Tools\n\nUse these tools via the Adaption MCP server:\n\n| Tool | Description |\n|------|-------------|\n| `list_datasets` | List datasets visible to the authenticated Adaption organization |\n| `get_dataset` | Get a concise public view of one dataset, including processing and column information |\n| `get_dataset_status` | Poll dataset processing status and progress after async operations |\n| `get_dataset_evaluation` | Get quality evaluation status and results for one dataset |\n| `get_dataset_export_links` | Poll saved external export URLs for a dataset after export |\n| `import_dataset` | Import a dataset from HuggingFace, Kaggle, or Google Sheets URL |\n| `run_dataset_adaptation` | Estimate or launch adaptation for an imported dataset |\n| `augment_dataset` | Create new dataset with source rows plus curated domain/general rows |\n| `translate_dataset` | Create new dataset with sampled rows translated into target languages |\n| `localize_dataset` | Create new dataset with rows localized for country/language pairs |\n| `combine_datasets` | Combine compatible ready datasets into one new dataset |\n\n## Example Workflows\n\n### Import and adapt a HuggingFace dataset\n\n1. Call `import_dataset` with a `source` object — all import options nest inside it:\n   ```json\n   {\n     \"source\": {\n       \"url\": \"https://huggingface.co/datasets/squad\",\n       \"files\": [\"plain_text/train-00000-of-00001.parquet\"]\n     }\n   }\n   ```\n   `files` is required for HuggingFace and Kaggle imports and rejected for Google\n   Sheets. Pass `processing_mode: \"raw\"` to import a trainable dataset without\n   running Adaptive Data.\n2. Poll `get_dataset_status` until processing completes\n3. Call `run_dataset_adaptation` with `estimate: true` to preview cost\n4. Call `run_dataset_adaptation` with column mapping to launch\n5. Poll `get_dataset_status` until adaptation completes\n6. Call `get_dataset_export_links` to download results\n\n### Augment a dataset with synthetic data\n\n1. Ensure source dataset status is `ready`\n2. Call `augment_dataset` with:\n   - `dataset_id`: Source dataset ID\n   - `domain_rows`: More samples from the domain already present in the dataset\n   - `general_rows`: Samples from other domains not present in it\n   - `training_type`: `instruction_dataset` (default) or `preference_pairs`\n   - `estimate: true` for cost preview\n3. Launch with `estimate: false`\n4. Poll `get_dataset_status` on the new dataset ID\n\nEach row count tops out at 100,000 per strategy. Augmenting creates a new dataset\nrather than modifying the source.\n\n### Translate dataset to multiple languages\n\n1. Call `translate_dataset` with:\n   - `dataset_id`: Source dataset ID\n   - `languages`: Array of target language codes\n   - `sample_rate`: Share of rows to translate, between 0.01 and 1\n   - `estimate: true` for cost preview\n2. Poll `get_dataset_status` until translation completes\n\n`localize_dataset` takes the same `sample_rate` but replaces `languages` with\n`pairs`, an array of `{ \"country\": \"RS\", \"language\": \"sr\" }` objects.\n\n### Combine multiple datasets\n\n1. Ensure all source datasets have compatible schemas and status `ready`\n2. Call `combine_datasets` with:\n   - `dataset_ids`: 2 to 10 unique dataset IDs, each a UUID v4\n   - `name`: Required name for the combined dataset\n   - `idempotency_key`: Required — reusing the same key returns the same result\n     instead of combining twice\n\nAll sources must belong to the same organization.\n\n## Column Mapping\n\nWhen calling `run_dataset_adaptation`, map dataset columns to roles. Full rules: https://docs.adaptionlabs.ai/adaptive-data/select-columns/\n\n- `prompt` and `completion` are column names. At least one is required. If the other is missing, Adaptive Data generates it.\n- `context` is a list of column names (background or metadata), not a single string.\n- `image` is one column of image bytes, URLs, or paths. Do not put images in `context`.\n- `chat` is one column of multi-turn message arrays. It replaces `prompt`, `completion`, and `context`.\n- `universal_prompt` is a shared instruction string for every row, not a column name. Use it when the dataset has no prompt column. It requires `context` or `image` to vary each row, and is mutually exclusive with `prompt`.\n\n```json\n{\n  \"column_mapping\": {\n    \"prompt\": \"question\",\n    \"completion\": \"answer\",\n    \"context\": [\"category\", \"source\"]\n  }\n}\n```\n\n## Adaptation Options\n\nBeyond `column_mapping`, `run_dataset_adaptation` accepts:\n\n| Parameter | Description |\n|-----------|-------------|\n| `training_type` | `instruction_dataset` (default) or `preference_pairs`. Use `preference_pairs` to prepare a dataset for an alignment training run |\n| `recipe_specification` | Toggles for `prompt_rephrase`, `deduplication`, and `reasoning_traces` under a `recipes` object |\n| `brand_controls` | `length` (`minimal`, `concise`, `detailed`, `extensive`), `safety_categories`, `hallucination_mitigation`, and `blueprint` |\n| `job_specification` | `max_rows` to cap the run, plus `idempotency_key` |\n| `language_expansion` | Same `translate`/`localize` spec as dataset generation |\n| `estimate` | Set `true` to preview cost without launching |\n\n## Tips\n\n- Use `estimate: true` parameter to preview costs before launching operations\n- Poll `get_dataset_status` every 5-10 seconds during processing — it returns\n  `progress_percentage` alongside the status\n- Filter `list_datasets` with `status` (`pending`, `running`, `awaiting_input`,\n  `succeeded`, `failed`) or a `query` string instead of paging through everything\n- Launch responses carry a `next_action` naming the tool to call next — follow it\n- Check `get_dataset_evaluation` for quality metrics after adaptation\n- Use `idempotency_key` for safe retries on launch operations\n"
}