Add dataset card

README.md

@@ -0,0 +1,48 @@
+---
+license: apache-2.0
+task_categories:
+  - text-generation
+language:
+  - en
+tags:
+  - tool-use
+  - function-calling
+  - zeroclaw
+  - agent
+  - synthetic
+size_categories:
+  - n<1K
+---
+
+# ZeroClaw Tool-Use Training Data
+
+Training dataset for teaching LLMs to use tools in the [ZeroClaw](https://github.com/lucasmcoleman/zeroclaw) autonomous agent runtime.
+
+## Format
+
+Standard chat-messages JSONL. Each line is a complete multi-turn conversation:
+
+```json
+{"messages": [{"role": "system", "content": "..."}, {"role": "user", "content": "..."}, {"role": "assistant", "content": "<tool_call>...</tool_call>"}]}
+```
+
+## Stats
+
+- **457 examples** with 1470 tool-call turns
+- **25 tools** covered: shell, file_read, file_write, file_edit, glob_search, content_search, memory_store/recall/forget, cron_add/list/remove, web_fetch, web_search, http_request, git_operations, browser, delegate, pushover, backup_tool, pdf_read, screenshot, project_intel, security_ops, tool_search
+- **Tool call format**: XML-wrapped JSON (`<tool_call>{"name": "...", "arguments": {...}}</tool_call>`)
+- Includes multi-tool chains, error recovery scenarios, and reasoning traces
+
+## Generation
+
+- 108 hand-crafted scenarios covering core tool patterns
+- LLM-generated scenarios (via local models) for diversity
+- Validated: no empty user messages, all results are strings, all tools are valid
+- Completion-only loss masking: only assistant turns contribute to training loss
+
+## Usage
+
+```python
+from datasets import load_dataset
+dataset = load_dataset("lmcoleman/zeroclaw-tool-use-training", split="train")
+```